Capacity and Level of Service at Finnish PDF

S12 Solutions to improve main roads
Capacity and Level of Service at Finnish

Unsignalized Intersections
Tiehallinnon selvityksiä 1/2004

S 12 Solutions to improve main roads
R. Tapio Luttinen
Capacity and Level of Service at Finnish

Finnra Reports
1/2004
Finnish Road Administration

Helsinki 2004
Finnra Reports 1/2004
ISBN 951-803-180-0
ISSN 1457-9871
TIEH 3200849E
PDF version (www.tiehallinto.fi/julkaisut )

ISBN 951-803-181-9
ISSN 1459-1553
TIEH 3200849E-v
Edita Prima Oy
Helsinki 2004
The publication is sold by

Email: asiakaspalvelut.prima@edita.fi
Tel. Int. +358 20 450 011
Telefax Int. +358 20 450 2470

Traffic Engineering
Opastinsilta 12 A
P.O. Box 33
00521 HELSINKI
FINLAND
Telephone: Int. +358 204 2211
R. TAPIO LUTTINEN: Capacity and Level of Service at Finnish Unsignalized Intersections.
Helsinki 2004, Finnish Road Administration, Finnra Reports 1/2004, ISBN 951-803-180-0, ISSN
1457-9871, TIEH 3200849E.
Keywords: capacity, level of service, unsignalized intersections
ABSTRACT
Since late 70’s capacity and delay at unsignalized intersections have been estimated
following the Swedish method adjusted for Finnish conditions. During the past 25 years
the performance of both drivers and vehicles has changed, and new calculation methods
have been developed. It has become necessary to update the analysis methodology of
Finnish unsignalized intersections.
Major attention has been given to capacity analysis methodology, because capacity
estimates have a central role in the estimation of other performance measures, such as
delays and queue lengths. In addition, delay estimates are very sensitive to inaccuracies
in capacity estimates, especially at high degrees of saturation.
Capacity at an unsignalized intersection has been described as a conditional expectation,
from which the capacity equations for the most common headway distributions are
easily obtained. For Rank 3 and Rank 4 capacity analysis a new theoretical approach
has been developed. The movement capacity of Rank 3 streams is adjusted both for
Rank 2 queues and for the modified headway distribution and lower flow rates during
the free-departure periods of Rank 2 streams. The Rank 4 stream adjustment factors
are shown to be multiplicative even across streams of different ranks.
The report presents the theoretical background required to manually perform opera-
tional analysis of unsignalized intersections and to understand and evaluate published
methodologies and capacity analysis software. The methods can be used both for simple
calculations with a calculator or a spreadsheet and for advanced analytical procedures.
The theoretical results and the results of three international methodologies—HCM2000,
Swedish Capcal 2, and Danish DanKap—have been compared with simulated capacities
and delays. HCM2000 gives higher potential capacities than the simulation results.
Movement capacities of Rank 3 and Rank 4 streams are, however, slightly too low. The
potential capacities of Capcal 2 are very similar to the simulated capacities. Rank 3
movement capacities are similar as in HCM2000, but Rank 4 capacities are too low.
DanKap follow the methodology of HCM2000.
Based on theoretical results and simulation experiments new capacity and delay es-
timation methods have been presented for ordinary unsignalized intersections and for
roundabouts. The new estimators have a more solid theoretical foundation and are in
better agreement with simulation results than the other methodologies evaluated.
R. TAPIO LUTTINEN: Capacity and Level of Service at Finnish Unsignalized Intersections
[Valo-ohjauksettomien liittymien välityskyky ja palvelutaso Suomessa]. Helsinki 2004, Tie-
hallinto, Finnra Reports 1/2004, ISBN 951-803-180-0, ISSN 1457-9871, TIEH 3200849E.
Asiasanat: välityskyky, palvelutaso, valo-ohjaukseton liittymä
TIIVISTELMÄ
Valo-ohjauksettomien liittymien välityskyvyn ja viipeiden arviointi Suomessa perustuu

70-luvun lopussa kehitettyyn ruotsalaiseen menetelmään. Kuluneiden runsaan 25 vuo-
den aikana sekä ajoneuvokanta, liikenekulttuuri että analyysitekniikat ovat kehittyneet.
Myös suomalaisten laskentamenetelmien uudistaminen on tullut ajankohtaiseksi.
Välityskykyestimaateilla on suuri vaikutus liikennevirran laadullisten mittareiden, ku-
ten viipeiden ja jonojen pituuksien estimaatteihin. Välityskyvyn laskentamenetelmillä
on siten toimivuustarkasteluissa keskeinen rooli.
Valo-ohjauksettoman liittymän välityskyky voidaan ilmaista ehdollisten todennäköi-
syyksien avulla, jolloin lauseke saa helposti ymmärrettävän muodon ja välityskyky-
lausekkeet pääsuunnan liikenteen eri aikavälijakaumille voidaan johtaa helposti. Kol-
mannen ja neljännen tason virtojen välityskyvyn analyysiin on esitetty uusi teoreettinen
lähestymistapa, jossa välityskykyä korjataan ottaen huomioon toisen tason virtojen jo-
not sekä aikavälijakauman muuttuminen ja saapuvien ajoneuvojen alhaisempi määrä
vapaan liikennevirran aikana. Neljännen tason korjauskertoimet osoittautuivat, vastoin
odotuksia, multiplikatiivisiksi.
Esitetty teoria tarjoaa tarvittavat taustatiedot toimivuustarkastelujen suorittamiselle se-
kä julkaistujen laskentamenetelmien ja -ohjelmien arvioimiselle. Esitettyjä menetelmiä
voidaan käyttää sekä yksinkertaisiin laskelmiin että vaativiin analyyseihin.
Teoreettisia tuloksia sekä kolmea laskentamenetelmää – HCM2000 (USA), Capcal 2
(Ruotsi) ja DanKap (Tanska) – on verrattu Hutsim-simulointiohjelmalla arvioituihin
välityskykyihin ja viipeisiin. HCM2000:n välityskykyarviot perustuvat oletukseen sa-
tunnaisesta liikennevirrasta, joka johtaa suurempiin välityskykyihin kuin sekä simu-
lointitulokset että Capcal 2 osoittavat. Kolmannen tason virtojen välityskyvyn arvioin-
timenetelmä on samanlainen sekä HCM2000:ssa että Capcal 2:ssa. Välityskykyarviot
ovat simulointituloksia alhaisempia. Neljännen tason virtojen välityskykyarviot ovat
Capcal 2:ssa selvästi liian alhaisia. HCM2000:ssa on korjaustermi, joka selvästi paran-
taa tuloksia. DanKapin laskentamenetelmä on saman kaltainen kuin HCM2000:ssa.
Teoreettisten tulosten ja simulointikokeiden perusteella on esitetty uudet välityskyvyn
ja viipeiden laskentamenetelmät sekä tavanomaisille valo-ohjauksettomille liittymille
että kiertoliittymille. Uusien laskentamenetelmien tuottamat välityskyvyn ja viipeiden
estimaatit ovat lähempänä simulointituloksia kuin vertailumenetelmien arvot. Esitetyil-
lä menetelmillä on myös vankka teoreettinen perusta, jonka johdosta ne eivät tarvitse
ylimääräisiä korjaustermejä.
PREFACE
Most of the intersections in Finland are unsignalized. The vast majority of them have
yield signs on minor-road approaches. Other unsignalized intersections are either stop
controlled, uncontrolled (observing the “right before left” principle), or roundabouts.
The operational analysis of unsignalized intersections is of great significance both to
highway authorities and to road users. This report presents a theoretical overview and
suggests a new methodology for capacity analysis of unsignalized intersections.
The description of an unsignalized intersection in terms of gap acceptance and queuing
models is rather detailed. It was considered necessary to define the concepts carefully,
because the literature contains numerous different and mutually incompatible modeling
approaches.
A rather extensive evaluation of theoretical models was considered necessary in order
to highlight the strengths and limitations of existing analysis methodologies. Some
new methods are presented, especially for capacity estimation. Minor modifications
are presented for the delay-estimation method.
The basic ideas necessary to understand capacity and delay models have been presented.
Although the text contains a considerable number of equations, mathematics has been
kept at a level accessible to anyone familiar with the basics of calculus and probability
theory. Some effort has been made to describe theories and formulas to a reader,
who is not an expert in the field. As far as possible, the text is self contained. The
style is similar as in the reports on two-lane highways (Luttinen 2001) and signalized
intersections (Luttinen & Nevala 2002) published earlier.
The bibliographic references have been selected, as far as possible, to credit the original
sources of the ideas under discussion. Some references to recent overviews and text-
books have been presented when considered appropriate. In many cases the models are
based on other systems having some analogies with unsignalized intersections. In fact,
observation of analogies between very different systems is in the core of mathematical
modeling. The reader is pointed to some of these analogies.
The report was written by Dr. R. Tapio Luttinen from TL Consulting Engineers Ltd,
and currently from the Helsinki University of Technology. The simulation experiments
were conducted with Hutsim software with help from Jouni Ojala and Nina Koivisto at
the Helsinki University of Technology. The Laboratory of Transportation Engineering
at the Helsinki University of Technology also collected and analyzed the gap acceptance
measurements. Dr. Iisakki Kosonen provided valuable information about some details
of the Hutsim model.
Discussions with professors Urho Pulkkinen (VTT Industrial Systems), Rod Troutbeck
(Queensland University of Technology), and Werner Brilon (Ruhr University Bochum)
are thankfully acknowledged. Professor Troutbeck also provided a spreadsheet con-
taining an implementation of the maximum likelihood method for the estimation of
critical gaps.
Under the name NORDKAP (NORDiskt KAPacitetssamarbete, Nordic capacity co-
operation) there has been considerable exchange of information and joint research
between Denmark, Finland, Norway, and Sweden. Because traffic conditions in these
countries have many similarities, the experiences and research results obtained in one
country are of particular interest to the other countries. For this research the results of
Dr. Ola Hagring (University of Lund, Sweden) have been very valuable.
This report is part of the Finnra strategic project S12 (Solutions to improve main roads).
The work has been coordinated by deputy director Pauli Velhonoja at Finnra Traffic
Engineering. Some discussion on the capacity theory presented here has been published
prior to this final report (Luttinen 2003, Luttinen 2004b, Luttinen 2004a).
Helsinki, 3rd July 2004

Traffic Engineering
Capacity and Level of Service at Finnish Unsignalized Intersections 9
CONTENTS
Contents
GLOSSARY 13
SYMBOLS AND ABBREVIATIONS 18
1 INTRODUCTION 22
1.1 Methodological approaches 22

1.2 State of the art in Finland 23
1.3 Research objectives 24
2 GAP ACCEPTANCE THEORY 25
2.1 Relative priority of traffic streams 25

2.2 Definition of priority streams 29
2.3 Gap availability 30
2.3.1 Gaps, lags, and headways 30
2.3.2 Geometric distribution 34
2.3.3 Negative exponential distribution 35
2.3.4 Shifted exponential distribution 37
2.3.5 Erlang distribution 39
2.3.6 Cowan’s M3 distribution 39
2.4 Gap acceptance and queue discharge 43
2.5 Fluid-analogy model 48
3 CAPACITY AT UNSIGNALIZED INTERSECTIONS 52
3.1 Definition of capacity 52

3.2 Potential capacity under one priority stream 54
3.2.1 Definition of potential capacity 54
3.2.2 Stepwise gap acceptance function 56
3.2.3 Linear gap acceptance function 62
3.3 Potential capacity under multiple priority streams 66
3.4 Potential capacity of roundabout approaches 73
3.4.1 Background 73
3.4.2 Historical overview 73
3.4.3 Current methodologies 76
3.5 Movement capacity of Rank 3 streams 77
3.5.1 Description of the procedure 77
10 Capacity and Level of Service at Finnish Unsignalized Intersections
CONTENTS
3.5.2 Queuing model for Rank 3 stream capacity 78

3.5.3 Rank 2 queue discharge 81
3.5.4 Rank 3 capacity with no Rank 1 conflict 82
3.5.5 Time available for gap acceptance 83
3.5.6 Gap availability for Rank 3 vehicles 86
3.5.7 Movement capacity under one Rank 2 stream 88
3.5.8 Movement capacity under two Rank 2 streams 90
3.6 Movement capacity of Rank 4 streams 92
3.7 Shared lanes 99
3.8 Pedestrian effects 100
3.9 Passenger-car equivalencies 101
3.10 Excursion: Additive Conflict Flows methodology 101
4 DELAYS AT UNSIGNALIZED INTERSECTIONS 105
4.1 Definitions of delay 105

4.2 Service times 109
4.2.1 Definition of service time 109
4.2.2 Average service time of queuing vehicles 110
4.2.3 Average service time of isolated vehicles 112
4.3 Delays under steady-state conditions 117
4.3.1 Queuing as a stochastic process 117
4.3.2 M/G/1 model 119
4.3.3 M/G2/1 model 123
4.3.4 Tanner’s delay model 124
4.4 Delays under transient conditions 128
4.4.1 Definition of a transient state 128
4.4.2 Oversaturated conditions 129
4.4.3 Coordinate-transformation method 134
5 PERFORMANCE ANALYSIS METHODOLOGIES 138
5.1 Performance measures 138

5.2 U.S. methodology 139
5.3 Danish methodology 143
5.4 Swedish methodology 145
5.4.1 Analysis procedure 145
5.4.2 Major flows 146
5.4.3 Critical gaps 147
5.4.4 Average service times 148
CONTENTS
5.4.5 Capacity 150

5.4.6 Delay 150
5.4.7 Short lanes 151
5.5 Finnish methodology 152
6 SIMULATION STUDIES 156
6.1 Hutsim as a simulation tool 156

6.2 Calibration of Hutsim for unsignalized intersections 159
6.2.1 Vehicle-dynamics parameters 159
6.2.2 Follower gaps 160
6.2.3 Driving speeds 161
6.2.4 Limited priority merge 161
6.2.5 The effect of right-turn vehicles 162
6.3 Simulated capacity of unsignalized intersections 163
6.3.1 Rank 2 capacity 163
6.4 Simulated capacity of roundabouts 173
6.5 Simulated delays 175
7 CRITICAL GAPS AND FOLLOW-UP TIMES 178
7.1 Estimation methods 178

7.2 Stop vs. yield control 178
7.3 Field studies 179
7.3.1 Data collection 179
7.3.2 Results 180
8 CONCLUSIONS AND RECOMMENDATIONS 183
8.1 Limitations of simulation experiments 183

8.2 Unsignalized intersections 183
8.3 Roundabouts 185
8.4 Level of service 188
8.5 Further research 188
REFERENCES 190
APPENDICES 205
CONTENTS
A CRITICAL GAP ESTIMATES 206
B HUTSIM INITIALIZATION FILE 209

GLOSSARY
GLOSSARY
The terminology of the Highway Capacity Manual (Transportation Research Board
2000), and American terminology in general, has been followed. Most of the defini-
tions below have been taken from the Highway Capacity Manual, possibly with slight
modifications. A major exception is the definition of a gap, which is similar to the
definition presented by (Drew 1968). The term “critical gap” has been used, as in
HCM2000, although it may be replaced with a more consistent “critical headway” in
the future. See discussion on page 30.
Acceleration delay: Time lost due to the limited acceleration capability of a vehicle.
Adjustment factor: A multiplicative factor that adjusts a capacity or service flow rate
from one representing an ideal or base condition to one representing a prevailing
condition.
Analysis period: A single time period during which a capacity analysis is performed.
Approach: A set of lanes at an intersection that accommodates all left-turn, through

and right-turn movements from a given direction.
Base conditions: Characteristics for a given type of facility that are assumed to be the
best possible from the point of view of capacity; that is, characteristics if further
improved would not result in increased capacity. (Ideal conditions.)
Base saturation flow rate: The maximum steady flow rate at which previously
stopped passenger cars can cross the stop or yield line under base conditions
and in the absence of conflicting higher priority vehicles.
Calibration: The process of adjusting model parameters in order to minimize the

difference between model results and measurements.
Capacity: The maximum hourly rate at which persons or vehicles can reasonably be
expected to traverse a point or uniform section of a lane or a roadway during a
given time period under prevailing roadway, traffic, and control conditions.
Conflicting movements: Traffic streams in conflict at an intersection.
Control condition: The traffic controls and regulations in effect for a segment of street
or highway, including the type, phasing, and timing of traffic signals; stop signs;
lane use and turn controls; and similar measures during the analysis period.
Control delay: The component of delay that results when a control causes vehicles on
a lane or a lane group to reduce speed or to stop; it is measured by comparison
with the uncontrolled condition.
Counting process: A continuous-time process that represents the number of occur-

rences of some event, such as arrival of vehicles, during a time period.
Critical gap: The minimum major-stream headway during which a typical minor-
stream vehicle can make a maneuver. (Critical headway)
Deceleration delay: Time lost due to deceleration.
Degree of saturation: Same as demand-to-capacity ratio.

GLOSSARY
Delay: The additional travel time experienced by a driver, passenger or pedestrian.
Demand flow rate: The flow rate expected to desire service past a point or segment
of the highway system at some future time, or the traffic currently arriving or
desiring service past such a point.
Demand-to-capacity ratio: The ratio of demand flow rate to capacity at a traffic fa-
cility. (D/C ratio)
Design application: Using capacity analysis procedures to assess a traffic facility for
a specified level of service.
Deterministic model: A mathematical model that is not subject to randomness.
Equilibrium condition: The expected state of a system that may have cyclic fluctu-
ations, but otherwise is time independent. (Steady state condition)
Exclusive turn lane: A designated lane used only by vehicles making a left or right
turn.
Flow rate: The expected rate at which vehicles pass a given point at a given time or
during a time interval. (Rate of flow)
Flow ratio: The ratio of the demand flow rate to the saturation flow rate for a lane
group at an intersection.
Follow-up time: Time between the departure of one minor-stream vehicle and the
departure of the next vehicle using the same gap under a condition of continuous
queuing.
Gap: A time interval between two consecutive major-stream vehicles evaluated by a

minor-stream driver for a crossing or merging operation.
Gap acceptance: The process by which a minor-stream vehicle accepts an available

gap to maneuver.
Geometric delay: The component of delay that results when geometric features cause
vehicles to reduce their speed in negotiating a facility.
Headway: Time between two vehicles passing a point measured from front bumper to
front bumper. (Interarrival time)
Heavy vehicle: Any vehicle with more than four wheels touching the pavement during
normal operation.
Ideal conditions: Base conditions.
Impedance: The reduction in the capacity of a minor stream, caused by congestion in

major streams.
Initial queue: The unmet demand from previous periods at the beginning of an analysis
period.
Lag: Time interval between a random event, such as the arrival of a minor-stream
vehicle, and the passage of the next priority-stream vehicle. (Forward waiting
time)
GLOSSARY
Lane group: A set of lanes established at an intersection approach for separate capacity
and level-of-service analysis.
Lane-group delay: The control delay for a given lane group.
Level of service: A qualitative measure describing operational conditions within a
traffic stream, generally described in terms of such factors as speed and travel
time, freedom to maneuver, traffic interruptions, comforts and convenience, and
safety.
Major road: The roadway of a higher importance in an intersection.
Major stream: A traffic stream or a set of traffic streams to which the subject stream
must give right of way.
Markov chain: A stochastic process in which the future state of the system depends
on its present state but not on its past history.
Measure of effectiveness: A quantitative parameter describing the performance of a
transportation facility or service.
Minor road: The road of lower importance at an unsignalized intersection, controlled
by yield or stop signs.
Minor stream: A traffic stream which must give right of way to a higher priority
stream.
Movement capacity: The capacity of a specific movement at an unsignalized inter-
section approach, assuming that the traffic has exclusive use of a separate lane.
Operational application: A use of capacity analysis to determine the level of service
on an existing or projected facility, with known or projected traffic, roadway, and
control conditions
Opposing flow rate: The flow rate at the direction of travel opposite to the direction
under analysis.
Overflow period: Time period during which demand exceeds capacity.
Passenger-car equivalent: The number of passenger cars having the same impedance
effect as a single heavy vehicle of a given type, under prevailing roadway, traffic,
and control conditions. (PCE)
Peak period: Time period during which arrival flow rate exceeds capacity.
Performance measure: A quantitative or qualitative characteristic describing the
quality of service provided by a transportation facility or service.
Planning application: A use of capacity analysis to estimate the level of service,
the volume than can be accommodated, or the number of lanes required, using
estimates, HCM default values, and local default values as input.
Platoon: A group of vehicles or pedestrians traveling together as a group, either vol-
untarily or involuntarily because of signal control or other factors.
Potential capacity: The capacity of a specific movement at an unsignalized inter-
section approach, assuming that it is unimpeded by other movements and has
exclusive use of a separate lane.
GLOSSARY
Prevailing condition: The geometric, traffic, and control conditions during the ana-
lysis period.
Priority stream: A traffic stream to which the subject stream must give right of way.
Quality of service: The capability of a transportation facility or system under given

conditions to meet the needs of its users.
Queue: A line of vehicles, bicycles or persons waiting to be served by a system.
Rank: A level of priority in the hierarchy of traffic streams at an unsignalized inter-

section.
Recreational vehicle: A heavy vehicle, generally operated by a private motorist, en-

gaged in the transportation of recreational equipment or facilities; examples in-
clude campers, boat trailers, and motorcycle trailers.
Renewal process: A counting process for which the interarrival times (headways) are
independent and identically distributed non-negative random variables.
Safety lag: Time interval at a conflict area from the departure of a minor-stream vehicle
to the arrival of a major-stream vehicle.
Saturation flow rate: The hourly rate at which previously queued passenger cars can
traverse an intersection approach under prevailing conditions in the absence of
conflicting higher priority vehicles.
Service flow rate: The maximum hourly rate at which persons or vehicles can reas-
onably be expected to traverse a point or uniform section of a lane or roadway
during a given time period under prevailing roadway, traffic, and control condi-
tions while maintaining a designated level of service.
Service measure: A specific performance measure used to assign a level of service.
Service time: The time period that a customer spends in a service facility. At unsignal-
ized intersections the time period during which a vehicle not blocked by vehicles
ahead blocks the vehicle behind from reaching the stop/yield line.
Simulation model: A computer program that uses mathematical models to conduct ex-
periments with traffic events on a transportation facility or system over extended
periods of time.
Spacing: The distance between two successive vehicles in a traffic lane measured from
front bumper to front bumper.
Stationary conditions: Conditions not changing during the analysis period.
Steady state conditions: The expected state of the system may have cyclic fluctu-
ations, but otherwise it is time independent. (Equilibrium conditions)
Stochastic model: A mathematical model that employs random variables for at least
one input parameter.
Stochastic process: A series of random variables indexed over time or space. In a

discrete time stochastic process the index set consists of discrete units of time.
In a continuous time stochastic process the index set is a subset of nonnegative
real numbers.
GLOSSARY
Stop line: A line behind which vehicles should stop as directed by a stop sign or a
traffic signal.
Stop time: A portion of control delay when vehicles are at complete stop. Also “stop
delay”.
Stream: A distinguishable element of the total traffic flow traversing an intersection—

for example, that performing a particular movement (OECD Road Research
Group 1974)
Time-in-queue delay: Time spent in a queue; from stopping at the end of queue to
passing the stop line.
Through vehicles: All vehicles passing directly through an intersection.
Traffic condition: A characteristic of traffic flow, including distribution of vehicle

types in the traffic stream, directional distribution of traffic, lane use distribution
of traffic, and type of driver population on a given facility during the analysis
period.
Transient conditions: The system is not in equilibrium, but its state depends on time.
(Time-dependent conditions)
Transition probability: The probability that a system at a given state at one time will
be at a certain state at a later time.
Truck: A heavy vehicle engaged primarily in the transport of goods and materials or
in the delivery of services other than public transport.
Undersaturation: A traffic condition in which the arrival flow rate is lower than the
capacity.
Utilization factor: The fraction of time that a server in a queuing system is busy or a
traffic facility is occupied.
Validation: Determining whether the selected model is appropriate for the given con-
ditions and for the given task. It compares model prediction with measurements
or observations.
Volume: The number of persons or vehicles passing a point on a lane, roadway, or

other trafficway during some time interval.
Waiting time in queue: Time spent in a queue waiting for service.
Waiting time in system: Time spent in queue and in service.
Yield line: A line of triangles at an unsignalized intersection to intensify the effect of

a yield sign. At a yield controlled approach the stop line is replaced by a yield
line.
SYMBOLS AND ABBREVIATIONS

a Acceleration or deceleration rate
Ai Number of arrivals between departures i and i − 1
A(τ ) Number of arrivals during (0, τ ]
A(τ1 , τ2 ) Number of arrivals during (τ1 , τ2 ]
ACF Additive Conflict Flows methodology
B(τ ) Backward waiting time at time τ
b Current duration of headway
C Capacity
Ci Capacity of stream i
Ca,i Potential capacity of stream i during unimpeded periods
CL,i Potential capacity of minor stream i under limited priority
Cm,i Movement capacity of minor stream i
Cp,i Potential capacity of minor stream i
c Cycle length (ta + tb )
D(τ ) Number of departures during (0, τ ]
D(τ1 , τ2 ) Number of departures during (τ1 , τ2 ]
d Central island diameter in a roundabout
D/C Demand-to-capacity ratio
EB Equivalency factor for buses
EHV Equivalency factor for heavy vehicles
ER Equivalency factor for recreational vehicles
ET Equivalency factor for trucks
E[X] Expected value of random variable X
E[ X | Y ] Expected value of random variable X under condition Y
f(·) Probability density function
fe (·) Probability density function of service times for isolated vehicles
fq (·) Probability density function of service times for queuing vehicles
fS (·) Probability density function of service times
fϒ (·) Probability density function of lags
F(·) Cumulative distribution function (see f(·) for subscripts)
fa,i Fraction of time available for gap acceptance in stream i
fe Adjustment factor for exiting traffic
fi Impedance factor of a Rank 2 stream i
fL Adjustment factor for limited priority merge
fq Adjustment factor for impedance effects in average service time
Finnra Finnish Road Administration
G(z) Probability-generating function
h̄ Average departure headway
hi,k Departure headway of vehicle k in stream i
h̄j Average departure headway of stream j
HBS Handbuch für die Bemessung von Straßenverkehrsanlagen
HCM Highway Capacity Manual
HUT Helsinki University of Technology
Ik The set of priority streams for minor stream k
i.i.d. Independent and identically distributed
La Queue length exceeded a percent of time
L(τ ) Number of customers in a queue at time τ
lf Effective entry flare length
lv Vehicle length
lw Length of weaving section

LOS Level of service
MOE Measure of effectiveness
N (τ ) Number of customers in a system at time τ
ng Average number of pedestrians in a group
P One-step transition probability matrix
PB Proportion or percentage of buses in traffic flow
PHV Proportion or percentage of heavy vehicles in traffic flow
PR Proportion or percentage of recreational vehicles in traffic flow
PT Proportion or percentage of trucks in traffic flow
P{A} Probability of event A
P{ A | Y } Probability of event A under condition Y
p0 Probability of a lag being accepted
p0,i Probability of empty system in stream i
pe Proportion of arriving flow exiting (turning right) before subject entry
pg Probability that a gap is rejected
pi Proportion of stream i in a set of streams
pij One-step transition probability from state i to state j
pk,j Fraction of pedestrians in stream j given priority by vehicle stream k
pl Probability that a lag is rejected
pn Probability of rejecting n consecutive gaps, or lag and n − 1 gaps
pz,i Probability of empty system in streams 1, 4, and i
pc Passenger car
PCE passenger-car equivalent
pcu Passenger car unit
q Flow rate, traffic volume (veh/h)
qa,i Average stream i flow rate (veh/h) during free-flow periods
qc Circulating flow at a roundabout
qe Rate of flow exiting (turning right) before subject entry
qg Hourly arrival rate of pedestrian groups
qII Flow rate of a Rank 2 movement
qIII Flow rate of a Rank 3 movement
qi Flow rate of movement i
qM Major-stream flow rate
qM,i Major-stream flow rate for movement i
qP Peak flow rate
qp Hourly arrival rate of pedestrians
q(τ ) Flow rate at time τ
R2 Coefficient of determination
R(·) Survivor function (see f(·) for subscripts)
rc Inscribed circle radius of a roundabout
re Entry radius at a roundabout
S Service time
S̄ Average service time
Se Service time for vehicles arriving to an empty system
Sq Service time for queuing vehicles
S̄i Average service time of stream i
S (i) Service time of vehicle i
si Saturation flow rate of stream i
T Headway (stochastic variable)
Ti Time interval between major-stream departures i − 1 and i

T (A, B) Time interval between events A and B
T (τ ) Current headway at time τ
Tbc,i Busy cycle in stream i
Tbp,i Busy period in stream i
Tip,i Idle period in stream i
t Length of a headway
t0 Queue accumulation and discharge time
ta Antiblock length
tb Block length
tc Critical gap
tf Follow-up time
tf(i) Follow-up time after vehicle i
ti Length of the i th headway
to Shortest acceptable major stream headway in a flow-analogy model
tp Intraplatoon headway
tpb Time interval of pedestrian blockage
v/c Volume-to-capacity ratio
va Approach cruise speed
ve Exit cruise speed
vn Negotiation speed
vp Pedestrian walking speed
W Control delay
W̄ Average control delay
W (i) Control delay of vehicle i
W (τ ) Cumulative delay during (0, τ ]
W̄2 Average delay according to Tanner (1962)
Wa Acceleration delay
We Waiting time of an isolated minor stream vehicle
W̄D Average waiting time in an M/D/1 queuing system
Wd Deceleration delay
W̄G Average waiting time in an M/G/1 queuing system
Wg Geometric delay
Wga Acceleration component of geometric delay
Wgd Deceleration component of geometric delay
W̄M Average waiting time in an M/M/1 queuing system
Wm Queue move-up time
Wn Delay due to low negotiation speed
Wo Overflow delay
WQ Waiting time in queue
Wq Waiting time at yield/stop line for queuing vehicles
WS Waiting time in system
Ws Stop delay
WT Waiting time in system under transient conditions
Wt Total delay
wa Approach width
we Entry width
wl Lane width
ww Width of a weaving section
α(t) Gap acceptance function; maximum number of departures in headway t
αk Probability of k arrivals during the service time of a queuing vehicle

β Unit extension interval
βk Probability of k arrivals during the service time of an isolated vehicle
γ Scale parameter of Cowan’s M3 distribution
δi Service departure time of vehicle i
δ(x − a) Unit impulse (Dirac delta) function
(x) Unit step (Heaviside) function
εi Time of entry of minor stream vehicle i
θ Scale parameter of shifted exponential distribution
κ Randomness constant of service times
κi Reserve capacity of stream i
λi Flow rate (veh/s) of stream i
λa,i Average stream i flow rate (veh/s) during free-flow periods
φi Proportion of free headways in stream i
ρe Service quotient for isolated (non-queuing) vehicles
ρi Utilization factor, demand-to-capacity ratio of stream i
ρi Demand-to-capacity ratio of stream i
ρP Peak demand-to-capacity ratio
ρq Service quotient for queuing vehicles
σ2 Variance
σe2 Variance of service times for vehicles arriving to an empty system
σq2 Variance of service times for queuing vehicles
τ Time
τi Arrival time of vehicle i
τo Length of overflow period
τP Length of peak period
ϒ(τ ) Lag (forward waiting time) at time τ
υ Length of lag
υs Safety lag
χ Probability of major stream antiblock and minor stream follow-up time
d
= Equal in distribution
(a, b] Time interval a < τ ≤ b
i∨j Stream i or j
i∧j Streams i and j
x Floor function (greatest integer not larger than x)
x Nearest integer of x
x+ Positive part of x; i.e., x+ = (x + |x|)/2
INTRODUCTION
1 INTRODUCTION
1.1 Methodological approaches
The majority of intersections in Finland are unsignalized. The vast majority of these
have yield signs on minor road approaches. Other unsignalized intersections are either
stop controlled, uncontrolled, observing the “right before left” principle, or round-
abouts, which are becoming increasingly popular.
At an unsignalized intersection each driver must find a safe moment for the movement
observing current traffic, traffic signs (stop or yield) and pertinent regulations (“right
before left”). Accordingly, modeling of operations at an unsignalized intersection is
about the interactions between vehicles under the geometric conditions of the intersec-
tion. The most important modeling approaches have been stochastic (gap acceptance
theory) and statistical (regression analysis).
In Britain the capacity analysis of unsignalized intersections has been based on statist-
ical models (Kimber & Coombe 1980, Kimber 1980, Semmens 1988, O’Flaherty 1997).
Regression analysis has been used to find the parameters and to estimate the values of
the coefficients which give a good fit between estimated and measured capacity. This
method is not sensitive to the headway distribution models, it gives a clear method to
estimate the effect of intersection geometry, and can model heavily congested condi-
tions under which the process of simple gap-acceptance gives way to a more interactive
one, in which major road vehicles adjust their headways to allow minor road vehicles to
enter, as well as situations, such as multi-lane roundabouts, where the assumed vehicle
interaction rules become very arbitrary (Kimber & Coombe 1980, Kimber 1989). Some
other examples of statistical capacity models have been presented by Louah (1988),
Simon (1991), Hakkert, Mahalel & Asante (1991), and Aakre (1998). When capacity
has been estimated, queuing theory can be used to estimate queue lengths and delays
(Kimber & Hollis 1979), although regression analysis has also been used in delay
estimation (e.g., Kyte, Zegeer & Lall 1991).
Most of the analysis procedures of unsignalized intersections are, however, based on
stochastic models; i.e., gap acceptance theory. It is assumed that a minor stream
vehicle can enter an intersection, when the time interval to the next arriving higher
priority vehicle is larger than a critical gap, and a safe time interval (follow-up time)
has passed since the departure of the preceding minor stream vehicle. Capacity and
delay are functions of follow-up times and the availability of headways larger than
critical gaps in higher priority streams. Statistical methods are applied to estimate the
parameters that describe “microscopic” properties of the process, such as critical gaps
and follow-up times of minors stream vehicles as well as headway distributions of major
streams. Whereas statistical models estimate the effects of intersection geometry on
capacity directly, in stochastic models gap acceptance parameters are adjusted for the
local conditions (Kimber 1980).
Gap acceptance theory has been developed especially in Germany by Grabe (1954),
Harders (1968), and Siegloch (1973), without forgetting the theoretical foundation
laid by Anglo-Saxon scientists, such as Major & Buckley (1962) and Tanner (1962).
Roundabouts are usually analyzed as series of intersections or short merging zones.
Gap acceptance theory can be adjusted for conditions, where an approaching major-
stream vehicle slows down to give way to a merging minor-stream vehicle. This type of
operation is called “limited priority merge” (Troutbeck & Kako 1999, Troutbeck 1999).
Priority stream headways are usually assumed to follow exponential distribution. This
approach has been used in many capacity manuals, such as in Germany (Brilon,
INTRODUCTION
Großmann & Blanke 1994, FGSV 2001) and in the U.S. (Transportation Research
Board 2000). A bunched exponential headway model used originally by Tanner (1962)
has received much attention later in a generalized form (Troutbeck 1986, Akçelik &
Chung 1994, Hagring 1998b). The Swedish method (Statens vägverk 1977, Hanson
1978, Bergh 1991) uses four different headway distributions, based on number of lanes,
distance from traffic signals, and type of movement.
Gap acceptance can also be described in terms of traffic signal analogy, as suggested
by Winsten (in Beckmann, McGuire & Winsten 1956), and recently developed by
Akçelik (1994a). Time periods blocked by higher priority vehicles are considered as
red intervals and time periods available for minor stream vehicles are considered as
green periods. Siegloch (1973) and McDonald & Armitage (1978) associated a lost
time with each major stream vehicle and assumed that at other time intervals minor
stream vehicles enter an intersection (roundabout) at constant saturation flow rate.
A third approach is to apply a model structure based on gap acceptance theory, but to
use regression analysis to fit the model to measured capacity values. Parameters, such
as critical gap and follow-up time, loose their behavioral (microscopic) meaning, and
become coefficients of a statistical (macroscopic) model. Examples of this approach
have been presented by Stuwe (1991), Brilon & Stuwe (1993), Guichet (1997), and
Aagaard (1995), among others.
The Additive Conflict Flows (ACF) procedure (Brilon & Wu 2001, Brilon & Wu 2002,
Brilon & Miltner 2002) combines traffic signal analogy with a statistical framework.
The D/C ratio of a major flow is the sum of the D/C ratios of component major flows,
and during the remaining fraction of time the minor stream can discharge at saturation
flow rate. Parameters are estimated using statistical procedures based on goodness of fit
between observed and estimated capacities. TheACF method was developed to simplify
the analysis of the hierarchy of priorities as well as the pedestrian and cyclist movements
at unsignalized intersections. The new draft chapter for unsignalized intersections in
the forthcoming German capacity manual follows the ACF procedure.
1.2 State of the art in Finland
In early 70’s the capacity analysis of Finnish unsignalized intersections (Häkli 1971,
Syyrakki, Lyly & Granberg 1975) was an implementation of the methodology of
Harders (1968) as presented by Jessen (1968)—see also Siegloch (1973), appendix 17.
Since late 70’s capacity and delay at unsignalized intersections have been estimated
following the Swedish method adjusted for Finnish conditions (Tiensuunnittelutoimisto
1978). These adjustments have also been implemented in the Finnish version of the
Swedish Capcal software (Tiensuunnittelutoimisto 1987, Kehittämiskeskus 1991). —
Pursula & Peltola (1997, 1998) have presented an overview of the capacity analysis
and design practice of Finnish unsignalized intersections.
The new intersection design manual (Tie- ja liikennetekniikka 2001) presents a set
of charts for capacity estimation in planning applications. The level-of-service clas-
sification follows the 1985 HCM (Transportation Research Board 1985, Lyly 1988).
Roundabouts are analyzed as series of T-intersections (Tie- ja liikennetekniikka 2001).
For the operational analysis of unsignalized intersections the old manual method
(Tiensuunnittelutoimisto 1978) and Capcal are still used. The old critical gaps
have not been updated, even though new measurements have suggested lower values
(Uusiheimala 1995, Niittymäki & Uusiheimala 1996).
INTRODUCTION
The Swedish Capcal has been updated to Capcal 2 and further to Capcal 3. Capcal 21
updated the calculation procedures, particularly for the vehicle-operating and emission
costs (SNRA 1995b, Hagring 1997). Small changes were also made to critical gaps
and to estimation of service times (SNRA 1995b). Capcal 3 is a Windows version
reproducing the methodology of Capcal 2. A new version is under development.
1.3 Research objectives

After the development of the current Finnish capacity estimation procedure in the 1970’s
the performance of both drivers and vehicles has changed, and new calculation methods
have been developed. It has become necessary to update the analysis methodology of
Finnish unsignalized intersections.
The methodological basis in this research is gap acceptance theory. It is widely used
worldwide. The most important advantage of the model is, however, in the parameter
estimation. Statistical capacity models require observations from congested intersec-
tions.2 Such data are difficult to obtain and were not available for this research. The
same problem has been reported in Denmark (Vejdirektoratet 1999a) and in Sweden
(Hagring 1996b). Parameters of stochastic (gap acceptance) models can be estimated
under uncongested conditions also, although there is always an element of error, when
results from one traffic condition are applied to another.
As Brilon, Wu & Bondzio (1997) have remarked, one cannot be sure that linear re-
gression models apply also in the areas of a capacity-flow diagram where only a few
measurement points have been observed. The structure of a gap acceptance model
is informative, and parameters have a physical meaning. This gives credibility to the
model even in those areas of the diagram where empirical evidence is scarce. It is easier
to transfer a gap-acceptance model to conditions different from a regression model.
The present report gives updated critical gap values and suggests a performance analysis
procedure for Finnish unsignalized intersections. New field measurements have been
made to update the methodology. The theoretical results and the results of three inter-
national methodologies—HCM2000 (Transportation Research Board 2000), Capcal 2
(SNRA 1995b), and Danish DanKap (Vejdirektoratet 1999b)—have been compared
with the capacities and delays simulated by Hutsim (Kosonen 1996, Kosonen 1999).
Based on these results a recommendation is made for a new analysis methodology.
Major attention has been given to capacity analysis methodology, because capacity
estimates have a central role in the estimation of other performance measures, such as
delays and queue lengths. In addition, delay estimates are very sensitive to capacity
estimates, especially at high degrees of saturation (Hagring 1998b). The report includes
some new theoretical results, especially for the capacity analysis of Rank 3 and Rank
4 streams. These results were made available in a preliminary report (Luttinen 2003),
and are presented here with some modifications.
1 The version referred to in this document is Capcal Dos 2.10.

2Akçelik (1998) has, however, suggested the estimation of capacity based on measuring departure flow
rates during queued portions of gap-acceptance cycles.
GAP ACCEPTANCE THEORY
2 GAP ACCEPTANCE THEORY

2.1 Relative priority of traffic streams
Unsignalized intersections can be either uncontrolled, yield-sign controlled, stop-sign
controlled, or roundabouts. Uncontrolled intersections follow the right-before-left
rule. The priority is implicit, given in the rules of the Highway Code only (OECD
Road Research Group 1974). Each vehicle must give way to vehicles approaching
from the right. Left turning vehicles must give way to opposing traffic.
At yield-sign controlled intersections the yielding approaches are referred to as minor-
road approaches. The approaches that are not controlled by yield signs are referred
to as major-road approaches. All minor-road vehicles yield the right-of-way to major-
road vehicles. Left-turning vehicles in all approaches yield right-of-way to opposing
through and right turning vehicles.
At stop-sign controlled intersections1 the minor-road approaches are controlled by stop
signs. The principles of operation are otherwise similar to yield controlled intersections,
but minor road vehicles must stop at the stop line before proceeding.
Roundabouts have a circulating roadway around a central island. All vehicles entering
a roundabout must yield to circulating vehicles. All vehicles entering or exiting a
roundabout make a right turn.
The theory describing the operations at an unsignalized intersection is called gap accept-
ance theory. It includes four basic elements (cf. Transportation Research Board 2000):
1. Relative priority of traffic streams
2. Distribution (availability) of gaps in higher priority streams
3. Usefulness of gaps to yielding vehicles (gap acceptance)
4. Queue discharge
The availability of gaps is described by a probability distribution of headways in higher

priority streams. A gap acceptance function describes the usefulness of headways as
well as queue discharge. The relative priority of traffic streams will be discussed next.
In an uncontrolled intersection of two one-way roads displayed in Figure 2.1 the south-
bound traffic (q11 ) must yield to east-bound traffic (q2 ). The intersection operates in
the same way, if the northern approach has a yield sign. (In Finland a yield sign is not
used in this case, because it is not necessary.) If there is a stop sign, vehicles must stop
before passing the sign. An intersection of two one-way roads is the simplest case to
analyze, and it is used in the derivation of potential capacity and delay equations.
The traffic steams at a four-leg intersection are presented in Figure 2.2. The num-
bering follows HCM2000 (Transportation Research Board 2000). Streams 13–16 are
pedestrians crossing the road.
The operation is more complicated than in the simple case of Figure 2.1. All minor
road vehicles yield the right-of-way to major road vehicles. Left-turning vehicles in all
approaches yield right-of-way to opposing through and right-turning vehicles. There
is a hierarchy of priorities. Through and right-turning movements on the major road
1 HCM2000 (Transportation Research Board 2000) calls this type of intersection two-way stop-
controlled (TWSC). All-way stop-controlled (AWSC) intersections,widely used in the U.S., are not used
in Finland.
11
Figure 2.1: An intersection of two one-way roads
12 11 10
16
6
5
4
13 14
1
2
3
15
Rank Traffic stream

1 2, 3, 5, 6
2 1, 4, 9, 12 789
3 8, 11
4 7, 10
Figure 2.2: Traffic streams at a four-leg intersection

(2, 3, 5, and 6) have absolute priority. This group of movements is called Rank 1. Left
turning traffic from the major road (1, 4) and right turning traffic from the minor road
(9, 12) have Rank 2. They must give priority to Rank 1 movements. Through traffic
on the minor road (8, 11) has Rank 3. It must give priority to all movements on the
major road (Ranks 1 and 2). Finally, left turning traffic from the minor road (7, 10) is
subordinate to all other streams, and it has Rank 4.
All yielding vehicles, which do not find an acceptable lag at arrival must decelerate
and/or stop to wait for the next safe gap. There may, however, also be vehicles or
pedestrians of higher rank waiting for an acceptable gap. Accordingly, a vehicle must
find an acceptable gap or lag after all waiting higher rank vehicles and pedestrians have
passed the conflict area.
Right turning stream 3 does not have a conflict with minor road streams 7, 8 and, 9.
It is, however, possible that minor road drivers are not confident that a major street
vehicle will turn right, until they see a clear indication of its intentions. Accordingly,
a stream may have an effect on the performance of another stream, even though no
conflict exists between the two streams.
The traffic streams at a T-intersection are presented in Figure 2.3. There are only three
hierarchical levels of priorities (ranks). Consequently, there are less possible conflicts
between streams, so that it is easier for a driver to observe the safe gaps in priority
streams.
5
4
13 14
2
3
15
Rank Traffic stream

1 2, 3, 5 7 9
2 4, 9
3 7
Figure 2.3: Traffic streams at a T-intersection
If the major road has a wide central island, one or more minor-road vehicles can be
stored between the two directions of major-road traffic streams. In this case minor
road through or left-turn vehicles may operate in two stages, if there is storage space
available (Brilon, Wu & Lemke 1996, Brilon & Wu 1999, Luttinen 1998)
In HCM2000 (Transportation Research Board 2000) pedestrian flows (15, 16) across
the minor road have Rank 1 and pedestrian flows across the major road (13, 14) have
Rank 2. HCM2000, however, states that the priority of pedestrian movements with
respect to vehicular movements may be a policy issue varying by jurisdiction. In the
Finnish guidelines (Tiensuunnittelutoimisto 1978) minor street movements are con-
sidered subordinate to pedestrian flows across the minor road. Pedestrian flows across
the major road are ignored.—The capacity and level of service of pedestrian facilities
is discussed e.g. in Chapter 18 of HCM2000 (Transportation Research Board 2000).

Current Finnish highway code states that vehicles must give priority to a pedestrian
who is on a crosswalk or is about to set foot on the crosswalk. A vehicle may not
pass without stopping another vehicle that has stopped in front of a crosswalk. A
turning vehicle must give priority to pedestrians crossing the road, where the driver is
turning to. However, the priorities between pedestrians and vehicles observed in reality
cannot be described with clear rules. Because pedestrians have a “higher” priority over
turning vehicles than over through vehicles, pedestrian flow 15 has a higher priority
over movement 3 (Rank 1) than over movement 8 (Rank 3). This breaks the hierarchical
structure of priorities. In addition, these rules are often not observed by or even known
to pedestrians and vehicle drivers (see also Brilon & Wu 2001, Brilon & Wu 2002).
Consequently, the Rank of pedestrian flows is not defined in Figure 2.2. The effect of
pedestrians should, however, be considered in the capacity calculations.
According to the hierarchy in Figures 2.2 and 2.3 the major street through vehicles have
absolute priority. They may, however, experience delays due to three reasons:
1. If major road through vehicles cannot pass queuing left-turn vehicles, they have
to wait for the clearance of the queue (Kimber & Coombe 1980, Bonneson &
Fitts 1997, Bonneson & Fitts 1999).
2. Even without queues a through vehicles may have to decelerate, if it follows a

vehicle decelerating for a turning maneuver.
3. At high traffic volumes major-road vehicles may be slightly delayed to accom-

modate minor-road vehicles (Troutbeck 1990, Troutbeck & Kako 1997, Trout-
beck & Kako 1999).
Wang (2000) has found that major and minor-stream traffic volumes do not have a sig-
nificant effect on critical gaps. However, if major-stream vehicles adjust their speeds
to accommodate minor-stream vehicles, the result can be observed also in the avail-
ability of gaps (headway distribution), not only in the gap acceptance behavior (see
Troutbeck 1997b, Troutbeck & Kako 1999).
At uncontrolled intersections all vehicles obey the right-before-left rule. It is possible
to have a situation where priorities are unspecified. Such a situation occurs, when
vehicles of streams 2, 4, and 7 simultaneously wait for an opportunity to proceed. Each
vehicle must yield to another vehicle, none has priority over all others. The methods
described in this report cannot be applied to uncontrolled intersections. The German
manual (FGSV 2001) suggests 600–800 pc/h as the sum of maximum flow rates of all
four approaches at an uncontrolled intersection.
At roundabouts (Fig. 2.4) vehicle circle counterclockwise. Approaching flows (qa ) give
priority to circulating flows (qc ). This ensures an uninterrupted flow in the circulating
roadway. Circulating and approaching flows merge immediately at the entrance to the
circulating roadway. Each vehicle must make two right turns. All other movements
are eliminated. As a subordinate vehicle enters the circulating roadway it becomes a
priority vehicle.
In Finland the geometric features of most roundabouts are designed so that speeds are
reduced to 20–40 km/h. Safety is the most important factor in the design, even if it
increases delays. Low speed of circulating traffic makes it easy and safe for approaching
vehicles to enter the roundabout.
qc
qa
Figure 2.4: Traffic streams at a roundabout
Because vehicles entering a roundabout make a right turn, the operation can be assumed
to be similar with the operation of right turning streams at yield (or stop) controlled
intersections. The intersection angle is, however, usually smaller in roundabouts, which
makes the operation similar to a (very) short merge area. If a roundabout is analyzed as
a series of unsignalized T-intersections, the roundabout must be so large that the merges
can be considered independent. According to the German manual (FGSV 2001), this
approach can be used for roundabouts having an inscribed circle diameter larger than
26 meters. For other cases Troutbeck (1991) suggests that this approach can be used
in an iterative manner, which quickly converges to solutions.
If pedestrian volumes are high, pedestrian crossings at roundabouts should be grade
separated. If this is not the case, crosswalks should be located so that exiting vehicles
stopped in front of a crosswalk do not disturb the circulating flow.
2.2 Definition of priority streams

A vehicle, whose driving path crosses or merges with a higher priority movement,
cannot enter an intersection, until these higher priority movements provide a safe gap.
For a minor stream k the set Ik includes all conflicting streams that have a higher
priority than stream k. For example, at a four-leg intersection (Fig. 2.2) the set of
conflicting priority streams for minor stream 8 includes streams 1, 2, 4, 5, and 6; i.e.,
{1, 2, 4, 5, 6} ∈ I8 . It is, however, necessary to take a closer look at the definition of
priority streams.
In the discussion above, it was observed that pedestrian flows do not conform to the
hierarchy of priorities of vehicular flows. It was not possible to assign a Rank to ped-
estrian flows. In section 3.8 the effect of pedestrian flows on vehicular capacities is
estimated using the fraction of pedestrians that have priority over the subject minor
stream. This fraction must be estimated for each location on the basis of field meas-
urements or knowledge of local conditions. Pedestrians are not included in the major
flow of a gap acceptance process.
So far it has been assumed that drivers know which higher priority vehicles are arriving
to the conflict area, and which are using a nonconflicting lane or are turning into an-
other direction. Special questions arise with major road right turning vehicles, through
vehicles on multi-lane highways, and circling vehicles on multi-lane roundabouts as
well as exiting vehicles on roundabouts.
Not all right turning vehicles display a turning sign. Some of those that do so, do
so very late. Even when the turning sign is displayed, minor-road drivers may be
hesitant and wait until there are also other indications that the major-road vehicle is
indeed turning. Accordingly, some part of the major road right-turn stream could be
included in the major flow, if it approaches the intersection on a lane sharing a through
movement in conflict with the subject movement. In many current methods half of the
right-turn streams (3 or 6) is included in the major stream. This principle is followed
in HCM2000 (Transportation Research Board 2000) and HBS 2001 (FGSV 2001),
and in the Norwegian manual (Statens Vegvesen 1985). On the other hand, Capcal 2
(SNRA 1995b) follows the old Swedish method (Statens vägverk 1977) and includes
in the major flow only those streams which are in actual conflict with the subject
movement. The same principle is followed in DanKap (Vejdirektoratet 1999b).
At roundabouts exiting vehicles have a similar role as major-road right turning vehicles
at three and four-leg intersections. According to HCM2000 at most well-designed
roundabouts the effect of vehicles exiting into the road, where subject vehicles are
entering, can be ignored. Exiting flows are also ignored in HBS 2001, Capcal 2, DanKap
and in the Norwegian manual. According to Hagring (1998b) most research reports
have not found any significant effect due to exiting flows. This was also his own
conclusion.
If a minor stream merges with a multi-lane stream, only that part of the higher priority
stream should be counted, which are in conflict with the subject stream. Traffic flows
on other lanes may have an impact on minor-stream drivers, but it is usually ignored.
This principle is followed in HCM2000, HBS 2001, Capcal 2, and DanKap.
If a major-road right turning movement (3 or 6) is separated by a triangular island and
face a stop or yield sign, priority is reversed. In this case major-road right turning
vehicles no longer have Rank 1, but they should be analyzed as a Rank 2 movement at
a T-intersection.
2.3 Gap availability
2.3.1 Gaps, lags, and headways
The time distance from front bumper to front bumper between two consecutive vehicles
passing an observation point is called a headway. It is the sum of the time used by
a vehicle to pass the observation point (occupancy time) and the time interval (gap)
to the arrival of the next vehicle (Luttinen 1996). At unsignalized intersections the
minor-road drivers consider gaps (unoccupied time intervals) as possible opportunities
for their crossing or merging maneuvers. However, it is easier to describe traffic flow
in terms of headways than in terms of gaps and occupancy times.2 In this discussion
gap is the headway between two consecutive major stream vehicles evaluated by a
minor-stream driver for the purpose of crossing or merging. The availability of gaps is
described by headway distributions of higher priority streams; i.e., gaps are considered
to be equal to headways (Troutbeck & Brilon 1997). It is assumed that the differences
in major stream vehicle lengths and speeds (i.e., occupancy times) can be ignored.
Mathematically the arrival of vehicles is described as a point process, where each
arrival is a point in the time axis (Fig. 2.5). Because vehicle length is ignored, gaps
and headways are equal. Headway (gap) tn is the time interval between vehicles n − 1
and n. Each headway is connected to the vehicle which terminates the time interval.
From the point of view of a minor-road driver it is thus possible to consider traffic flow
on a major road as a succession of moving gaps (Raff 1950, Buckley 1962).
Vehicle count
0
Time
t1 t2 t3 t4
0 τ1 τ2 τ3
Figure 2.5: Vehicle arrivals as a point process (Luttinen 1990)
Let us assume that a vehicle enters a stop line or a yield line at time τ , the preceding
priority vehicle has passed the conflict point at time τi , and the next priority vehicle will
pass at time τi+1 (Fig. 2.6). The time from the passage of the previous priority vehicle,
B(τ ) = τ − τi , is called backward waiting time or current life (Karlin & Taylor 1975).
The time to the passage of the next priority vehicle, ϒ(τ ) = τi+1 − τ , is called forward
waiting time or excess life in the theory of stochastic processes and lag in traffic flow
theory. The current headway at time τ is the headway of priority stream vehicle i + 1;
i.e., T (τ ) = Ti+1 = τi+1 − τi = B(τ ) + ϒ(τ ).
Some of the major road gaps are not accepted by minor road vehicles, some gaps are ac-
cepted by one vehicle, and some gaps allow the entry of several vehicles. Consequently,
more detailed information than a deterministic headway model ( ∀i : ti = t¯ ) is required
(see Watson 1933, Clayton 1941, Catchpole & Plank 1986). For realistic estimation of
capacity and delays it is necessary to have a realistic model of the frequency of gaps
of different lengths. In mathematical models the availability of gaps is described by a
probability distribution of vehicle headways (see Luttinen 1996).
A cumulative distribution function gives the probability that a randomly3 selected head-
2 Headway can be considered as the inverse individual flow rate of each vehicle (Dawson & Chimini

1968). The sum of headways τn = ni=1 ti is the time interval between the arrivals of vehicles 0 and
n, so that the number of headways (vehicles) divided by the sum of headways gives a proper estimate of
flow rate (λ̃ = nτn−1 ).
3 It should be emphasized that the headway that occurs for a randomly arriving vehicle or pedestrian
is not randomly selected. It is more probable that a random observer arrives during a long headway
than during a short headway. This length-biased sampling is the reason for the well-known waiting time
Number of priority vehicles
B(τ ) ϒ(τ )
i+1
i−1
τi−1 τi τ τi+1 Time
Figure 2.6: Backward waiting time B(τ ) and lag ϒ(τ ) (Luttinen 1990)
way is not larger than a given value
t
F(t) = P{T < t} =f(u) d(u) (2.1a)
∞ 0
F(∞) = P{T < ∞} = f(u) d(u) = 1, (2.1b)

0
where f(t) is the probability density function. Because headways cannot have negative
length, the lower limit of integration is zero. In conventional probability analysis F(t)
is defined as P{T ≤ t}. The definition given above is more useful in gap acceptance
theory. For distributions continuous at t, both definitions give identical numerical
results.
The probability that a randomly selected headway is larger than a given value is called
survivor function
∞
R(t) = P{T ≥ t} = f(u) d(u) = 1 − F(t) . (2.2)
t
In reliability and life data analysis this function gives the probability that an object
survives or operates reliably for a given time (see e.g. Nelson 1982, Crowder, Kimber,
Smith & Sweeting 1991).
The expectation E[T ] is the average headway from the distribution:
∞
E[T ] = t¯ = tf(t) dt. (2.3)
0
Because f(t) = 0 when t < 0, expectation can also be expressed in terms of the
paradox (Feller 1966, Daganzo 1997).

survivor function (Grimmett & Stirzaker 2001):4

∞ ∞
E[T ] = t¯ = [1 − F(t)] dt = R(t) dt. (2.4)
0 0
If we count n consecutive headways, the expected duration of the count is E[τn ] = nt¯.
Flow rate (λ = q/3600) is the inverse of average headway: λ = n(E[τn ])−1 =
t¯−1 . This is an unconventional definition of flow rate, but it helps to demonstrate the
relationship between flow rate and average headway. For a more detailed analysis of this
relationship the reader is referred to the discussions by Haight (1963) on synchronous
and asynchronous counting and by Luttinen (2001) on doubly synchronous counting,
as well as to a discussion on renewal processes in any standard textbook on stochastic
processes (e.g., Ross 1996).
The lags experienced by randomly arriving vehicles have a different distribution than
headways. If R(υ) = P{T > υ} is the survivor function of the headway distribution
and E[T ] is the expected headway, the probability density function of lags can be
expressed using the Palm-Khintchine equation (Palm 1943, Khintchine 1960, Cox &
Isham 1980):
R(υ)
fϒ (υ) = = λR(υ) . (2.5)
E[T ]
Cox & Miller (1965) have presented an application of this equation for the estimation
of delays at unsignalized intersections.
The survivor function of lags is obtained directly from equation (2.5) as
∞
Rϒ (υ) = λ R(y) dy. (2.6)
υ
The expected value of a lag can be expressed in terms of the moments of the headway
distribution:
∞
E[ϒ] = υfϒ (υ) dυ
0
∞
1
= υR(υ) dυ
E[T ] 0
∞ ∞
1
= υ f(y) dy dυ
E[T ] 0
∞ yυ
(2.7)
1
= υ dυ f(y) dy
E[T ] 0 0
∞ 2
1 y
= f(y) dy
E[T ] 0 2

E T2 t¯ σ 2
= = + T.
2 E[T ] 2 2t¯
See Kleinrock (1975) for a different approach.
4 This equation can be proved by integration with change of order (Ross 2002)
∞ ∞ ∞
R(t) dt = f (u) du dt
0 0 t
∞ u
= dt f (u) du
0 0
∞
= uf (u) du.
0
For a geometric interpretation see Gnedenko (1962).
2.3.2 Geometric distribution

Winsten (in Beckmann et al. 1956) has estimated the delay at an unsignalized intersec-
tion considering time in finite sized steps. If the probability of an arrival during a time
interval of unit length is p, the probability of a headway of k time units is (1 − p)k−1 p.
During one time unit, the number of arrivals can be 0 or 1. The probability p of an
arrival can be interpreted as the flow rate; i.e., number of arrivals per time unit. The
arrival process is binomial, and headways follow the geometric distribution, which has
density, distribution, and survivor functions as follows:
f(k) = P{K = k} = (1 − p)k−1 p (2.8a)

k
k−1
F(k) = P{K ≤ k} = p (1 − p)i−1 = p (1 − p)i = 1 − (1 − p)k (2.8b)
i=1 i=0
R(k) = P{K > k} = 1 − F(k) = (1 − p)k , (2.8c)
where p is the probability of arrival during the minimum time interval, and K is the
headway (measured in discrete time intervals). The expected length of a headway is
(Matloff 1988)
∞

E[K] = p iq i−1
i=1
∞
d i
=p q
dq i=1

d 1 (2.9)
=p −1
dq 1 − q
1
=p
(1 − q)2
1
= ,
p
where q = 1 − p. Average headway is the reciprocal of the flow rate, as expected.

If headways are i.i.d. and a randomly selected headway is less than t with probab-
ility p = F(t), the probability of k consecutive headways T < t follows geometric
distribution:
pk = p k (1 − p) = [F(t)]k R(t) . (2.10)
The expected number of consecutive headways T < t is
∞

E[K] = (1 − p) ipi
i=1
∞
= (1 − p)p ip i−1
i=1
1 (2.11)
= (1 − p)p
(1 − p)2
p
=
1−p
F(t)
= .
R(t)
Geometric distribution is a discrete-time headway model. Modern microscopic traffic

models are, however, based on continuous-time headway distributions, the most im-
portant of which will be discussed below. Even in continuous-time models geometric
distribution can be used to estimate probabilities and expectations of i.i.d. events, as
shown above.
2.3.3 Negative exponential distribution

If vehicles are assumed to move at their desired speeds and independently of each other,
traffic volumes follow Poisson distribution, and headways are exponentially distributed.
The probability density, cumulative distribution, and survivor functions are
f(t) = λe−λt (2.12a)

−λt
F(t) = 1 − e (2.12b)
−λt
R(t) = 1 − F(t) = e , (2.12c)
where λ is the scale parameter (see Fig. 2.7). The expected headway is equal to the
inverse scale parameter:
∞ ∞
1
E[T ] = R(t) dt = e−λt dt = . (2.13)
0 0 λ
This indicates that the scale parameter is equal to the flow rate expressed in veh/s; i.e.,
λ = q/3600.
Negative exponential distribution is the interarrival time distribution of totally random
arrivals. The counting process follows the Poisson distribution. At low flow rates the
negative exponential density function is highest, and distribution function (Fig. 2.7)
rises steeply. For long headways the density approaches asymptotically zero and the
distribution function unity. This indicates both a substantially high probability of very
short headways, even at low flow rates, and the possibility of very long headways, even
at high flow rates.
Several authors (e.g., Weiss & Herman 1962, Breiman 1963, Thedéen 1964) have
demonstrated that under rather weak assumptions (no vehicle interaction, identically
and independently distributed speeds) the number of vehicles in an arbitrary time inter-
val will be asymptotically Poisson distributed as time tends to infinity. Thus, headways
will asymptotically follow the exponential distribution. If the combined headways of
several independent streams passing a reference point are considered, the limitations
of the Poisson assumption become less severe.
Exponential distribution has the Markov (“lack of memory”) property (Luttinen 1990):
e−λ(b+υ)
P{ T > b + υ | T > b } =
e−λb
−λυ (2.14)
=e
= P{T > υ} ,
where b is the current duration of headway (backward waiting time) and υ is the lag
(Fig. 2.6). This indicates that lags have the same distribution as headways:
RT (υ)
fϒ (υ) = = λe−λυ . (2.15)
E[T ]
In fact, exponential distribution is the only continuous-time distribution that has the
Markov property (Feller 1957).
0.8
0.6
F(t)
0.4
0.2
0
1400
1000 10
8
600 6
4
200 2
Flow rate (veh/h) 0
Headway (s)
Figure 2.7: Negative exponential cumulative distribution function for vehicle headways
Because of the Markov property it is not necessary to keep track of the time elapsed since
the last arrival. This makes the Poisson process mathematically tractable (Tijms 2003).
Therefore, the negative exponential distribution is often preferred over other, more
realistic but also more complex headway distributions. In many cases explicit solutions
can only be found when the arrival process is assumed Poisson. Since Adams (1936)
the negative exponential distribution has played a central role in traffic flow theory.
If the arrival process is Poisson and the number of arrivals during a given time interval
(0, τ ] is n, the distribution of the arrival times is similar as the distribution of n uniformly
distributed events in time interval (0, τ ] (Karlin & Taylor 1981, Tijms 2003). This indic-
ates that Poisson arrivals are completely random in time. An important consequence of
this is the PASTA (Poisson Arrivals See Time Averages) property (Wolff 1989). When
the arrival process is Poisson, the average state of a system during arrival times is equal
to the time average state of the system.
The exponential model has two major limitations: The model allows unrealistically
short headways, and it does not describe platooning. The model gets more distorted
as flow rates increase. Consequently, the exponential distribution can be considered
as a realistic headway model under very low-flow conditions only, approximately q <
150 veh/h (Luttinen 1996).
The exponential distribution can also be used to describe the headways of free vehicles,
which are not driving in platoons. This indicates that empirical headway distributions
have an “exponential tail” (Luttinen 1996). Because very short headways are not likely
to be “free”, the exponential distribution should be modified.
2.3.4 Shifted exponential distribution

If headways have a minimum length of tp , and headways larger than tp are exponen-
tially distributed, the resulting model is called shifted exponential distribution. It has
probability density
0, if t < tp
f(t) = −θ(t−tp )
(2.16)
θe , if t ≥ tp ,
where θ is the scale parameter and tp is the location parameter. The distribution function
(Fig. 2.8) is
0, if t < tp
F(t) = −θ(t−tp )
(2.17)
1−e , if t ≥ tp ,
and the survivor function is

1, if t < tp
R(t) = −θ(t−tp )
(2.18)
e , if t ≥ tp .
0.8
0.6
F(t)
0.4
0.2
0
1400
1000 10
8
600 6
4
200 2
Flow rate (veh/h) 0
Headway (s)
Figure 2.8: Shifted exponential cumulative distribution function with minimum headway tp =
1s
Shifted exponential distribution is the conditional distribution of exponential headways

larger than tp :
P{T > t} e−λt

P T > t T > tp = = −λt = e−λ(t−tp ) , tp ≤ t. (2.19)
P T > tp e p
This distribution avoids the problem of extremely short headways, but does not model
platooning.
The shape of the distribution function is similar to the negative exponential distribution
(Fig. 2.7), but the probability of headways smaller than the minimum headway (tp ) is
null. After the minimum headway the distribution function rises even steeper than the
negative exponential distribution function, and approaches asymptotically unity. The
model does not include platooning, but the density is highest at headways just larger
than the minimum headway.
The expected headway is
∞ tp ∞
1
E[T ] = R(t) dt = 1 dt + e−θ(t−tp ) dt = tp + . (2.20)
0 0 tp θ
Scale parameter θ can now be expressed in terms of flow rate as

1 λ q
θ= = = . (2.21)
E[T ] − tp 1 − λtp 3600 − qtp
Thus, the average flow rate of an arrival process with shifted exponential time headways
is
1 θ
λ= = . (2.22)
E[T ] 1 + θ tp
Oliver (1961) has derived the explicit counting distribution.
The probability density function of the lag distribution is

 θ

 , if υ < tp
R(υ) θ R(υ) 1 + θ tp
fϒ (υ) = = = θe−θ(υ−tp ) (2.23)
E[T ] 1 + θ tp   , if υ ≥ tp .

1 + θ tp
The survivor function of the lag distribution can be obtained by integrating the density
function. When υ < tp , the survivor function is
tp ∞
θ 1
Rϒ (υ) = dy + θe−θ(υ−tp ) dy
1 + θ tp υ 1 + θ tp tp
θ (tp − υ) 1
= + (2.24)
1 + θ tp 1 + θ tp
θυ
=1− , υ < tp .
1 + θ tp
When υ ≥ tp , the survivor function is
∞
1
Rϒ (υ) = θe−θ(υ−tp ) dy
1 + θ tp υ
(2.25)
e−θ(υ−tp )
= , υ ≥ tp .
1 + θ tp
Consequently, the survivor function is

 θυ

1 − , if υ < tp
1 + θ tp
Rϒ (υ) = (2.26)

 e−θ(υ−tp )
 , if υ ≥ tp .
1 + θ tp
When the location parameter (minimum headway) is zero, we obtain the survivor func-
tion of negative exponential distribution, as expected.
Once the survivor function of the lag distribution is known, the expected lag is obtained
as
∞
E[ϒ] = Rϒ (υ) dυ
0 tp ∞
θυ 1
= 1− dυ + e−θ(υ−tp ) du
0 1 + θ tp 1 + θ tp tp
(2.27)
θ tp2 1
= tp − +
2(1 + θ tp ) θ (1 + θ tp )

1 1
= (1 + θ tp ) + .
2θ (1 + θ tp )
Equation (2.7) gives naturally the same result, when the substitutions for mean t¯ =
tp + θ −1 and variance σT2 = θ −2 of the shifted exponential headway distribution are
made. When tp = 0, the result is equal to the expectation of negative exponential
distribution (θ −1 ).
2.3.5 Erlang distribution

Another simple model used in the analysis of unsignalized intersections is the Er-
lang distribution (Erlang 1917a, Erlang 1917b), which is a special case of the
gamma distribution. It is easily tractable in mathematical analysis, but not as
easy as the exponential distribution, and not particularly good as a headway model
(Luttinen 1991, Luttinen 1996). The Erlang distribution has not been used in this
research, but it has been applied in the old Swedish method (Nordqvist, Bång &
Hansson 1973, Statens vägverk 1977). See also Yeo (1962), Voigt (1973), and Pre-
vedouros (1988).
2.3.6 Cowan’s M3 distribution

A headway distribution can also be described as a mixture of follower and free-vehicle
headways. If headways in platoons are assumed to be constant (tp ) and free-vehicle
headways follow shifted exponential distribution, the cumulative distribution function
is obtained as (Cowan 1975)

0, if t < tp
F(t) = −γ (t−tp )
(2.28)
1 − φe , if t ≥ tp ,
where φ is the proportion of free-vehicle headways, and γ is the scale parameter.
Buckley (1962) called this a regular-random distribution, but nowadays it is most often
known as Cowan’s M3 distribution according to (Cowan 1975). The survivor function
is
1, if t < tp
R(t) = −γ (t−tp )
(2.29)
φe , if t ≥ tp ,
The expectation is
∞ tp ∞
φ
E[T ] = R(t) dt = 1 dt + φ e−γ (t−tp ) dt = tp + . (2.30)
0 0 tp γ
Maximum flow rate qmax = 3600 (min E[T ])−1 = 3600tp−1 is attained when all vehicles
move in platoons (φ = 0). The method-of-moments estimator for the scale parameter
is obtained as
φ φq
γ̃ = = . (2.31)
E[T ] − tp 3600 − qtp
Because the proportion 1 − φ of vehicles are followers and have a headway of size tp ,
the density function has a mass of 1 − φ at tp . This point mass can be described using
a unit impulse (Dirac delta) function (Spanier & Oldham 1987)

∞
∞ if t = tp
δ(t − tp ) = and δ(t − tp ) dt = 1, (2.32)
0 otherwise 0
which gives the probability density function (Plank 1982, Luttinen 1999)

0, if t < tp
f(t) = −γ (t−tp )
(2.33)
(1 − φ)δ(t − tp ) − φγ e , if t ≥ tp .
Figure 2.9 displays the coefficient of variation
√ √
var(T ) φ(2 − φ)
C(T ) = = (2.34)
E[T ] γ tp + φ
of Cowan’s M3 distribution. It is zero for deterministic headways (φ = 0), reaches the
maximum [γ tp (2 + γ tp )]−1/2 at φ = γ tp (1 + γ tp )−1 , and decreases to (1 + γ tp )−1 as
φ reaches unity and the distribution becomes shifted exponential (Luttinen 1999). At
low values of γ the peak rises above unity, but it stays below unity when γ is large.
Although M3 gives a very crude approximation of short headways, it can produce
moment characteristics very similar to real headway distributions (Luttinen 1999).
1.5
γ = 0.1
γ = 0.2
Coefficient of variation
1
γ = 0.3
γ = 0.4
γ = 0.5
0.5
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Proportion of free headways, φ
Figure 2.9: Coefficient of variation of Cowan’s M3 distribution with tp = 1.8 s (Luttinen 1999)
The density function of the lag distribution is

 γ

 if υ < tp
R(υ) γ R(υ) φ + γ t ,
p
fϒ (υ) = = = φγ e−γ (υ−tp ) (2.35)
E[T ] φ + γ tp   , if υ ≥ tp .

φ + γ tp
The survivor function of the lag distribution can be obtained by integrating the density
function. When υ < tp , the survivor function is5
tp ∞
γ φ
Rϒ (υ) = dy + γ e−γ (υ−tp ) dy
φ + γ tp υ φ + γ tp tp
γ (tp − υ) + φ
= (2.36)
φ + γ tp
γυ
=1− , υ < tp .
φ + γ tp
When υ ≥ tp , the survivor function is

∞
φ
Rϒ (υ) = γ e−γ (υ−tp ) dy
φ + γ tp υ
−γ (υ−tp )
(2.37)
φe
= , υ ≥ tp .
φ + γ tp
The survivor function can now be expressed as

 γυ

1 − φ + γ t , if υ < tp

p
Rϒ (υ) = φe−γ (υ−tp ) (2.38)

 if υ ≥ tp .
 ,
φ + γ tp
When the proportion (φ) of free headways is unity, we obtain the survivor function of
shifted exponential distribution.
Because the flow rate λ (veh/s) is the inverse of the expected headway E[T ], the lag
density can be expressed as (Hagring 1998a):

λ, if υ < tp
fϒ (υ) = λR(υ) = −γ (υ−tp )
(2.39)
λφe , if υ ≥ tp .
The survivor function of the lag distribution can likewise be expressed as (see Hagring
1998a)

1 − λυ, if υ < tp
Rϒ (υ) = λ −γ (υ−tp ) (2.40)
 φe , if υ ≥ tp .
γ
A model similar to Cowan’s M3 was presented earlier by Tanner (1953, 1961a, 1967)
and (Miller 1961). Tanner modeled traffic flow as departures from an M/D/1 queuing
system (exponential interarrival times and one server with constant service times).
Because the arrival process is Poisson (Khintchine 1960), the proportion of constant
(tp ) headways is equal to the utilization factor of the server (Luttinen 1990)
qtp
1 − φ = P T = tp = λtp = . (2.41)
3600
5 The survivor function can be obtained more easily as follows:
υ
γ γυ
Rϒ (υ) = 1 − Fϒ (υ) = 1 − dy = 1 − , υ < tp .
φ + γ tp 0 φ + γ tp
Using equations (2.30) and (2.41) parameters φ and γ can be expressed as
qtp
φ = 1 − λtp = 1 − (2.42)
3600
φ q
γ = =λ= . (2.43)
1
λ
− tp 3600
The last result indicates that the scale parameter γ of the shifted exponential distribution
of free headways is equal to the scale parameter λ of the negative exponential interarrival
time distribution of the underlying M/D/1 queuing process. The distribution function
is

0, if t < tp
F(t) = −λ(t−tp )
(2.44)
1 − (1 − λtp )e , if t ≥ tp .
The headways are no longer independent; i.e., platoon lengths do not follow geometric
distribution, as in the M3 model, but the Borel-Tanner distribution (Tanner 1953, Tanner
1961a, Tanner 1961b, Haight & Breuer 1960, Prabhu 1965). The autocorrelation of
headways does not have any influence on capacity estimates, but delays are sensitive
to the order of headways (Kyte, Tian, Mir, Hameedmansoor, Kittelson, Vandehey,
Robinson, Brilon, Bondzio, Wu & Troutbeck 1996).
Figure 2.10 displays the cumulative distribution function of headways from an M/D/1
queuing process with service time (i.e., headway in platoon) tp = 2 s. There is a dis-
continuity at T = tp . The probability of shorter headways is null. The proportion of
follower headways increases linearly with increasing flow rate, as indicated by equa-
tion (2.42). When headways are larger than tp , the cumulative distribution function
approaches asymptotically unity. Maximum flow rate is λ = tp−1 or q = 3600tp−1 ,
when all vehicles are followers (φ = 0).
M3 distribution is a good headway model, if there is no need for accurate modeling
of short gaps. Such is the case in the analysis of unsignalized intersections. M3 can
reproduce data with moment characteristics very similar to real headway distributions,
and it gives good results in capacity analysis for unsignalized intersections (Luttinen
1999). Properties of the M3 distribution and the estimation of its parameters have
been discussed by several authors (see Brilon 1988a, Akçelik & Chung 1994, Sullivan
& Troutbeck 1994, Hagring 1996b, Troutbeck 1997b, Luttinen 1999, Tanyel & Yayla
2003).
The M3 distribution has been widely used in the capacity and delay analysis of unsig-
nalized intersections and roundabouts (Akçelik 1994a, Hagring 1996b, Hagring 1998b,
Luttinen 1999, Sullivan & Troutbeck 1994, Troutbeck 1986, Troutbeck 1991, Troutbeck
& Kako 1997), especially in Australia. The headway distribution for a two-lane case;
i.e., a superposition of two M3 distributions, has been derived by Troutbeck (1986) and
Hagring (1998b). See also section 3.3.
More advanced mixed distributions, such as the semi-Poisson distribution (Buckley
1962, Buckley 1968, Luttinen 1994) and the generalized queuing model (Cowan 1975,
Branston 1976), use a specific distribution for headways in platoons and exponential
distribution for free-vehicle headways. These models are more realistic, but they are
also more difficult to use in mathematical analysis (Luttinen 1996).
Most of the current unsignalized intersection models use either negative exponential
distribution or Cowan’s M3 distribution for major stream headways. For a more detailed
description of headway distributions the reader is referred to Luttinen (1996).
0.8
0.6
F(t)
0.4
0.2
0
1400
1000 10
8
600 6
4
200 2
Flow rate (veh/h) 0
Headway (s)
Figure 2.10: Cumulative distribution function of headways from an M/D/1 queuing process
with service time tp = 2 s
2.4 Gap acceptance and queue discharge

The arrivals and departures at an unsignalized intersection can be described by a queuing
model demonstrated in Figure 2.11. When a minor-stream vehicle enters the stop line
or yield line (first position in a queue, time 0 in Figure 2.11) the driver must evaluate
whether the lag in priority streams is safe for entry. If the driver considers the lag
unsafe, the vehicle waits for a safe headway. The minimum acceptable major-stream
headway is assumed to be equal for all vehicles, and it is called the critical acceptance
gap, or simply the critical gap (tc ).
Minor stream entry times
0 ε1 ε2 ε3 (ε3 + tf ) ε4 ε5 (ε5 + tf ) ε6
Wait at stop line tf tf tf Wait at stop line tf tf
(vehicle 1) (vehicle 4)
Block Block Block

Antiblock tc Antiblock tc
Lag υ1 Headway t4
0 τ1 τ2 τ3 τ4 τ5 τ6
Major stream arrival times
Figure 2.11: Queue discharge at an unsignalized intersection
A minor stream driver attempts to maximize her/his utility by accepting a major stream
headway having a safety risk lower than the value of the expected delay resulting from
headway rejection. Critical gap can be regarded as a compromise between the demand
for safe entry to an intersection and for minimizing delay (Hagring 1998b). Gap
acceptance theory is a microscopic (stochastic) theory, which describes the behavior
of individual drivers, as opposed to macroscopic (statistical) models describing traffic
flow at an aggregate level.
Each major-stream vehicle blocks minor-stream vehicles from entering the intersection.
The block, as Raff (1950) called this time interval, starts tc seconds before the major-
stream vehicle arrives and lasts until the beginning of the next major-stream headway
ti ≥ tc (Fig. 2.11). The time interval between two blocks is called an antiblock. An
antiblock starts when a headway ti ≥ tc starts and ends tc before the arrival of the
next major-stream vehicle. Minor-stream vehicles can enter the intersection during
antiblocks. A minor-stream vehicle is assumed to enter the intersection immediately at
the beginning of an antiblock or immediately after having reached the stop line during
an antiblock (Weiss & Maradudin 1962).—Assuming a traffic signal analogy, suggested
first by Winsten (in Beckmann et al. 1956), blocks can be considered red intervals and
antiblocks green intervals.
When a minor stream vehicle enters the intersection, it takes some time before the next
vehicle can take its position at the head of the queue. After this the process is repeated.
The minimum headway between two minor stream vehicles entering the same major
stream gap is called follow-up time (tf ). During an antiblock vehicles from a minor
stream queue enter the intersection separated by follow-up times.6 An interval tf after
the queue has discharged, minor stream vehicles can proceed without delay until the
next block starts. At a yield controlled intersection a vehicle approaching an intersection
or discharging from a queue can proceed without stopping, if the lag is long enough.
The operation can be described as a single server queuing system with deterministic
(tf ) service times. The service is, however, occasionally interrupted, because customers
can enter service during antiblock intervals only. A more comprehensive description
of several conflicting streams would be a single server system with priority queues.
Each customer type (rank) has a separate queue. The first customer in the highest
rank queue is always served first. Switching from one queue to another requires a
“reorientation time” tc − tf (Jaiswal 1968). The hierarchy of conflicts in an entire
unsignalized intersection is, however, too complicated to be described by a single
queuing system. For example, in a four-legged intersection (Fig. 2.2) streams 4 and
8 have a conflict, as do streams 3 and 4, but stream 3 and 8 vehicles can enter the
intersection simultaneously. Queuing models are limited, but nevertheless useful tools
in the analysis of unsignalized intersections.
It is assumed that service starts as a vehicle crosses stop line. Service time is equal to
the follow-up time. When service ends, the next vehicle can enter service (cross stop
line).
6 According toAkçelik (1994a) the first minor stream vehicle enters an intersection t after the beginning
f
of an antiblock. Consequently, the last vehicle can enter the intersection tc − tf before the arrival of the
next major stream arrival. See also Kyte et al. (1996).
According to Tanner (1962) and Yeo & Weesakul (1964) a minor stream vehicle can depart tc after the
last vehicle in a block, and the next block starts when the next major stream vehicle arrives. Mathematically
the model is unchanged if the block starts a constant time u < tc before the first vehicle and terminates
tc − u after the last vehicle (Tanner 1962). By setting u = tc − tf , we obtain the model of Akçelik (1994a).
The model in Figure 2.11 was suggested by Weiss & Maradudin (1962), and it is obtained by setting
u = tc .
In terms of a queuing process, entry time is the time of departure from service. In Figure 2.11 major
stream arrival times can be interpreted as service departure times, while minor stream entry times should
be considered as service entry times. See Figure 3.16 on page 79 for more details.
The reference point of a priority stream is the conflict point between the priority and
minor streams. However, in the gap acceptance process no service times are usually
assumed for priority streams. The entry of minor-stream vehicles is blocked by critical
gaps and follow-up times only.
In Figure 2.11 the first vehicle enters the intersection at the same time as the major
stream vehicle arrives (ε1 = τ3 and ε4 = τ5 ); i.e., passes the conflict point. When the
“service is completed”, the minor stream vehicle passes the conflict point. Thus, tf is
also the minimum time interval between a major stream vehicle and a merged minor
stream vehicle. This is, however, an extension of the basic queuing model.
The service of a minor stream vehicle may overlap the critical gap. In Figure 2.11
vehicle 3 departs at ε3 + tf , when major stream vehicle 4 has already blocked the entry
of minor stream vehicle 4. The minimum headway between a minor stream vehicle
and a major stream vehicle is tc − tf .
This discussion raises three issues concerning time intervals between vehicles.
1. Critical gap (tc ) cannot be shorter than the follow-up time (tf ). Otherwise the
arrival time of the next major stream vehicle (such as τ6 in Figure 2.11) might
take place before a minor stream vehicle has departed from service (ε5 + tf ). This
is an important condition for capacity models, which assume a constant queue
of minor stream vehicles. The condition tf ≤ tc ensures that there is always a
minor stream vehicle waiting at the stop line when an antiblock starts (Plank &
Catchpole 1984).
2. Even if tc > tf , the headway between the merged vehicle and the next major
stream vehicle (τ6 −ε5 −tf ) may become too short, in which case the major stream
vehicle has to decelerate in order to delay its arrival and to keep the headway safe.
Because the priority of major stream vehicles is limited, this type of operation
is called “limited priority merge” (see Troutbeck 2002). Major stream priority
becomes limited, if tc < tf + tp , where tp is the follower headway between major
stream vehicles.—To make the argument simpler, follower headway (headway
between vehicles in a platoon) has been assumed constant.
3. The minor stream follow-up time (tf ) cannot be shorter than the major stream
follower headway (tp ). Otherwise the minor stream queue discharge rate would
be higher than the major stream capacity. The major stream cannot receive
merging vehicles at this rate.
Because drivers and vehicles differ, the minimum gaps and lags accepted by drivers
and the headways between discharging vehicles differ from vehicle to vehicle. Driver
behavior is heterogeneous. In addition, even under similar conditions a driver may
behave differently at different times. A driver may accept a gap that is shorter than a
gap rejected by the same driver earlier. This kind of behavior is called inconsistent.
It is apparent that inconsistency is related to real inconsistency in driver behavior, but
it is possible that most of the observed “inconsistent” behavior can be explained by
situation-specific factors, such as waiting time and variation in speed and type of major
stream vehicles (Hagring 1998b).
In mathematical models driver behavior is usually assumed to be both consistent and
homogeneous. A consistent driver behaves in the same way every time at similar
situations. An inconsistent driver may approve a gap that is shorter than a gap that the
same driver had rejected earlier. An inconsistent driver determines a new critical gap for
each major stream headway (Ashworth 1969, Plank & Catchpole 1984), or assigns to
each gap a probability of crossing (Weiss & Maradudin 1962, Hawkes 1968). Typically
the acceptable headway decreases as the number of rejected gaps increases. Such a
driver is called impatient. (If critical gap increases as the number of rejected gaps
increases, the behavior is not inconsistent according to the definition.) Accordingly,
inconsistency increases capacity.
If drivers are homogeneous, there is no difference in behavior among different drivers
under similar conditions. Homogeneous but inconsistent drivers have the same way of
selecting a critical gap for each individual headway. In the case of heterogeneous and
consistent drivers, each driver (or vehicle type) has an individual critical gap sampled
from a common distribution. This critical gap is applied consistently to all major
stream headways. Vehicles with large critical gaps are more likely to be found at the
head of the queue (Catchpole & Plank 1986), which leads to a decrease in capacity
(Wegmann 1991).
When a minor stream consists of heterogeneous and inconsistent drivers each driver has
an individual way to sample critical gaps for each individual headway or lag. Because
inconsistency increases capacity and heterogeneity decreases it, the overall effect of the
assumption of consistent and homogeneous drivers has been considered to be minimal
compared to the case of heterogeneous and inconsistent drivers (Catchpole & Plank
1986, Troutbeck 1988, Transportation Research Board 1997, Hagring, Rouphail &
Sørensen 2003).
In summary, the major assumptions in basic gap acceptance models are
1. Similar (consistent) behavior of each driver under similar conditions

2. Similar (homogeneous) behavior among all drivers
3. Minor stream vehicles have similar critical gaps in all priority streams and
between vehicles of different priority streams
4. Gap acceptance is similar to lag acceptance.
5. Gaps of all independent priority streams are superimposed into a single stream
of gaps
6. The availability of priority-stream gaps is described by a statistical headway
distribution model
7. Minor-road vehicles arrive randomly
8. Priority stream vehicles do not adjust their behavior to accommodate an entering
vehicle; i.e., gap-acceptance behavior is similar at all traffic volumes
9. Minor stream vehicles can distinguish priority vehicles from vehicles exiting
before the conflict area
It is also assumed that follow-up time is shorter than critical gap. This assures that under
a continuous queue there is always a vehicle waiting for entry as a new gap begins. If
critical gap was shorter than follow-up time, the mathematical models would allow the
possibility of queue discharge headways shorter that follow-up times.
Several models have relaxed one or more of these assumptions. Herman & Weiss (1961)
and Weiss & Maradudin (1962) introduced a gap acceptance probability; i.e., the prob-
ability that a waiting driver accepts a headway of given length. Ashworth (1969) and
Plank & Catchpole (1984) have developed a capacity formula for heterogeneous and
consistent drivers. Catchpole & Plank (1986) and Madanat, Cassidy & Wang (1994)
have analyzed the operation of unsignalized intersections for heterogeneous and incon-
sistent drivers; thus relaxing assumptions 1 and 2. Troutbeck has relaxed assumption
8 and studied gap acceptance under “limited priority merge” (Bunker & Troutbeck
1994a, Bunker & Troutbeck 1994b, Troutbeck 1995, Troutbeck 1997b, Troutbeck &
Kako 1999, Troutbeck 1999).
Tanner (1967) was the first to relax assumption 5. He estimated minor-stream capacity
assuming two or more independent priority streams with different flow rates. See also
Troutbeck (1986). Gazis, Newell, Warren & Weiss (1967), Fisk (1989) and Hagring
(1998b) have extended the model to include the case where different lanes (priority
streams) have different critical gaps (assumption 3).—Only streams without mutual
conflicts can be considered independent. When priority streams are from more than
one rank, special analysis methods are required.
The effect of nonpriority vehicles, such as major-road right turners, can be taken into
a model by adding a proportion of this stream to a priority stream.
In advanced methodologies many of these assumptions have been relaxed, but in the
core of these methods is usually a potential capacity or delay formula with most of these
assumptions. Some adjustments are then presented to make the method more general.
The nature of a problem becomes often clearer when it is compared with other, but
similar problems. In a mathematical sense the following traffic engineering problems
have many similarities with the unsignalized intersection case:7
Pedestrians crossing a road also face a gap acceptance problem. Haight (1963) refers
to the items delayed as vehicles or pedestrians “according as they do or do not form
queues”. Early studies of the pedestrian-crossing problem have been presented by
Adams (1936), Tanner (1951), and Mayne (1954). Cowan (1984) has generalized
Adams’ result for the M3 headway distribution. See also section 4.2.3.
If all waiting pedestrians can cross the road when a headway larger than the critical
gap starts, the follow-up time is zero. This makes the analysis simpler. However, if
all waiting pedestrians can cross the road during a headway larger than the critical
gap, the capacity of a pedestrian crossing is either infinite or zero. Such a model
makes more sense in delay studies than in capacity studies.—Detailed information
of pedestrian gap acceptance and vehicle headway distributions can also be used to
evaluate the accident risk of pedestrians (Cohen, Dearnaley & Hansel 1955, Song,
Dunne & Black 1993, Pasanen 1991).
Tanner (1951) remarked the similarity between the pedestrian crossing problem and the
operation of vehicle-actuated traffic signals, as described by Garwood (1940). Beck-
mann et al. (1956) presented the operation of a stop-controlled intersection in terms of
red (block) and green (antiblock) interval sequences generated by the priority stream
gaps. A minor stream has “green” when conflicting major streams have no “green
demand”. The generation and extension of blocks (Fig. 2.11) is mathematically very
similar to the operation of vehicle detectors (Fig. 2.12). In the theory of stochastic
processes this type of operation is known as a Type II counter (Parzen 1962). Akçelik
(1994a) has described a traffic-signal analogy in modeling gap-acceptance at unsignal-
ized intersections.
7 It is also worth mentioning that queuing theory was first developed in the context of telephone traffic
(Erlang 1909, Erlang 1917a, Erlang 1917b).
Extension Extension
β β
Interarrival time t4 > β
0 τ1 τ2 τ3 τ4 τ5
Arrival times to detector
Figure 2.12: Extension principle of vehicle detectors
Another model similar to the unsignalized intersection is the operation of merging zones
(see Haight, Bisbee & Wojcik 1962, Haight 1963, Drew 1967). Unlike the minor road
drivers at unsignalized intersections, merging drivers can change the gap acceptance
criterion by changing their speeds, and they have a limited opportunity to determine
the location of the merge. In more advanced models the effect of turbulence on priority
streams should be considered (Transportation Research Board 2000).
Passing on two-lane highways is a gap-acceptance process, where a passing vehicle
moves in the opposite direction to the “priority stream”. Drivers can determine the
time and location of passing considering gap availability and sight restrictions. The
order in which vehicles pass each others is not fixed, and several vehicles on a road
segment may pass simultaneously. Consequently, this problem is usually described at a
macroscopic level. Tanner (1953) presented an early description of the problem stating
also the similarities with unsignalized intersection and pedestrian crossing problems.
For an overview of these models see McLean (1989) and Luttinen (2001). Miller &
Pretty (1968) have discussed the gap-acceptance of inconsistent drivers.
2.5 Fluid-analogy model

The number of customers in a system at time τ is the initial number of customers in
system plus the difference between arrivals and departures: N (τ ) = N (0) + A(τ ) −
D(τ ). When A(τ ), or D(τ ), is large, the law of large numbers indicates that with
probability one
A(τ ) − E[A(τ )]
lim = 0. (2.45)
τ →∞ E[A(τ )]
This suggests that the arrival and departure processes can be approximated by their
time-related average values:
E[N (τ )] = E[N (0)] + E[A(τ )] − E[D(τ )] . (2.46)
In a fluid-analogy model a discontinuous stochastic process is replaced by a continuous

deterministic process (see Kleinrock 1976, Newell 1982, Wolff 1989).
A fluid-analogy model is useful especially in capacity estimation and in the estimation
of delays at oversaturated conditions. Once a queue has grown so large that its vanishing
probability is negligible, a fluid approximation will give a reliable estimate of the future
growth of the expected queue (Newell 1971).
According to the central limit theorem the change in queue length approaches normal
distribution. This suggests a diffusion approximation (Glynn 1990, Newell 1982) for
cases where the stochastic process becomes mathematically intractable or too complic-
ated and the random variation cannot be ignored. The delay estimation methods of
unsignalized intersections use both stochastic and fluid-analogy approaches. The dif-
fusion approach has found some applications in the analysis of signalized intersections
(e.g. Newell 1965), but little or none in the analysis of unsignalized intersections.
The basic characteristics of a fluid-analogy model and its use in capacity estimation will
be discussed next. A fluid analogy model for the analysis of oversaturated conditions
will be presented in section 4.4.2.
Let us assume that vehicles arrive at an intersection at arrival rate q(t). The cumulative
number of arrivals is τ
A(τ ) = q(u) du. (2.47)
0
If the arrival rate is constant, the cumulative arrival curve is linear: A(τ ) = qτ (Fig.
2.13). Cumulative demand is the number of arrivals plus the initial queue8 at the
beginning of the observation period (Gazis & Potts 1965):
A (τ ) = L(0) + A(τ ). (2.48)
Number of vehicles
(τ )
L(τ )
s, D
)
(τ
A
W (i)
cle
s,
le
ehi
ic
h
gv
ve
ng
rtin
vi
rri
pa
A
De
Antiblock 0 Block tb Antiblock t0 τ
Figure 2.13: Fluid analogy model for a minor stream
The cumulative number of departures is

τ
D(τ ) = d(u) du, (2.49)
0
where d(τ ) is the departure rate. During a block of length tb no vehicles depart the
intersection; that is
d(τ ) = 0, 0 < τ ≤ tb
(2.50)
D(τ ) = 0, 0 < τ ≤ tb .
8 Fluid analogy models do not consider individual vehicles. So, no distinction can be made between
vehicles in queue an vehicles in service.
When an antiblock starts at τ = tb , the queue starts to discharge at saturation flow rate
s = tf−1 :
d(τ ) = s, tb < τ ≤ t0
τ
(2.51)
D(τ ) = s du = s(τ − tb ), tb < τ ≤ t 0 ,
tb
where t0 is the time when the queue has discharged. After t0 the discharge rate is equal
to the arrival rate:
d(τ ) = q(τ ), t0 < τ ≤ c
(2.52)
D(τ ) = A(τ ), t0 < τ ≤ c,
assuming that t0 < c and q(τ ) ≤ s, and the cycle time c is the sum of the block (tb )
and antiblock (ta ) periods.
The length of a queue (in vehicles) is the initial queue L(0) at τ = 0 plus the difference
between cumulative arrivals and departures. Graphically it is the vertical difference
between cumulative arrival and departure curves. The queue has discharged at time t0 ,
when the number of arrivals is equal to the number of departures (Newell 1965)
L(t0 ) = L(0) + A(t0 ) − D(t0 ) = L(0) + qt0 − s(t0 − tb ) = 0. (2.53)
During a differential time interval (t, t + dt] the delay experienced is L(t) dt. The
cumulative delay during the cycle length c is equal to the area of the triangle in Figure
2.13 and can be obtained by a method analogous to the uniform delay in traffic signal
control (Newell 1965)
c t0 t0
W (tc ) = L(u) du = qu du − s(u − tb ) du
0 0 tb
qstb2 tb2 sta (2.54)
= = , q≤ .
2(s − q) 1 1 c
2 −
q s
The average delay per vehicle can be obtained by dividing the cumulative delay by the
number of arriving (and departing) vehicles during an average cycle (Newell 1965):
tc
L(u) du W (tc ) t2 sta
W = 0tc = = b q , q≤ . (2.55)
0 q(u) du
qtc 2tc 1 − tc
s
The horizontal distance between the arrival and departure curves gives the waiting time
of a “vehicle”.
Capacity is reached when the number of arrivals during an average cycle (sum of the
average block length tb and the average antiblock length ta ) is equal to the number of
vehicles discharging during an average antiblock tb :
ta
Cp c =
tf
ta (2.56)
Cp = ,
tf c
where the proportion of average antiblock (ta ) from the average cycle time (c) is the
probability of a lag larger than or equal to a minimum acceptable lag to . Accordingly,
the potential capacity is
Rϒ (to )
Cp = = sRϒ (to ) , (2.57)
tf
where s = tf−1 is the saturation flow rate. If Figure 2.13 describes average conditions,
capacity is reached when the time (t0 ) required for queue build-up and discharge is
equal to the cycle time. When q ≤ Cp , the deterministic system is in equilibrium. A
more detailed discussion on capacity will be presented next.
CAPACITY AT UNSIGNALIZED INTERSECTIONS
3 CAPACITY AT UNSIGNALIZED INTERSECTIONS

3.1 Definition of capacity
According to HCM2000 (Transportation Research Board 2000) capacity is
the maximum sustainable flow rate at which vehicles or persons reasonably

can be expected to traverse a point or uniform segment of a lane or roadway
during a specified time period under given roadway, geometric, traffic,
environmental, and control conditions; usually expressed as vehicles per
hour, passenger per hour, or persons per hour.
Thus, capacity has three basic elements:
1. Capacity is a location specific measure.
2. Capacity is a maximum flow rate.
3. Capacity describes steady-state conditions.
According to gap acceptance theory the capacity of a minor stream (k) is limited by
−1
its saturation flow rate (sk = tf,k ) and the availability of antiblocks at conflict areas.
A movement may have several conflict areas. It is, however, assumed that a vehicle
waits at a stop line or yield line until an antiblock is available at all conflict areas. If
a median supplies storage spaces for vehicles, so that a movement can be performed
in two stages, capacity of that movement is measured at the median stop or yield line
considering the bottleneck effect of first stage gap acceptance and the limited number of
storage spaces in the median. Accordingly, the capacity at an unsignalized intersection
is defined as the stop-line capacity.
Besides interactions between a minor stream and higher priority movements capacity
can be limited by interactions between movements in the same approach. When a lane
is shared by several movements and/or one movement occupies several lanes, it is most
informative to use as a unit of analysis the group of lanes which are occupied by the
interacting traffic streams in the approach. From this analysis it is also possible to
obtain lane-specific performance measures.
Roundabouts may have additional bottlenecks (see Hagring 1998b): i) In narrow weav-
ing areas heavy vehicles may have operational difficulties or they may prevent the sim-
ultaneous use of an adjacent lane. ii) Pedestrians may limit the capacity of exiting
vehicles.
The maximum flow rate is attained, when there is a continuous queue of subject move-
ment vehicles arriving to the intersection. Because of possible downstream bottlenecks
due to short lanes or two-stage gap acceptance there may be periods of empty queues
at some lanes of a lane group.
Capacity is estimated for one minor stream or a group of movements in one approach.
The flow rates of higher priority streams are given, and the minor stream capacity is
estimated under these higher priority flow rates. A capacity of an entire intersection is
not defined, unless it is defined as the sum of arrival flow rates, which gives D/C ratio
ρi = 1 at the most saturated lane group (see OECD Road Research Group 1974). This
requires that the proportions of arrival flow rates for each movement can be estimated
under the conditions when capacity is reached (Nordqvist et al. 1973).
At roundabouts the circulating flow is generated by merging processes at each entry

point. Thus, the flow rate at subject approach has an effect on merging processes and
capacities at other approaches, and eventually on the major flow of the subject approach
itself (see Akçelik 1998). The procedures below assume that the properties of major
flow do not depend on the flow rate in the subject approach. The capacity of an approach
is estimated assuming that all other traffic flows and driver behaviour attributes remain
the same. This assumption gives capacity estimates suitable for the estimation of
degrees of saturation, and eventually of delays and queues under prevailing conditions
(Troutbeck 1997b).
Capacity is the reasonably expected maximum sustainable flow rate under given con-
ditions. Capacity does not describe maximum observed flow rate during some short
period of time. It describes the average (expected) maximum flow rate that can be
sustained over long time periods under given conditions. Mathematically the capa-
city of a minor stream k can be defined as (Plank & Catchpole 1984, Catchpole &
Plank 1986, Wegmann 1991)
E[Dk (τ )]
Ck = lim , (3.1)
τ →∞ τ
where Dk (τ ) is the number of stream k vehicles crossing a stop line during time interval
(0, τ ] assuming a continuous queue. When the process has been going on for a long
time, the system has reached a steady state and the effect of initial state can be neglected.
During the analysis period (0, τ ] traffic and roadway conditions are assumed to stay
unchanged.
The expected number of departures during a long time interval (0, τ ] is1
E[Dk (τ )] = E[AM (τ )] E[α] , (3.2)
where AM (τ ) is the number of major stream vehicles (gaps) during time interval (0, τ ]
and E[α] is the expected number of stream k departures during a randomly selected
major stream headway. Capacity can now be expressed as (Plank & Catchpole 1984)
E[AM (τ )]
Ck = lim E[α] = qM E[α] , (3.3)
τ →∞ τ
where qM is the major stream flow rate.

Because in real world traffic conditions change frequently, analysis must be based on
short time intervals, which may not be in a steady state. In HCM2000 (Transportation
Research Board 2000) the analysis interval is 15 minutes. This is a compromise between
considerations on short term fluctuations and stability of macroscopic traffic flow meas-
ures. Shorter time interval would allow better adjustment for short traffic fluctuations.
The generalization of the results to stable flow conditions would, however, be problem-
atic. Longer interval length would describe stable conditions better, but nonstationarity
of traffic flow within the time interval could cause bias to the results. When steady state
analyses are applied to finite time intervals, the performance of the facility during these
periods will have considerable stochastic variation around the estimated performance
1 Let the length of a time period be the sum of n headways (τ = t + t + · · · + t ) and let the
1 2 n
expected
number
of minor stream departures during a randomly selected major stream headway
ti be
E Dk (ti ) = E[α]. The expected number of departures during a time interval (0, τ ] is E Dk (τ ) =

E Dk (ti ) = n E[α]. For a long time interval limτ →∞ E[N (τ )] = τ/ E[t] = n.
measures. Also a bias due to initial conditions, such as long or short initial queues2 on
major streams, cannot be excluded.
Following HCM2000 two further capacity concepts are defined:
Potential capacity (Cp ) is the capacity of a specific movement at an unsignalized

intersection approach, assuming that it is unimpeded by other movements and
has exclusive use of a separate lane or lanes.
Movement capacity (Cm ) is the capacity of a specific movement at an unsignalized

intersection approach, assuming that the traffic has exclusive use of a separate
lane or lanes.
Capacity estimation proceeds in four stages:
1. Critical gaps and follow-up times are estimated under prevailing conditions.
2. Potential capacity is estimated for each movement assuming that they have ex-
clusive use of separate lanes and there is no queuing in higher priority streams.
3. Movement capacity is estimated taking into account the impedance effects of

higher priority queues.
4. Capacity of the movements on a lane group is estimated considering also the

effects of shared and short lanes.
Stages 2–4 are described below. Stage 1 is described in chapter 7.

For an overview of unsignalized intersection and roundabout capacity analysis theory
and methods the reader is referred to Brown (1995), Kyte et al. (1996), Transportation
Research Board (1997), Troutbeck & Brilon (1997), and Louah (1986). The discussion
below follows the report of Luttinen (2003).
3.2 Potential capacity under one priority stream

3.2.1 Definition of potential capacity
The theoretical foundation for capacity analysis of unsignalized intersections was laid in
the 1960’s especially by Tanner (1962), Major & Buckley (1962), and Harders (1968).
Consequently, the first edition of HCM (Bureau of Public Roads 1950) did not discuss
unsignalized intersections. Even the second edition (Highway Research Board 1965)
did not present any analysis methodology for unsignalized intersections. Two tables
were presented as examples of capacities of “four-way stop intersections”. By the
time of the third edition (Transportation Research Board 1985) theoretical models and
empirical field data were at such a level that enabled the presentation of specific analysis
procedures. The HCM methodology was adopted from German guidelines (FGS 1972),
which were based on the research by Harders (1968). See Baass (1987), Brilon (1988a),
and Kyte et al. (1991).
Capacity is the expected hourly rate at which vehicles can enter an intersection under
prevailing conditions. Capacity is estimated for streams of ranks lower than 1. The
2 Steady state capacity estimates should be adjusted for finite analysis periods with longer or shorter
than expected initial major stream queues; i.e., expected queues under equilibrium conditions. Longer
(shorter) initial queues indicate impedance effects higher (lower) than under steady state conditions. If all
initial major stream queues are considered as excess flows, minor stream capacity will be underestimated.
methods described below cannot be applied to uncontrolled intersections or Rank 1

flows. The German manual (FGSV 2001) suggests 1,800 pc/h as a maximum flow rate
for Rank 1 through streams. At uncontrolled intersections the manual estimates the
maximum number of vehicles entering the intersection from all four approaches to be
in the range 600–800 pc/h. In the Norwegian guidelines (Statens Vegvesen 1985) the
capacity estimates for uncontrolled intersections are much higher (Table 3.1).
Table 3.1: Capacity (veh/h) of uncontrolled intersections in the Norwegian guidelines (Statens
Vegvesen 1985) as a function of minimum and maximum flow rates in the intersection approaches
Percent qmin /qmax
left turns 1.0 0.8 0.6 0.4 0.2 0.0
0 1,700 1,650 1,600 1,600 1,650 2,000
< 10 1,600 1,550 1,550 1,500 1,500 1,750
< 20 1,500 1,500 1,500 1,450 1,450 1,600
This section describes the basic theoretical formulas used to estimate the potential
capacity of unsignalized intersections. Only a simple crossing operation, as in Figure
2.1, or a simple merge operation, as in a roundabout entry (Fig. 2.4), is discussed. It is
assumed that merging operations between the legs of a roundabout can be ignored. The
road and traffic conditions are assumed optimal. Unless stated otherwise, the models
follow the assumptions presented above (page 46).
Potential capacity describes the capacity of a minor stream under ideal conditions
assuming that it is unimpeded by other movements and has exclusive use of a separate
lane (Transportation Research Board 2000). When the impedance effects (queuing
in higher priority streams) are considered, HCM2000 calls the capacity estimate a
movement capacity. Because Rank 2 vehicles do not face any impedance effects due to
queuing at higher priority streams, the movement capacity of a Rank 2 stream is equal
to the potential capacity:
Cm,k = Cp,k , k ∈ {1, 4, 9, 12} . (3.4)
It is assumed that there is a continuous queue of minor stream vehicles (stream 11 in

Fig. 2.1) waiting to enter the intersection. The expected number of vehicles departing
from a queue during a headway t is given by a gap acceptance function
∞

α(t) = ipi (t) (3.5)
i=1
where pi (t) is the probability of i departures during headway t.

The expected number of departing vehicles during a randomly selected headway is
obtained by integration ∞
E[α] = α(t)f(t) dt, (3.6)
0
where f(t) is the probability density function of major stream headways. Potential
capacity is obtained by multiplying E[α] by the number (qM ) of major stream headways
per time unit (Siegloch 1973)
∞
Cp = qM E[α] = qM α(t)f(t) dt. (3.7)
0
Accordingly, capacity is expressed in terms of traffic flow rate, gap-acceptance function

α(t), and the headway distribution.
3.2.2 Stepwise gap acceptance function

A minor stream vehicle enters the intersection, if there is a headway or lag larger or
equal to the critical gap tc in the priority stream 2. Critical lag is assumed to be equal
to the critical gap. Vehicles entering the intersection during the same priority stream
headway have discharge headways equal to the follow-up time tf .
During a headway of 0 < t < tc no vehicles can enter the intersection. One vehicle can
enter during a headway of tc ≤ t < tc + tf . Two vehicles can enter during a headway
tc + tf ≤ t < tc + 2tf . Accordingly

1, if tc + (i − 1)tf ≤ t < tc + itf

pi (t) = (3.8)
0, otherwise.
Gap acceptance function is a step function

t − t c + tf 0, if t < tc
α(t) = = t−tc +tf (3.9)
tf + tf
, if t ≥ tc ,
where X is the floor function; i.e., greatest integer not larger than X (Spanier &
Oldham 1987).
The probability of i arrivals during a randomly selected headway is the probability that
the headway has an appropriate length:
P{α = i} = P{tc + (i − 1)tf ≤ t < tc + itf }

tc +itf
= f(u) du
tc +(i−1)tf (3.10)
= F(tc + itf ) − F(tc + (i − 1)tf )
= R(tc + (i − 1)tf ) − R(tc + itf ) ,
where f(t) is the probability density function, F(t) is the cumulative distribution func-
tion, and R(t) = 1−F(t) is the survivor function of the major road headway distribution.
Capacity (maximum number of entering vehicles per time unit) is obtained by multiply-
ing the expected number of minor road vehicles entering during a major road headway
by the expected number of major road headways (vehicles) per time unit:
∞

Cp = qM i P{tc + (i − 1)tf ≤ t < tc + itf }
i=0
∞ tc +itf
= qM i f(u) du
i=0 tc +(i−1)tf
∞
(3.11)

= qM i [R(tc + (i − 1)tf ) − R(tc + itf )]
i=0
∞

= qM R(tc + itf ) (veh/h).
i=0
Capacity is high, if the frequency of large headways is high. Consequently, capacity

increases as the coefficient of variation of priority stream headways increases (Luttinen
1990). If priority stream flow rate is zero, the minor road queue discharges at headways
tf , and the capacity is tf−1 veh/s or 3600tf−1 veh/h.
If major stream headways follow the negative exponential distribution, the minor stream
capacity is (Major & Buckley 1962, Drew 1968, Harders 1968)3
∞

Cp = qM R(tc + itf )
i=0
∞
= qM e−λ(tc +itf )
i=0
∞

= qM e−λtc e−iλtf (3.12)
i=0
qM e−λtc
=
1 − e−λtf
qM e−qM tc /3600
= , 0 ≤ tf ≤ tc .
1 − e−qM tf /3600
See Figure 3.1. The result was obtained using the convergence property of a geometric
series; i.e.,
∞
1
ai = , a < 1. (3.13)
i=0
1−a
If follow-up time is zero, the entire queue can discharge simultaneously, and the capacity
is infinite. This is the capacity of a simple pedestrian crossing model.
2000
2s
1800
1600
1400
Capacity, C/(veh/h)
1200
1000 3s
800
600 4s
400 5s
6s
200 7s
8s
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Conflicting flow rate, q /(veh/h)
M
Figure 3.1: Minor-stream potential capacity with different critical gaps, tf = 0.6tc , and major-
stream headways following the negative exponential distribution
Equation (3.12) is used in HCM 1985 (Transportation Research Board 1985, Baass
1987), HCM2000 (Transportation Research Board 2000), DanKap (Vejdirektoratet
3According to Major & Buckley (1962) this equation was first obtained by Fisher (1960). However,
the equation was presented already by Nordqvist (1958). For a different kind of proof see Heidemann
(1991).
1999b) and in the Norwegian guidelines (Statens Vegvesen 1985,Aakre 1998). Capcal 2
uses this equation for minor through and left turn movements, and when the priority
streams occupy at least four lanes (SNRA 1995b, Hagring 1997).
This approach can be used to generalize the potential capacity equation for headway
distributions having an exponential tail; i.e., the distribution of headways larger than
tc is exponential. Because exponential distribution has independent increments, that is
R(a + b) = R(a) R(b), equation (3.11) can be expressed as
∞

i=0
∞

= qM R(tc ) e−iξ tf (3.14)
i=0
qM R(tc )
= ,
1 − e−ξ tf
where ξ is the scale parameter of the exponential tail distribution. The expected num-
ber of departing vehicles during a randomly selected headway, E[α], is the expected
number of departures (α) during a headway larger than the critical gap multiplied by
the probability of a headway larger than the critical gap. Accordingly, equation (3.14)
gives capacity as
Cp = qM E[α] = qM E[ α | T ≥ tc ] P{T ≥ tc } . (3.15)
This expression is valid for any headway distribution. When headways T ≥ tc are
exponentially distributed,
1
E[ α | T ≥ tc ] = (3.16)
1 − e−ξ tf
is obtained as the expectation (2.9) of a geometric distribution. For an approach to
derive capacity as the inverse of average service time for queuing vehicles see section
4.2.2.
For priority stream headways following the shifted exponential distribution the minor
stream potential capacity can be obtained as (Gipps 1982, Luttinen 1990)
∞

i=0
∞

= qM e−θ(tc +itf −tp )
i=0 (3.17)
∞

= qM e−θ(tc −tp ) e−iθtf
i=0
−θ(tc −tp )
qM e
= , 0 ≤ tp ≤ tc , 0 ≤ tf ≤ tc .
1 − e−θtf
If tp = 0, the equation gives the potential capacity for exponential headways (3.12).
When substitution (2.21) for θ is made, capacity can be expressed as

−qM (tc −tp )
qM exp 3600−q M tp
Cp = (veh/h). (3.18)
−qM tf
1 − exp 3600−q M tp
As Figure 3.2 displays, the minor-stream potential capacity decreases as the major
stream minimum headway (tp ) increases and variance decreases (see Luttinen 1990,
Luttinen 1996). The loss in capacity is largest at high major stream flow rates.
1200
1000
800
Capacity, C/(veh/h)
600
0s
1s
400
2s
200 3s
4s
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Conflicting flow rate, qM/(veh/h)
Figure 3.2: Minor stream potential capacity when critical gap is 5 s, follow-up time 3 s, and
major stream headways follow shifted exponential distribution with minimum headway (location
parameter) ranging from zero to four seconds
The survivor function (2.29) of Cowan’s M3 distribution differs from the shifted ex-
ponential survivor function (2.18) by the factor φ. Thus, the potential capacity is
(Plank 1982):
∞

i=0
∞
= qM φ e−γ (tc +itf −tp ) (3.19)
i=0
qM φe−γ (tc −tp )

= , 0 ≤ tp ≤ tc , 0 ≤ tf ≤ tc , 0 ≤ φ ≤ 1.
1 − e−γ tf
It can be interpreted as minor stream capacity across a stream, where the proportion
φ of headways are available for gap acceptance and these headways follow the shifted
exponential distribution. This is a special case of the capacity equation of Tanner
(1967). If φ = 1, we get the capacity for shifted exponential headways (3.17). For
exponential headways (3.12) we set φ = 1 and tp = 0.
In the Australian roundabout capacity guideline (Austroads 1993) parameter γ has been
defined by the method of moments (2.31) as
φqM
γ = , (3.20)
3600 − qM tp
where qM = qc is the circulating flow rate at the entry. The proportion of free headways
is lower than predicted by Tanner’s M/D/1 model (Akçelik & Troutbeck 1991):

q M tp
φ = 0.75(1 − λtp ) = 0.75 1 − . (3.21)
3600
See also Troutbeck (1993) and Troutbeck & Kako (1997).
Troutbeck & Kako (1997) have modified capacity equation (3.19) for “limited priority
merge”. It describes the reduced capacity due to major stream vehicles slowing down
in order to allow enough space for a merging minor-stream vehicle. The delay of a
major-stream vehicle reduces the next gap. On the other hand, shorter critical gaps (tc )
during limited priority operation increase capacity.
Some major-stream vehicles have to slow down, if critical gap tc is less than tf + tp .
These vehicles have to delay their entry in order to keep the headway to a merged
vehicle at tp (see Figure 2.11 and related discussion above). The capacity of limited
priority merge is
qM fL φe−γ (tc −tp )
CL = , tp < tc < tf + tp , (3.22)
1 − e−γ tf
where
1 − e−γ tf
fL = (3.23)
1 − e−γ (tc −tp ) − γ tc − tp − tf e−γ (tc −tp )
is the adjustment factor for limited priority merge (Fig. 3.3). Parameter γ is estimated
as in equation (3.20). If tc > tf + tp , major stream vehicles have absolute priority and
fL should be set to 1. See also Bunker & Troutbeck (1994a), Bunker & Troutbeck
(1994b), Troutbeck (1995), Troutbeck (1997b), and Troutbeck & Kako (1999).
1 4.0 s
3.5 s
0.9
3.0 s
0.8
0.7 2.5 s
L
Adjustment factor F
0.6
2.0 s
0.5
0.4
0.3
0.2
0.1
0
0 200 400 600 800 1000 1200 1400 1600 1800
Figure 3.3: Adjustment factor fL with tf = 2.5 s, φ = 0.8, and tp = 2 s (Troutbeck 1995)
If the traffic flow is considered as departures from an M/D/1 queuing system, substi-
tutions (2.42) and (2.43) are made, and we obtain the capacity formula presented by
Tanner (1962)

qM 1 − λtp e−λ(tc −tp )
Cp = , (3.24)
1 − e−λtf
where λ = qM /3600 veh/s.4 If the minimum headway condition is removed (tp = 0),
equation 3.24 is equivalent to the capacity equation (3.12) for exponential headways.
Tanner’s formula has been used in numerous manuals (e.g. Road Research Laboratory
1965, Hobbs 1974, Austroads 1988). Capcal 2 uses this model for major road left turns
and minor road right turns when the major flow occupies one lane only (SNRA 1995b,
Hagring 1997).
Figure 3.4 displays minor stream capacity according to Tanner (1962) with different
critical gaps and tf = 0.6tc . The follower headway in a platoon is tp = 1.8 s, as in
Capcal 2 (SNRA 1995b, Hagring 1997). The platooning model decreases the frequency
of large headways at high flow rates more steeply than the exponential headway model
(Fig. 3.1). When all vehicles travel in platoons, qM = 3600/1.8 = 2000 veh/h, and
there is no capacity left to a minor stream.
2000
2s
1800
1600
1400
Capacity, C/(veh/h)
1200
3s
1000
800
4s
600
5s
400
6s
7s
200 8s
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
M
Figure 3.4: Capacity according to Tanner with different critical gaps, tf = 0.6tc , and platoon
headways tp = 1.8 s
Figure 3.5 shows the effect of major stream bunching on minor stream capacity. As
bunching increases (φ decreases) the proportion of short intra-bunch headways in-
creases, and so does also the average length of free headways. This improves capacity.
If φ = 1.0, there is no bunching, and the major stream headways follow the shifted
exponential distribution. Shifted exponential distribution has a lower coefficient of
variance than the negative exponential distribution and gives lower capacity for minor
stream vehicles. This is, however, just a feature of the M3 model, and it cannot be taken
as an evidence of capacity reduction at low platoon percentages. By more extensive
4 Tanner obtained the equation as a limiting case when delay approached infinity. Brennan & Fitzgerald
(1979) presented a direct derivation of the capacity equation (3.24).
comparisons Siegloch (1973) concluded that slight disturbances (bunching) reduce ca-
pacity, medium disturbances increase capacity slightly, and severe disturbances increase
capacity significantly (Brilon 1988a).
1200
1000
800
Capacity, C/(veh/h)
600 0.2
0.4 Tanner
0.6
400
0.8
1.0
200
0
0 200 400 600 800 1000 1200 1400 1600
M
Figure 3.5: Capacity with different proportions of free vehicles, critical gap tc = 5 s, follow-up
time tf = 3 s, and intra-platoon headway tp = 1.8 s compared with the Tanner model (dashed
curve)
A comparison of figures 3.1 and 3.4 indicates that at high major road flow rates, when
the variation of headways decreases, the minor stream capacity (3.19) decreases below
the capacity of the exponential model (3.12). This could be expected by the shape of
the coefficient of variation curve (Fig. 2.9). Absolute priority is assumed.
Figure 3.6 displays the capacity of a minor stream under limited priority conditions
based on Tanner’s headway distribution (M/D/1 output) with flow-dependent critical
gaps (tc = 5.0 − 6.0λ) and follow-up times (tf = 0.6tc ) and constant (tp = 1.8 s)
follower headway in platoons. It is compared with Tanner’s capacity formula (absolute
priority) with the same critical gaps, follow-up times, and follower headways. The
parameters are not necessarily realistic and they may produce excess curvature, but they
highlight the difference between the two models. Under flow-dependent gap acceptance
the capacity curve has steepest slopes at low and high flow rates. This can be inferred
from Tanner’s capacity formula (Fig. 3.4) by assuming large (5 s) critical gaps at low
flow rates and small (2 s) critical gaps at high flow rates. If critical gaps decrease with
increasing flow rate the capacity curve becomes more linear, especially under limited
priority. This gives some theoretical support for the British linear regression models
(Kimber & Coombe 1980), as suggested by (Troutbeck 2002).
3.2.3 Linear gap acceptance function

In a fluid analogy model gap acceptance function is continuous for headways larger
than the shortest acceptable headway to (see section 2.5). A linear gap acceptance
1200
1000
800
Capacity, C/(veh/h)
600
Tanner
Limited priority
400
200
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
M
Figure 3.6: An example of minor stream capacity under flow-dependent gap acceptance and
major stream arrivals from an M/D/1 queuing system with limited (solid curve) or absolute
(dashed curve) priority
function has been suggested by Siegloch (1973) as
 tf 
t − tc +
t − to  2 ,
α(t) = =  (3.25)
tf + tf
+
where to = tc − tf /2 is the shortest acceptable headway. The minor-stream departure

process is considered as a fluid flowing into the headways of the priority stream. As
long as the lag is larger than to the minor stream departure flow rate is tf−1 .
As Figure 3.7 indicates, equation (3.25) is a linear approximation of the step function
(3.9). Linear gap acceptance function can be interpreted as a simplified means to
describe the average gap acceptance behaviour of heterogeneous drivers. The potential
capacity is obtained as
∞
C p = qM α(t)f (t) dt
0
qM ∞
= (t − to ) f (t) dt (3.26)
tf to
∞
qM
= tf (t) dt − to R(to ) .
tf to
For negative exponential priority stream headways and linear gap acceptance function
Number of departures
0
to tc tc + tf tc + 2tf Headway (s)
Figure 3.7: Step function and a linear model for gap acceptance
the potential capacity is obtained as (Siegloch 1973, Schnabel & Lohse 1997)

qM ∞
Cp = (t − to ) λe−λt dt
tf to
∞ ∞
λqM −λt −λt
= te dt − to e dt
tf to to
qM −1 ∞
= λ − t − to e−λt t=to (3.27)
tf
qM −λto
= e
λtf
3600 −qM to /3600
= e .
tf
Because tf is the minimum headway and the exponential part is the probability (pro-
portion of time) of a lag larger than to , the equation can be expressed as
Cp = s P{ϒ > to } = sRϒ (to ) , (3.28)
where s is the maximum flow rate of the minor stream in the case of no higher priority
flow; i.e., saturation flow rate. P{ϒ > to } is the proportion of time that minor stream
vehicles find an acceptable lag. During this time minor stream vehicles can discharge
at maximum rate. For a fluid analogy model, capacity can be expressed as the product
of minor stream saturation flow rate (s) and the probability of a lag larger than the
minimum acceptable gap (to ) (Akçelik 1998). The same result (2.57) was obtained in
section 2.5 by a heuristic method.
In the German guidelines (Brilon et al. 1994, FGSV 2001) equation (3.27) has been
used, but the difference between equations (3.27) and (3.12) has been considered in-
significant (see Brilon 1991). The comparison of Figures 3.1 and 3.8 demonstrates that
the capacities of equations (3.12) and (3.27) are virtually identical (see also Troutbeck
& Brilon 1997). According to Brilon & Stuwe (1993) Siegloch’s equation (3.27) with
parameters estimated by regression analysis was slightly better as a capacity estim-
ate for German roundabouts than a linear regression equation. — HCM has applied
equation (3.27) in the 1994 update (Transportation Research Board 1994).
2000
2s
1800
1600
1400
Capacity, C/(veh/h) 1200
1000 3s
800
600 4s
400 5s
6s
200 7s
8s
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 3.8: Minor-stream capacity with different critical gaps, tf = 0.6tc , linear gap acceptance
function and major stream headways following the negative exponential distribution
Equation (3.27) has a simpler form than equation (3.12). Because it can be transformed
into a linear equation by taking a logarithm, it is easy to apply in linear regression
analysis (Guichet 1997, Aagaard 1995). The model also has a simple logic: During
the time available for gap acceptance minor stream discharges at maximum rate. The
operation of individual vehicles can be ignored.
For shifted exponential headways the potential capacity is
∞
qM −θ(t−tp
Cp = tθ e dt − to R(to )
tf to

qM 1 −θ(to −tp ) −θ(to −tp )
= to + e − to e (3.29)
tf θ
qM −θ(to −tp )
= e , to > tp .
θ tf
The application of equation (2.21) gives (Brilon 1988a)
3600 − qM tp −θ(to −tp )

Cp = e , to > tp . (3.30)
tf
The equation can also be expressed in the familiar form
3600e−θ(to −tp )
Cp = = s P{ϒ > to } , to > tp . (3.31)
tf (1 + θ tp )
Because the smallest acceptable headway should be larger than the follower headway in
a platoon (to > tp ), in Cowan’s M3 distribution only the exponential tail of the distribu-
tion has an effect on the capacity. It differs from the shifted exponential distribution by
factor φ, thus giving the following potential capacity equation (Troutbeck 1997a, Trout-
beck & Brilon 1997)
∞
qM −γ (t−tp )
Cp = φγ te dt − to R(to )
tf to
(3.32)
φqM −γ (to −tp )
= e , to > tp .
γ tf
Equation (2.43) indicates that for an M/D/1 departure process φγ −1 = t¯−tp = λ−1 −tp .
When this substitution and the substitution for qM = 3600λ are made, the modification
of Tanner’s capacity model (3.24) is obtained as (Brilon 1988a):
3600(1 − λtp ) −λ(to −tp )

Cp = e , to > tp , (3.33)
tf
where parameter λ = qM /3600 veh/s (Brilon 1988a, Transportation Research Board
1997). This equation can also be presented as
3600φe−γ (to −tp )

Cp = = s P{ϒ > to } , to > tp . (3.34)
tf (φ + γ tp )
The results are virtually identical with the original Tanner model (3.24) as displayed
in Figure 3.4. McDonald & Armitage (1978) have used equation (3.33) as a capacity
model for a roundabout entry. Following the terminology of signalized intersections
they interpreted tf−1 as saturation flow rate and to as a “lost time” associated with each
circulating vehicle and blocking the entry of minor-stream vehicles.
3.3 Potential capacity under multiple priority streams

The potential capacity models described above assume a single conflicting priority
(major) stream on one lane. In most cases a minor stream faces many priority streams,
some of which may share lanes with other streams or occupy several lanes. This
section discusses the effect of multiple independent priority streams. Sections 3.5 and
3.6 discuss the capacity under dependent priority streams. The effect of shared lanes
is discussed in section 3.7.
Let us consider a T-intersection (Fig. 2.3 on page 27) with an exclusive left-turn lane
in the major road. The major road through movements (2 and 5) can be considered as
two independent priority streams for the minor road left turn stream (7). The set of
independent priority streams for minor stream 7 is I7 = {2, 5}. Stream 4 (major road
left turn) is not independent, because it must give way to streams 2 and 3.
On multi-lane highways (Fig. 3.9) traffic flows on adjacent through and right turn lanes
can be considered independent. Minor road right turning vehicles (stream 9) must give
way to traffic flow on one major road lane; i.e., I9 = {2a}. Minor road left-turn vehicles
(7) must give way to other major road through streams, but they can ignore stream 5a;
i.e., I7 = {2a, 2b, 5b}.
If the independent priority streams are described as point processes, the combined
priority stream is a superposition of the component point processes.5 As Figure 3.10
indicates, the superimposed process does not have the minimum headway restrictions
that the component processes may have.
5 In reliability theory a superimposed process is a system having several components in series. When
a component fails, it is immediately replaced with a new component of the same type. (Thompson 1988,
Høyland & Rausand 1994)
5a
5b
2b
2a
7 9
Figure 3.9: T-intersection of a four-lane major road
Time
Figure 3.10: Superposition of three independent point processes (cf. Cox & Smith 1954)
A superimposed process is more like a Poisson process than the component processes
(Cox & Isham 1980). Palm (1943) proposed that the superposition of a large number
of independent sparse point processes leads to a Poisson process in the limit. This first
limit theorem of point processes was proved by Khintchine (1960). See also Gnedenko
(1967), Çinlar (1972), and Cox & Isham (1980). The theorem explains the central
role of Poisson processes in traffic flow theory. If the arrivals of each vehicle are
considered as a separate process, the arrival process of all vehicles at a given location is
a superposition of thousands of component processes, which can be approximated as a
Poisson process. However, physical lengths and safety intervals of vehicles as well as
limited passing opportunities make the component processes dependent on each others,
which dependency the Poisson approximation ignores.
Another important result in the theory of point processes is that a superposition of
two independent renewal processes is a renewal process if and only if the components
are Poisson processes (Çinlar 1972, Cox & Isham 1980). As we shall see below, this
superimposed process is also a Poisson process. Memoryless component processes
produce a memoryless superimposed process. If component processes are renewal
processes, but their future depends on history (such as a minimum headway condition),
all points in the superimposed process are no longer renewal points.
The number of points in a superimposed process during a (long) time interval is the
sum of points in the component processes during the same time interval. Thus, the flow
rate of a superimposed process is the sum of the component flow rates

qM,k = qi , (3.35)
i∈Ik
where Ik is the set of independent priority streams for minor stream k.

Let us assume a major road with multiple lanes (i ∈ Ik ) of independent traffic streams,
which a minor stream k must give way to. When a minor stream vehicle reaches the
first position in the queue at time τ , the lag is the time interval to the arrival of the first
vehicle in any component process i ∈ Ik . The survivor function of a lag distribution in
a superimposed process is (Thompson 1988)

Rϒ,Ik (υ) = P ϒIk (τ ) > υ
!
= P{ϒi (τ ) > υ}
i∈Ik (3.36)
!
= Rϒ,i (υ) ,
i∈Ik
where ϒi (τ ) is the lag in component process i starting from time τ , and ϒIk (τ ) is the
lag in the superimposed process. The calculation of probability P{ϒi (τ ) > υ} may
require information about arrivals preceding time τ on lane (component process) i.
If τ is the time of last arrival in a superimposed process, the headway over all lanes is
larger than t, if the headway on lane j (the lane of last arrival) and lags on other lanes
are all larger than t (Cox & Smith 1954):
!
RIk ,j (t) = P Tj > t P{ϒi (τ ) > t}
i∈Ik
i=j
! (3.37)
= Rj (t) Rϒ,i (t) , j ∈ Ik .
i∈Ik
i=j
The probability that a random point is from a component process j ∈ Ik is the expected
proportion of process j points in the superimposed process (Weiss & Maradudin 1962):
qj qj
pj = = . (3.38)
i∈Ik qi qM,k
The survivor function of headways in a superimposed process can now be presented as

!
RIk (t) = pj P Tj > t P{ϒi (τ ) > t}
j ∈Ik i∈Ik
i =j
!
= pj Rj (t) Rϒ,i (t)
j ∈Ik i∈Ik
(3.39)
i=j
!
−1
= qM,k qj Rj (t) Rϒ,i (t) .
j ∈Ik i∈Ik
i=j
For the probability density function of superimposed headways on a multi-lane highway,

the reader is referred to Weiss & Maradudin (1962).
If stream k has a stepwise gap acceptance function (3.9), the potential capacity is
obtained following equation (3.11):
∞

Cp,k = qM,k RIk tc,k + ltf,k
l=0
∞
(3.40)
!
= qj Rj tc,k + ltf,k Rϒ,i tc,k + ltf,k .
l=0 j ∈Ik i∈Ik
i=j
It is assumed that there is no serial correlation between consecutive headways (see

Luttinen 1996). Then the history
of headways on lane j preceding time τ does not
affect the probability P Tj > t . It is also assumed that the history of other processes
in Ik can be ignored, which is not always a realistic assumption.
In a Poisson process headways follow negative exponential distribution, and the Markov
property simplifies equations considerably. Lag distributions on each lane are identical
with headway distributions and they do not depend on previous history of the process:
P{ϒi (τ ) > υ} = P{ϒi > υ}

= P{Ti > υ} (3.41)
= e−λi υ .
The survivor function of headways in a superimposed process is

!
RIk (υ) = e−λi t = e−t i∈Ik λi
. (3.42)
i∈Ik

Because this is the negative exponential distribution with parameter λ = i∈Ik λi =
qM,k /3600, the capacity of stream k can be calculated using equation (3.12) as follows:
qM,k e−tc qM,k /3600

Cp ,k = . (3.43)
1 − e−tf qM,k /3600
The allocation of major flow among component streams has no effect on capacity,
unless these streams have different critical gaps (Hagring 1998b).
The theory of superimposed point processes gives credibility to Poisson process (ex-
ponential headways) as a gap availability model under multiple independent priority
streams. The results also indicate that under Poisson assumption the superimposed
process is mathematically no more complicated than the component processes. A su-
perposition of Poisson processes is a Poisson process having as a parameter the sum of
component-process parameters.
If headways follow shifted exponential distribution the survivor functions of headways
and lags in the component processes are
Ri (t) = e−θi (t−tp,i ) , tp,i < t (3.44)

−θi (υ−tp,i )
e
Rϒ,i (υ) = , tp,i < υ. (3.45)
1 + θi tp,i
The capacity of a minor stream yielding to several independent major streams having
shifted exponential headway distributions can now be expressed as
∞

!
Cp,k = qj Rj tc,k + ltf,k Rϒ,i tc,k + ltf,k
l=0 j ∈Ik i∈Ik
i=j
∞ ! e−θi (tc,k +ltf,k −tp,i )
−θj (tc,k +ltf,k −tp,j )
= qj e
l=0 j ∈I i∈I
1 + θi tp,i
k k
i=j

∞
e− i∈Ik θi (tc,k +ltf,k −tp,i ) (3.46)
= " qj (1 + θj tp,j )
l=0 i∈Ik (1 + θi tp,i ) j ∈Ik
#∞ $
e− i∈Ik θi (tc,k −tp,i ) −ltf,k i∈I θi
=" e k qj (1 + θj tp,j )
i∈Ik (1 + θi tp,i ) l=0 j ∈Ik
− i∈I θi (tc,k −tp,i )

e i∈Ik qi (1 + θi tp,i )
k
= " , ∀i : tp,i < tc,k .
1 − e−tf,k i∈Ik θi i∈Ik (1 + θ t
i p,i )
Because the flow rate of a shifted exponential headway process can be expressed as
3600 θi
qi = 3600λi = , (3.47)
1 + θi tp,i
the capacity of minor stream k is

3600 e− i∈Ik θi (tc,k −tp,i )
i∈Ik θi
Cp,k = "
−tf,k θi
1−e i∈Ik
i∈Ik (1 + θi tp,i )
"
3600 θ i∈Ik Rϒ,i tc,k (3.48)
= , ∀i : tp,i < tc,k
1 − e−θtf,k

3600 θ Rϒ,Ik tc,k
= ,
1 − e−θtf,k

where θ = i∈Ik θi , and the lag distribution of a superimposed process is

e− i∈Ik θi (υ−tp,i )
Rϒ,Ik (υ) = " . (3.49)
i∈Ik (1 + θi tp,i )
If all minimum headways (tp,i ) are zero, the result is equal to a capacity across random
(Poisson) traffic streams.
Tanner (1967), Troutbeck (1986), Fisk (1989), and Hagring (1998b) have derived the
capacity of a minor stream crossing or merging independent major streams each having
a Cowan’s M3 headway distribution. They assumed that minimum headways in all
streams are identical. With the help of equation (3.40) the potential capacity can be
obtained directly as:

∞
!
Cp,k = qj Rj tc,k + ltf,k Rϒ,i tc,k + ltf,k
l=0 j ∈Ik i∈Ik
i =j
∞
! φi e−γi (tc,k +ltf,k −tp,i )
−γj (tc,k +ltf,k −tp,j )
= q j φj e
l=0 j ∈Ik i∈Ik
φi + γi tp,i
i =j

∞
e− i∈Ik γi (tc,k +ltf,k −tp,i )
= " qj (φj + γj tp,j ) (3.50)
γi tp,i
l=0 i∈Ik 1 + φi j ∈I k
#∞ $
e − i∈I γi (tc,k −tp,i )
k −lt γ
=" e f,k i∈Ik i 3600γj
γi tp,i
i∈Ik 1 + φi l=0 j ∈Ik

e− i∈Ik γi (tc,k −tp,i ) i∈Ik γi ! φi
= 3600 , ∀i : tp,i < tc,k ,
1−e −tf,k i∈I γi
k
i∈I
φi + γi tp,i
k
where flow rate is expressed as

3600 3600 γj
qj = = . (3.51)
E Tj φj + γj tp,j
Equation (3.50) can also be expressed as

"
3600γ i∈Ik Rϒ,i tc,k
Cp,k =
1 − e−γ tf,k (3.52)
3600γ Rϒ,Ik tc,k
= ,
1 − e−γ tf,k

where γ = i∈Ik γi , and
! φi !
Rϒ,Ik (υ) = e− i∈Ik γi (υ−tp,i )
= Rϒ,i (υ) (3.53)
i∈Ik
φi + γi tp,i i∈I k
is the lag distribution of the superimposed process.

The model of Tanner (1967) can be obtained from equation (3.50) by acknowledging
that tp,i = tp , φi = 1 − λi tp , and γi = λi = qi /3600:
e−λ(tc,k −tp ) !
Cp,k = qM,k −λt
1 − λ i tp , (3.54)
1−e f,k
i∈I k

where qM,k = 3600λ = i∈Ik qi . See also Gipps (1982). Fisk (1989) has extended
Tanner’s result to include the case where different lanes have different critical gaps.
Hagring (1998a, 1998b) has extended Tanner’s formula for superimposed M3 headway
distributions having equal follower headways, but allowing lanes to differ in critical
gaps and follow-up times. Fisk and Hagring observed that capacity is greatest when
traffic is evenly distributed among lanes. Then bunching is least and bunch overlap is
greatest for a given total major flow rate.
For a single priority stream the equation 3.50 reduces to equation 3.19. Equation 3.50
can be used to obtain the capacity across multiple independent streams having any com-
bination of negative exponential, shifted exponential, and M3 headway distributions.
The potential capacity across two streams i and j having Cowan’s M3 headway distri-
bution with equal minimum headways (tp ) is
3600φi φj (γi + γj )e−(γi +γj )(tc,k −tp )
Cp,k = , tp < tc,k . (3.55)
(φi + γi tp )(φj + γj tp ) 1 − e−tf,k (γi +γj )
This equation is identical with the equation 8 of Troutbeck (1986). Figure 3.11 displays
a set of curves identical with Fig. 1 of Troutbeck (1986). It shows that major stream
bunching increases minor stream capacity, even in the case of two independent priority
streams.
1800
1600
1400 φ = 0.01
φ = 0.20
Entry capacity (veh/h)
1200 φ = 0.40
φ = 0.60
1000
φ = 0.80
φ = 1.00
800
600
400
200
0
0 200 400 600 800 1000 1200 1400 1600 1800
Flow in each lane (veh/h)
Figure 3.11: Entry capacity at an intersection having two identical priotity streams with min-
imum headway tp = 2 s and different proportions (φ) of free vehicles, and minor stream critical
gap tc = 4 s and follow-up time tf = 2 s (see Troutbeck 1986)
Troutbeck (1991) described a modification of a single-lane capacity model, which

gives good estimates of the two-lane model. He presented an iterative procedure to
estimate the parameters for the modified model. Although Troutbeck & Brilon (1997)
suggested that there are no practical reasons to increase the complexity of calculations
by using multi-lane models, it can be questioned whether the two-lane model (3.55)
is more complicated to use than a one-lane model (3.19) with parameters estimated
by an iterative procedure. Brilon (1988a) has favored simple capacity formulas (3.12,
3.27, and (3.30)) based on the summation of conflicting major streams, because it gives
lower capacity estimates than multi-lane model, and thus does not overestimate actual
capacity.
Hagring (1998b) has demonstrated the superiority of multi-lane models, especially
when critical gaps are lane specific. In this case equation (3.40) should be modified for
lane-specific critical gaps and follow-up times:
∞
!
Cp,k = qj Rj tc,j + ltf,j Rϒ,i tc,i + ltf,i . (3.56)
l=0 j ∈Ik i∈Ik
i=j
The capacity model for M3 distributed component-stream headways becomes

e− i∈Ik γi (tc,i −tp,i ) i∈Ik γi ! φi
Cp,k = 3600 , ∀i : tp,i < tc,i . (3.57)
1−e − i∈I tf,i γi
k
i∈I
φi + γi tp,i
k
When the intra-bunch headways tp,i are equal in all component streams, the capacity
equation by Hagring (1998b) is obtained:

e− i∈Ik γi (tc,i −tp ) i∈Ik γi ! φi
Cp,k = 3600
1−e − i∈I tf,i γi
k
i∈Ik i
φ + γi tp,i
(3.58)
3600e− i∈Ik γi (tc,i −tp ) ! φi qi
= ,
1 − e− i∈Ik tf,i γi i∈Ik
3600γ i

where = i∈Ik γi . Different critical gaps for each lane may reflect the assumption
that the values increase as the distance of the conflicting lane from the stop line increases,
and that crossing and merging operations may have different critical gaps for the same
minor movement (Fisk & Tan 1989). Li, Wang & Jiang (2003) have extended the
multi-lane model to a case where the minor flow consists of two classes of vehicles.
3.4 Potential capacity of roundabout approaches

3.4.1 Background
The capacity of a stream or a lane at an unsignalized intersection is usually defined as the
stop-line capacity. It is restricted by the arrivals and queues of higher priority vehicles
and pedestrians and by minimum headways between departing vehicles. In addition,
limited length of turning lanes may restrict capacity upstream of an intersection.
Hagring (1996a) has pointed out that roundabouts can have additional capacity re-
strictions at weaving sections, especially due to heavy vehicles, and at exits6 , where
capacity may be limited by crossing pedestrians. That is, capacity restrictions apply
to both inflows, outflows, and circulating flows of a roundabout.7 However, most the-
oretical studies of roundabout capacity have focused on entry or stop-line capacity.
Consequently, the discussion below describes the capacity of an approach at a round-
about. The effect of pedestrians on exit flows can be estimated using the procedure in
section 3.8. If the capacity of weaving sections is critical, the use of simulation models
is suggested.
All circulating vehicles in a roundabout have merged to the circulating stream. The
gap-acceptance process at roundabout entries increase bunching in the circulating flow.
On the other hand, exiting vehicles “combine headways”. The processes of entering and
exiting streams modify the shape of the headway distribution of circulating flow (see
Hagring 2001). Although theoretical methods usually assume that the minor stream
does not modify the circulating flow rate, iterative techniques can be used to address
the interdependency of entering and circulating flows.
3.4.2 Historical overview

The first estimates of roundabout capacity were more or less rules of thumb. In 1930’s
Watson (1933) suggested that there should be a four-second headway between success-
ive vehicles in each traffic stream. This would allow 900 pc/h to merge under favorable
6 Preliminary results in Germany indicate that the maximum outflow from a one-lane roundabout with
a rather narrow exit is about 1,200 veh/h (Brilon, Wu & Bondzio 1997).
7 This is of course a simplistic description of roundabout capacity. The operation of traffic streams at
a roundabout is a process of many complex merging and weaving interactions.
conditions. Watson suggested a 15 % margin for pedestrian traffic, obtaining a capacity

estimate of 1,500 pc/h as the maximum sum of major and minor flows at a roundabout
merge. Heavy vehicles required a six-second time interval.
Another early rule of thumb was given by Hammond & Sorenson (1941), who suggested
a capacity limitation of about 3,000 vehicle per hour at any one point. The minimum
radius for the central island was 150 feet (50 m) and the width of the circulatory roadway
25–35 feet (8–11 m). Shrope (1952) provided an extensive empirical analysis of one
roundabout in Latham, New York.
One of the first capacity models was proposed by Clayton (1941, 1945) as:

α 4
Cp = s n − n− . (3.59)
90 3
The parameters of the model were saturation flow rate (s = 1200 veh/h), number of
entry lanes (n), and “maximum weaving angle” (α, angle between the diagonals of a
weaving area). This model was later recommended by the British Ministry of Transport
(Clayton 1955).
The most significant of the early roundabout capacity models was developed by Wardrop
(1957) at the British Road Research Laboratory. Based on measurements under test
conditions the maximum flow through a weaving section was found to be
4.9(ww + wa )(4lw − 3ww )(3 − pw )
Cp = pc/h, (3.60)
l(0.56 + pHV )
where ww and lw are the width and the length of a weaving area (Fig. 3.12), wa is
average approach width (Fig. 3.13), pw is the proportion of weaving traffic to total
traffic in weaving section, and pHV is the proportion of “medium or heavy” vehicles,
including busses. The equation was further developed to the following form (Road
Research Laboratory 1965):

wa p
108ww 1 + 1−
ww 3
Cp = ww pc/h. (3.61)
1+
lw
This “practical capacity” was about 80 per cent of the capacity formula (3.60) above.
At old British roundabouts, described by these models, neither vehicles already at a
roundabout nor those entering were given priority (Ashworth & Laurence 1978). The
operations were best described as weaving rather than merging or gap acceptance man-
euvers (McDonald & Armitage 1978). It was possible that vehicles “blocked back”
from an adjoining weaving section and so impeded the exiting flow (Road Research
Laboratory 1965). This “locking effect” limited the capacity of a roundabout, as illus-
trated in the delay/flow curve of Figure 3.14 (Webster 1960).
When circulating flow was given priority, the locking problem was avoided and capacity
was increased approximately by 10 % (Blackmore 1963), being comparable to the best
traffic signal systems (Blackmore 1970). Consequently, an offside priority rule was
introduced in Britain in November 1966 (Duff 1968, Kimber 1980). It required entering
drivers to give way to circulating vehicles. Long weaving sections to absorb queues
were no longer necessary, enabling the design of smaller roundabouts (Kimber 1980).
This offside-priority rule is nowadays the prevailing policy worldwide.
In the 70’s the Wardrop’s model was found to be no longer applicable in Britain
(Ashworth & Field 1973, Philbrick 1977), and some gap-acceptance models based
2r c
lw
ww
we
re
Figure 3.12: Roundabout geometry parameters
wa
wa
lf
we − w a
2
Figure 3.13: Flared entry parameters
Figure 3.14: The effect of “locking” at a roundabout (Webster 1960, Road Research Laboratory
1965)
on Tanner’s model (3.24) were developed (Armitage & McDonald 1974, McDonald
& Armitage 1978, Ashworth & Laurence 1978) as well as linear regression models
(Philbrick 1977). The current British methodology is based on linear regression mod-
els (O’Flaherty 1997).
3.4.3 Current methodologies

The British linear model for roundabout capacity is (Kimber 1980)

a1 (a2 − a3 qM ), if a3 qM ≤ a2
Cp =
0, if a3 qM > a2 ,
where

a1 = 1 − 0.00347(φe − 30) − 0.978 re−1 − 0.05
a2 = 303a4
a3 = 0.210a5
we − w a
a4 = wa +
1 + 2a6
0.5
a5 = 1 +
1 + e c −60)/10
(2r
we − wa
a6 = .
lf
Parameters are described in Figures 3.12 and 3.13. Parameter φe is the entry angle;
i.e., conflict angle between entering and circulating streams. Parameters we , wa , lf , rc ,
and re are in meters, φe is in degrees, and Cp and qM are in pc/h. Kimber (1980) has
presented more detailed descriptions of the parameters as well as their ranges in the
data base.
In Norway the potential capacity of a roundabout is estimated as (Statens Vegvesen
1985, Aakre 1998)
Cp = 275a1 − 0.282qM (1 + 0.2a1 ) pc/h. (3.62)
The coefficient a1 is defined as

we − w a
a1 = w a +
1 + 2a2
1.6(we − wa )
a2 = ,
lf
where we is the entry width (Fig. 3.12) and lf is the entry flare length (Fig. 3.13).
Other examples of linear capacity models have been presented by Louah (1988), Simon
(1991), and Hakkert et al. (1991). Regression models based on exponential capacity
equations (3.27) have been presented by Stuwe (1991), Brilon & Stuwe (1993), Guichet
(1997), and Aagaard (1995), among others.
A gap-acceptance model for roundabouts requires an appropriate headway distribution
for circulating streams and the estimation of critical gaps and follow-up times, adjusted
for the roadway (geometric) conditions of the roundabout. If a roundabout is analyzed
as a series of T-intersections, the equations for potential capacity in section 3.2 can be
applied. HCM2000 (Transportation Research Board 2000, Troutbeck 1997b) suggests

the potential capacity model (3.12) based on exponential headways and a stepwise gap-
acceptance function. The same equation is used in DanKap (Vejdirektoratet 1999b).
German HBS 2001 applies the Siegloch model (3.27). All these manuals apply the same
model for roundabouts as for other unsignalized intersections. The Swedish Capcal 2
has a special headway distribution for circulating flow (SNRA 1995c). The method is
described in section 5.4.
The Australian method (Austroads 1993) applies the M3 model (3.19), which is also
used in the SIDRA software package (Akçelik 1998). The parameters γ and φ are
estimated using equations (3.20) and (3.21). The intra-bunch headway tp is two
seconds for one-lane roundabouts and one second for multi-lane roundabouts (Akçelik
& Troutbeck 1991, Austroads 1993). Multi-lane traffic flows are reduced to one-lane
flows modified to have a shorter intra-bunch gap and flow rate equal to the total flow
rate on all circulatory lanes (Troutbeck 1991). In later SIDRA versions the intra-bunch
headways have been defined as follows (Akçelik 1998):
tp = 2.0 s for single-lane circulating flow

= 1.2 s for two-lane circulating flow
= 1.0 s for circulating flow with more than two lanes
If lane flows are not equal, the intra-bunch headway is calculated as a flow-weighted
average of intra-bunch headways of the streams contributing to the circulating flow.
The effect of vehicles exiting to the subject approach is ignored (Troutbeck 1991). The
effect of limited priority has also been evaluated (see page 60).
Hagring (1998b) has developed a new Swedish method to estimate roundabout capacity.
He uses the M3 model (3.19), which he has extended for multi-lane highways with lane-
dependent critical gaps and follow-up times. The capacity equation (3.58) and its de-
rivation has been presented above. The intra-bunch headway is 1.8 seconds, which has
been used in the old Swedish capacity manual (Statens vägverk 1977, Anveden 1988)
and verified by Hagring (1996b). The proportion of free headways is obtained by a
relation φi = 0.910 − 1.545qi /3600 (Hagring 1998b, Hagring 2000). The Tanner dis-
tribution (2.41) with φi = 1 − tp qi /3600 was also found to give good results in capacity
estimation. In a recent paper (Hagring et al. 2003) estimated the proportion of free
headways as φi = 0.914 − 1.549qi /3600. This model indicated that the allocation of
traffic on lanes and thus the origin-destination flows in a roundabout have a considerable
effect on capacity (Hagring 2000), as Akçelik (1997) had suggested earlier.
3.5 Movement capacity of Rank 3 streams

3.5.1 Description of the procedure
The estimation of Rank 3 stream k movement capacity proceeds in four stages:
1. The potential capacity (Cp,i ) of a conflicting Rank 2 stream i is determined using

the procedures described above.
2. An adjustment factor (fa,i , fraction of time available for gap acceptance in stream
−1
i) is evaluated using Cp,i as the average service time of stream i server. This
procedure is repeated for all conflicting Rank 2 streams.
3. The gap acceptance capacity Ca,k of stream k is estimated assuming that no stream
i vehicles are waiting for entry and all major streams are independent.
4. The movement capacity of stream k is the gap acceptance capacity multiplied by

the adjustment factor
Cm,k = fa,i Ca,k , (3.63)
or
Cm,k = fI,i Cp,k , (3.64)
where fI,i is the impedance factor describing the fraction of potential capacity
available as movement capacity. In case of several conflicting Rank 2 streams
the adjustment or impedance factors are combined.
The estimation of factors fa,i and fI,i , and gap-acceptance capacity Ca,k is discussed
below, but first it is necessary to take a closer look at queue discharge processes.
3.5.2 Queuing model for Rank 3 stream capacity

Figure 3.15 displays a simplified presentation of traffic streams at a T-intersection.
The minor-road left-turn vehicles (stream 7, Rank 3) must give way to streams 2 and
4. Streams 2 and 4 are, however, not independent. The Rank 2 stream 4 must yield
to Rank 1 stream 2. Thus stream 7 vehicles can enter the intersection only if there
is a gap larger than the critical gap in the combined major stream (2 and 4) and no
stream 4 vehicles are waiting for entry. Potential capacity, which assumes independent
priority streams (Section 3.3), would overestimate the capacity. A modified approach
is required.
4 (R2)
2 (R1)
7 (R3)
Figure 3.15: Three ranks of streams at a T-intersection
The time available for Rank 3 vehicles consists of two types of periods: i) Rank 2 stream
queuing and queue discharge periods, and ii) gap acceptance periods. During gap
acceptance periods all major-stream (Rank 1 and Rank 2) vehicles can proceed without
delay, and priority streams can be considered independent. When the proportion of time,
number of major-stream headways, and the headway distribution of the superimposed
process available for Rank 3 gap acceptance are known, the capacity of a Rank 3 stream
can be calculated. In order to do this, the queuing model has to be described in detail.
Figure 3.16 displays a queuing model description of Rank 2 stream i vehicles blocking
the entry of Rank 3 stream k vehicles. As vehicle 1 enters the queue of stream i, Rank
1 vehicles block the entry of both Rank 2 and Rank 3 vehicles. When the antiblock
starts, vehicle 1 enters service, and the busy period of the server begins. During the
service time tf,i no other stream i vehicle can enter service. Vehicle 2 has to wait for
the service of vehicle 1 to complete. As vehicle 1 departs, vehicle 2 enters service.
Busy period ends as vehicle 2 departs service. The length of the busy period is 2tf,i .
An isolated vehicle 3 arrives to an empty system. As it enters the queue, the service
starts immediately. It blocks stream k for the time of the critical gap tc,k .
Time to
depart service k
Time to
enter service k
tf,k
Minor stream k blocked

Time to
depart service i
tc,k tf,i tc,k
1 2 3 Time to
enter service i
tf,i tf,i
Busy period
Block
1 2 3 Time to
enter queue i
1 2 3
Figure 3.16: A queuing model having service times interrupted by blocks
The block of minor stream k terminates as the server of stream i becomes idle. After
that the server of stream k starts service. The blocks and antiblocks in Figure 3.16 are
described in terms of priority stream service departure times. Accordingly, the major
stream arrival times in Figure 2.11 on page 43 should also be interpreted as service
departure times. As Rank 1 flows decrease, the Rank 2 stream approaches an M/D/1
queuing system, and the departure flow approaches the model of Tanner (see page 41).
If tf,i is interpreted as the travel time from the stop line to the conflict point, the service
starts when a vehicle crosses the stop line and ends when the vehicle has passed the
conflict point. In this case the utilization factor of the server is the sum of utilization
factors for all component streams (see section 3.10).
The queuing analogy should not be stretched too far. When an acceptable gap follows
a stream i vehicle, a stream k vehicle does not—safety permitting—have to wait for
tf,i before it can enter the intersection. The possibility of overlapping service times
between conflicting streams cannot be excluded.
In the basic queuing model of Figure 3.16 each stream has a separate server with service
time equal to the follow-up time. Service is interrupted by blocks created by higher
priority vehicles. Haight (1963) described this model by stating that in an unsignalized
intersection a vehicle arriving to an intersection with no queue does not necessarily
receive immediate service.
If interruptions in the operation of stream i server are included in the service time, the
average service time of queued vehicles is the inverse of capacity (S̄q,i = Ci−1 ). As
Figure 3.17 displays, in this model vehicle 1 enters service immediately, and service
(1)
time includes the time (We,i ) spent waiting for the block termination. Service begins
when a vehicle enters the stop line and terminates tf,i after the vehicle has crossed the
stop line. Neither a Rank 2 nor a Rank 3 vehicle can cross the stop line before the
follow-up time tf,i has elapsed.
Vehicle 1 departs service tf,i after the termination of the block, and the next vehicle
enters service. Major stream arrivals frequently extend the service times of minor
stream vehicles beyond the follow-up time. The dashed arrow displays the virtual
departure time; i.e., the departure time assuming that vehicle 1 had arrived during an
antiblock.
Minor stream k blocked

Time to
depart service i
tf,i tf,i tc,k
Block
1 1 2 3 Time to
(1) enter service i
We,i
tf,i tf,i
Busy period
1 2 3 Time to
enter queue i
1 2 3
Figure 3.17: A queuing model having service times extended by blocks
In both models the service departure times of stream i vehicles are equal. In both
models the busy period ends as server i becomes idle. During busy periods the entry
of stream i vehicles may be blocked by Rank 1 vehicles. The starting time of a busy
period is, however, different in the models. The model in Figure 3.17 includes in the
busy period the time that the first vehicle is waiting for an antiblock, while this time is
excluded from the busy period in model 3.16. Because a Rank 2 vehicle waiting for
entry prevents the entry of a Rank 3 vehicle, this time period is not available for gap
acceptance. Accordingly, the capacity of a Rank 3 stream should be estimated using
the model in Figure 3.17. In this model service time is the sum of two components:
waiting time at yield/stop line and follow-up time (Yeo & Weesakul 1964). See Section
4.2 for more details.
According to Kimber, Summersgill & Burrow (1986) a queuing vehicle enters service
when the previous vehicle has crossed the yield/stop line. Follow-up time is at the
beginning of a service time, before a queuing vehicle reaches the yield/stop line. For
non-queuing vehicles service begins when they arrive to the yield/stop line, and service
time does not include the follow-up time. In this model follow-up time is interpreted
as the time required for a queuing vehicle to move to the yield/stop line. The length of
the service time before a minor stream vehicle reaches the yield/stop line depends on
the departure time of the vehicle ahead. See also Kremser (1964).
In this report it is assumed that service of all vehicles begins when they reach the
yield/stop line. After crossing the yield/stop line a vehicle is being served until the end
of the follow-up time, as displayed in Figure 3.17 (see Weiss & Maradudin 1962, Dunne
& Buckley 1972, Daganzo 1977, Pöschl 1983). Follow-up time is interpreted as a safety
zone behind a vehicle crossing a yield/stop, which postpones the opportunity of the next
vehicle to enter the yield/stop line, and also the departure of a lower priority vehicle.
The definition of service time is the same for all vehicles, also for vehicles arriving
during the follow-up time of the previous vehicle. The measurement of a follow-up
time is initialized when a vehicle crosses the yield/stop line.
3.5.3 Rank 2 queue discharge

Let us assume that the arrival processes of all streams are Poisson (see Section 2.3.3).
Minor stream i vehicles accept lags and headways that are not shorter than the critical
gap (tc,i ). The minimum interdeparture time of minor-stream vehicles is equal to the
follow-up time (tf,i ).
The utilization factor of a server in a Rank 2 stream i is
qi S̄i qi
ρi = ≈ = ρi , (3.65)
3600 Cp,i
where ρi is the D/C ratio, and the potential capacity Cp,i is calculated using the equation
(3.12) above. Utilization factor ρi is the fraction of time that server i is busy. The
average service time S̄i is approximately equal to the inverse capacity. (See section 4.2
for details.) For the sake of simplicity, the utilization factor (ρ) and the D/C ratio (ρ )
are here assumed equal, unless otherwise stated.
The fraction of randomly arriving (Poisson) vehicles in stream i having to wait for
service is equal to the fraction of vehicles arriving when the server is busy, which is
equal to the utilization factor ρi (Khintchine 1960). These vehicles discharge from a
queue and their headways cannot be used by minor stream drivers. Other (1 − ρi )
vehicles enter the intersection without queuing, although they may have had to wait for
an antiblock at the stop line.
The number of queue discharge headways per hour is ρi qi . Because the average
−1
queue discharge headway is Cp,i , the fraction of time required for queue discharge
−1
is ρi qi Cp,i = ρi . During queue discharge the departure flow rate of stream i is
2
Cp,i = qi ρi−1 . The flow rate can be expressed as the sum of flows during the queue
discharge and free departure periods:
qi
qi = ρi2 + (1 − ρi2 )qi∗ , (3.66)
ρi
where qi∗ is the average flow rate during the free departure periods. The vehicles
departing during the free departure periods do not have to wait for service, although
the service times may be extended by higher priority vehicles. The average flow rate
during the free departure periods can be solved from equation (3.66) as
qi
qi∗ = . (3.67)
1 + ρi
The average free departure headway of stream i can now be expressed as

3600 1 1
¯ti∗ = ∗ = 3600 + . (3.68)
qi qi Cp,i
It is the sum of average arrival headway and average service time. If Rank 1 flows
−1
approach zero, Cp,i approaches tf,i and the average free departure headway in stream
−1
i approaches λi + tf,i .
82 Example 3.1 (No Rank 1 flows) Capacity and Level of Service at Finnish
Following Figure 3.18 an analysis period τ of stream i can be divided into two distinct
time periods: queue discharge and free departures. During a queue-discharge period of
length ρi2 τ the departure flow rate is Cp,i . The free departure period lasts for (1 − ρi2 )τ ,
and has flow rate qi (1 + ρi )−1 .
Flow rate
Cp,i
qi
1 + ρi
ρi2 τ τ Time
Figure 3.18: Flow rates during queue-discharge and free-flow periods
As presented in equation (3.7), capacity is the product of expected departures during a

randomly selected headway multiplied by the expected number of headways per time
unit. The number of queue discharge headways in stream i is ρi qi . The number of
headways available for gap acceptance is (1−ρi )qi (Gipps 1982). Because no headway
can
be
shorter than
the follow-up time, the headway distribution should be modified as

fi t T ≥ tf,i (Dunne & Buckley 1972, Siegloch 1973).
3.5.4 Rank 3 capacity with no Rank 1 conflict

Dunne & Buckley (1972) have studied the capacity of a Rank 3 stream with no Rank
1 conflict, but Rank 2 stream modified by filtering through a Rank 1 stream. A similar
model is discussed here, but in this model a Rank 3 vehicle cannot proceed, if a Rank
2 vehicle is waiting for entry.
If Rank 3 vehicles can ignore other priority streams besides stream i (as in Fig. 3.19),
the movement capacity of a Rank 3 stream k is:
∞

Cm,k = (1 − ρi )qi α(t)fi t T ≥ tf,i dt
0
(1 − ρi )qi ∞
= α(t)fi (t) dt (3.69)
Ri tf,i 0
(1 − ρi )
= Cp,k ,
Ri tf,i
where Cp,k is the potential capacity obtained by the methods above. The capacity is
adjusted for Rank 2 queues by (1 − ρi ) and for the modified headway distribution by
−1
Ri tf,i . For a stepwise gap acceptance function (3.9) the capacity is obtained as
∞

Cm,k = (1 − ρi )qi Ri tc,k + j tf,k T ≥ tf,i
j =0
∞
(3.70)
(1 − ρi )qi
= Ri tc,k + j tf,k .
Ri tf,i j =0
5 (R1)
1 (R2)
7 (R3)
Figure 3.19: Three ranks of streams with no conflict between ranks 1 and 3
For random (exponential) headways in stream i and no Rank 1 flow the movement capacity of
a Rank 3 stream k is
(1 − ρi )qi e−λi tc,k
Cm,k = ×
e−λi tf,i 1 − e−λi tf,k
(3.71)
(1 − ρi )qi e i (tc,k −tf,i )
−λ
= .
1 − e−λi tf,k
The equation has the same form as Tanner’s capacity formula (3.24). This was to be expected,
because in the absence of Rank 1 flows ρi = λi tf,i , and the Rank 2 flow i is a departure process
from an M/D/1 queuing system (see also Dunne & Buckley 1972). Constrained headways have
a fixed length, and unconstrained headways follow the shifted exponential distribution.
3.5.5 Time available for gap acceptance

Harders (1968) proposed that, when no Rank 2 vehicles are queuing, the movement
capacity of Rank 3 streams can be estimated assuming independent priority streams.8
When there is a queue of Rank 2 stream i vehicles, no Rank 3 stream k vehicles can
enter the intersection. Thus the movement capacity of stream k can be expressed as
Cm,k = p0,i Cp,k , (3.72)
where p0,i is the probability of empty queue in stream i, and Cp,k is the potential
capacity of stream k assuming independent priority streams.
Harders (1968) observed that the flow rate in a Rank 2 stream i can be expressed as the
sum of delayed and undelayed vehicles. During a proportion of time 1 − p0,i vehicles
discharge from queue at flow rate Cp,i . When there is no queuing, a vehicle can enter
the intersection without delay, if there is an antiblock in the Rank 1 stream j and the
headway in the Rank 2 stream i is longer than the follow-up time tf,i . Assuming Poisson
arrivals Harders expressed flow rate qi as
qi = (1 − p0,i )Cp,i + p0,i χ qi , (3.73)
where χ is the probability of an antiblock in the Rank 1 stream j and a headway larger
than the follow-up time in Rank 2 stream i:

χ = P ϒj ≥ tc,i P Ti ≥ tf,i
= e−tc,i qj /3600 e−tf,i qi /3600 (3.74)
(tc,i qj +tf,i qi )/3600
=e .
8 This assumption is correct for Poisson arrivals only. As Rank 2 queue clears, the first vehicle in a
Rank 3 stream queue faces not a gap but a lag.
The probability of empty queue in stream i was then obtained as
Cp,i − qi
p0,i = . (3.75)
Cp,i − χ qi
Harders called p0,i an impedance factor. Figure 10-5 in HCM 1985 (Transportation
Research Board 1985) is based on this equation (see Brilon 1988a).
In HCM2000 (Transportation Research Board 2000) the impedance factor p0,i is the
probability that at a random time instant the system in stream i is empty; i.e.,
qi
p0,i = 1 − ρi ≈ 1 − = 1 − ρi , ρi ≤ 1, (3.76)
Cp,i
where the utilization factor (ρi ) of the queuing system in stream i is approximated as
the D/C ratio (ρi ), qi is the arrival flow rate, and Cp,i is the potential capacity of stream
i. Brilon (1988a) and Grossmann (1991) have demonstrated that equation (3.76) gives
acceptable results. Equation (3.72) is used to calculate the movement capacity.
This method has three problems:
1. The impedance factor p0 is interpreted as the fraction of time available for gap
acceptance, whereas it should be interpreted as the fraction of headways available
for gap acceptance, as shown in Section 3.1.
2. The utilization factor ρ overestimates the impedance effect of a Rank 2 stream.
3. The distribution of Rank 2 stream headways during the gap acceptance periods
should be modified to exclude headways shorter than the follow-up time tf,i and to
adjust the flow rate during free-departure conditions accordingly (Siegloch 1973).
These modifications are discussed below.

When a priority vehicle in queuing models of Figures 3.16 and 3.17 has departed service,
the next vehicle in the same stream or in a minor stream can enter service. The critical
gaps for Rank 3 vehicles are evaluated backward from these time instants.
The critical gap of a minor stream includes the follow-up times of priority stream
vehicles. Thus, the blocking effect of vehicle 1 in Figure 3.17 does not start as a
vehicle enters service, but from the “virtual departure time” (dashed arrow) that it
would have departed service if the entry was not blocked by Rank 1 vehicles. It is
also obvious, that the blocking effect of an isolated vehicle 3 is tc,k , not tc,k + tf,i . The
follow-up time of an isolated vehicle should be included as time available for Rank 3
gap acceptance. The time period blocked by a freely departing vehicle is equal to the
critical gap, and it is considered in the gap acceptance process.9 This is evident, if we
consider a situation having no Rank 1 flow and no queuing in a Rank 2 stream. The
impedance factor should be equal to unity. Consequently, the time period blocked by
queuing Rank 2 vehicles is the busy period minus the follow-up time.
If a busy period in stream i ends at time τ , the following idle time is the lag ϒi (τ )
remaining at time τ . For random (Poisson) arrivals the average idle period is equal to
the average lag, which is equal to the average headway λ−1
i . If the average
length of a
busy period is E Tbp,i and the average length of an idle period is E Tip,i = λ−1
i , then
9Akçelik (1994a) includes the follow-up times of all major stream vehicles in antiblocks, but subtracts
follow-up time from critical gap.
for a queuing process with Poisson arrivals holds (Gross & Harris 1974)

ρi E Tbp,i
= = λi E Tbp,i . (3.77)
1 − ρi E Tip,i
The expected length of a busy period is now obtained as

ρi
E Tbp,i = . (3.78)
λi (1 − ρi )
The average busy cycle is the expected sum of busy and idle periods:
ρi 1 1
E Tbc,i = E Tbp,i + E Tip,i = + = . (3.79)
λi (1 − ρi ) λi λi (1 − ρi )
The proportion of randomly arriving (Poisson) vehicles having to wait for service is
the proportion of vehicles arriving during a busy period (see Khintchine 1960):

λi E Tbp,i
pw,i = = ρi . (3.80)
λi E Tbc,i
This proportion does not include the vehicles, whose service initiated the busy periods.
The proportion of time available in a Rank 1 stream for gap acceptance is obtained as

E Tbp,i − tf,i
fa,i = 1 −
E Tbc,i
tf,i
= 1 − ρi +
E Tbc,i (3.81)
= (1 + λi tf,i )(1 − ρi )

qi tf,i qi
= 1+ 1− .
3600 Ci
This is the proportion of time during which queuing stream i vehicles do not block the
entry of Rank 3 vehicles; i.e., the proportion of time during which the Rank 2 departure
headways (real and virtual) follow the shifted exponential distribution. As capacity
−1
approaches tf,i , the impedance factor approaches the probability 1 − ρi2 of no queue (at
most one customer in system) in an M/G/1 queuing system.10 If stream i is congested
(ρi > 1), the server does not have idle periods and the adjustment factor is zero. In
order to avoid negative values of fa,i , the equation should be slightly modified:

qi tf,i qi
fa,i = 1+ 1− . (3.82)
3600 Ci +
As Figure 3.20 displays, the factor fa,i assigns a larger proportion of time to gap
acceptance than p0,i = 1 − ρi . The difference between p0,i and f0,i increases as
potential capacity Cp,i increases, because then busy cycles become shorter.
10 Kendall (1951) has shown that for Poisson arrivals the probability of a queue length left behind by
a departing customer is geometric: pn = (1 − ρ)ρ n . Thus the probability that a departing customer
leaves behind at most one customer is (1 − ρ) + (1 − ρ)ρ = 1 − ρ 2 . For Poisson arrivals this is also
the queue size distribution for a outside observer (see Cooper 1981); i.e., the probability that at a random
time instant no vehicles are waiting for service. A process having Poisson arrivals also has the PASTA
(Poisson arrivals see time averages) property (Wolff 1989).
0.9
C=1800 veh/h
0.8
1200
0.7 600
Impedance factor
0.6
0.5
0.4
p = 1−ρ
0
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Degree of saturation, ρ
Figure 3.20: Two impedance factors: p0 (dashed line) and fa (solid curves)
In Figure 3.20 the follow-up time is tf,i = 2 seconds. Capacity 1,800 veh/h is attained,
when there is no conflicting higher priority traffic. Even then the impedance factor
decreases to zero as the D/C ratio increases to unity. This indicates that the model
accounts for a bottleneck effect. Because the discharge headway of vehicles entering the
intersection cannot be shorter than tf,i , a turning movement may have queue formation
even though there is no conflicting higher priority traffic. Furthermore, at low degrees
of saturation the drop in the time available for gap acceptance is not as steep as p0,i
indicates.
Because fa,i is the time available for gap acceptance, the number of Rank 1 stream j
headways available for gap acceptance11 is fa,i qj . As noted above, the number of Rank
2 stream i headways available is (1 − ρi )qi . The adjustments are different, because
Rank 2 headways during a gap acceptance period are all larger than tf,i , including the
first one.
3.5.6 Gap availability for Rank 3 vehicles

The discussion above has demonstrated that the fraction of Rank 2 stream i headways
available for Rank 3 gap acceptance is 1 − ρi . These may be headways of either real
or virtual stream i departures. The fraction of time fa,i available for gap acceptance is
given by equation (3.82), which also gives the fraction of Rank 1 stream j headways
available for gap acceptance. The total number of available headways per hour is

∗ qi tf,i
qM = (1 − ρi )qi + fa,i qj = (1 − ρi ) qi + 1 + qj . (3.83)
3600
11 In capacity estimation the lags and headways available for gap acceptance can be assumed to form a
continuous time period during which the nonqueuing major-stream vehicles arrive. See Figure 3.18. The
resulting headways are available for gap acceptance.
In order to derive the equation for movement capacity, it is necessary to obtain the
distribution of headways during the gap acceptance periods. As indicated earlier these
are headways of either real of virtual priority stream departures.
Traffic flow is described as a point process, where the location of a point on a time
axis describes the departure time of a vehicle from a service process. A Rank 3 vehicle
observes two point processes: process ‘I’ for Rank 1 departures and process ‘II’ for
Rank 2 departures. Let us assume that all arrivals can be described as a Poisson process.
Process I is independent of other streams, so that its headways follow the negative
exponential distribution with location parameter λI . The headways available for gap
acceptance in process II follow shifted exponential distribution with scale parameter λII ,
which is equal to the scale parameter of the exponential arrival headway distribution.
The location parameter (minimum headway) is equal to the follow-up headway tp = tf,II .
Because departure times of delayed vehicles have been replaced by virtual departure
times, the virtual flow rate of stream II during gap acceptance periods is
3600λII
qa,II = 3600λa,II = , (3.84)
1 + λII tf,2
which is the flow rate of the shifted exponential headway process.

Because all service times in the Rank 2 stream exceeding tf,2 have been excluded, these
two processes can be considered independent. The process observed by the Rank 3
vehicles is the superposition of processes 1 and 2. The survivor function of intervals
(headways) in a superimposed process of two component processes is
R(t) = pI RI (t) Rϒ,II (t) + pII RII (t) Rϒ,I (t) , (3.85)
The proportions of points from both component processes are
λI λI (1 + λII tp )
pI = = (3.86)
λI + λa,II λI (1 + λII tp ) + λII
λa,II λII
pII = = . (3.87)
λI + λa,II λI (1 + λII tp ) + λII
where λa,II = λII (1+λII tp )−1 . The same results are obtained, if pI and pII are calculated
with λa,I = fa,II λI = (1+λII tp )(1−ρII )λI and λa,II = (1−ρII )λII . The survivor function
can now be expressed as

pI
R(t) = + pII e−(λI +λII )t+λII tp
1 + λII tp
(3.88)
λI + λII
= e−(λI +λII )t+λII tp , t ≥ tp .
λI (1 + λII tp ) + λII
If tp = 0, we get the survivor function of negative exponential distribution with scale

parameter λ = λI + λII , as expected.
3.5.7 Movement capacity under one Rank 2 stream

When minor stream gap acceptance follows the step function (3.9), the movement
capacity of a Rank 3 stream k can be obtained as
∞

Cm,k = qM R tc,k + itf,k
i=0
∞
3600(1 − ρII )(λI (1 + λII tp ) + λII )(λI + λII ) −(λI +λII )(tc,k +itf,k )+λII tp
= e
λI (1 + λII tp ) + λII i=0
∞

−(λI +λII )tc,k +λII tp
= 3600(1 − ρII )(λI + λII )e e−(λI +λII )itf,k
i=0
−(λI +λI )tc,k +λII tp
e
= 3600(1 − ρII )(λI + λII )
1 − e−(λI +λII )tf,k
e[−(qI +qII )tc,k +qII tp ]/3600
= (1 − ρII )(qI + qII ) , tp < tc,k
1 − e−(qI +qII )tf,k /3600
= fII Cp,k ,
(3.89)
where
(1 − ρII )
fII = (3.90)
e−qII tp /3600
is the impedance factor due to Rank 2 stream II queuing, tp = tf,II is the follow-up
time of the Rank 2 stream, and Cp,k is the potential capacity (3.12) assuming a Poisson
priority stream with flow rate qI +qII , critical gap tc,k , and follow-up time tf,k . The Rank
1 priority streams used in the analysis of ρII may be different than the Rank 1 priority
streams of stream k. The denominator in the impedance factor is the probability of a
headway or lag larger than tp in the Poisson process of the Rank 2 stream. It adjusts
the capacity for the filtered headway distribution (and lower flow rate) of the Rank 2
stream during the periods available for Rank 3 gap acceptance. The result is similar to
equation (3.70). The relation between fII and the HCM2000 impedance factor p0,II is
p0,II
fII = −q
. (3.91)
e tf,II /3600
II
If qII = 0, equation (3.89) is equal to the capacity equation (3.12) for Poisson priority
flow. If qI = 0, equation (3.89) is equal to Tanner’s formula (3.24), as indicated in
example 3.1 on page 83.
Capacity equation (3.89) can be obtained directly from equation (3.46) by setting num-
ber of headways qI∗ = (1+λII tp )(1−ρII )qI , qII∗ = (1−ρII )qII , scale parameters θI = λI ,
θII = λII , and minimum headways tf,I = 0 and tf,II = tp :
e−(λI +λII )tc,k +λII tp (qI + qII )(1 + λII tp )(1 − ρII )
Cm,k =
1 − e−(λI +λII )tf,k (1 + λII tp )
(3.92)
e[−(qI +qII )tc,k +qII tp ]/3600
= (1 − ρII )(qI + qII ) , tp < tc,k
1 − e−(qI +qII )tf,k /3600
Figure 3.21 displays the capacity of a Rank 3 stream with different Rank 2 and total
(Rank 1 + Rank 2) major-stream flow rates. The critical gap of both Rank 2 and Rank
1200
1000 q =0→
II
Rank 3 stream capacity (pc/h)

800
qII=300→
600
400
qII=600→
200
q =900→
II
0
0 500 1000 1500
Total priority flow rate (pc/h)
Figure 3.21: Movement capacity of a Rank 3 stream with different Rank 2 (qII ) and total priority
flow rates. Critical gaps and follow-up times of both Rank 2 and Rank 3 streams are 5 s and
3 s, respectively. Capacities were calculated with equation (3.89) and Monte Carlo simulation
(markers). Dashed curve shows Tanner’s capacity (3.24) with no Rank 1 flow and Rank 2
stream minimum headway equal to the follow-up time.
3 vehicles is five seconds and follow-up time three seconds. The capacities of equation
(3.89) are compared with results of Monte Carlo simulations.
If there are no priority flows, the capacity is 3600/tf,k = 1200 veh/h. For zero Rank
2 flows Rank 3 capacity is obtained with the equation (3.12) for Poisson priority flow,
displayed as the upper curve in Figure 3.21. For zero Rank 1 flows the capacity is
obtained with Tanner’s capacity formula (3.24), displayed as a dashed curve in Figure
3.21. When qI = 0 veh/h and qII = 1200 veh/h, no headways are available for Rank 3
gap acceptance, and Rank 3 capacity is zero. The Rank 3 capacities for other proportions
of Rank 1 and Rank 2 flows are found approximately between the capacity formulas
(3.12) and (3.24).
Figure 3.22 displays a comparison of equation (3.89) with HCM2000 capacities. (The
HCM adjustment factor 2 for left-turn flows has not been applied.) Dashed curves
display the respective capacities with no Rank 1 flows. This is the major difference
between equation (3.89) and HCM2000. The latter excludes the follow-up times of
all Rank 2 vehicles from the time available for Rank 3 gap acceptance and assumes
exponential priority stream headways. Equation (3.89) includes the follow-up times
of non-queuing Rank 2 vehicles in the time available for Rank 3 gap acceptance and
assumes shifted exponential priority stream headways. It gives a lower Rank 2 flow
rate during the gap acceptance periods than the average arrival rate of the (exponential)
Rank 2 stream. Consequently, HCM2000 gives much lower capacities at low Rank 1
flows.
Stream 7 at a T-intersection (Fig. 2.3) must give way to Rank 1 streams 2 and 5, and
1200
1000
HCM→
800 q =300→
II
600 qII=300→
qII=600→
400
qII=600→
200 q =900→
II
q =900→
II
0
0 500 1000 1500
Figure 3.22: Movement capacity of a Rank 3 stream: Equation 3.89 (blue curves) compared
with HCM2000 capacities (magenta). Dashed curves display capacities with no Rank 1 flows.
Rank 2 stream 4. Movement capacity is calculated as
Cm,7 = f4 Cp,7 , (3.93)
where
1− q4
Cp,4
f4 = . (3.94)
e−q4 tf,4 /3600
The priority streams of stream 4 used in the calculation of Cp,4 are I4 = {2, 3}. Potential
capacity Cp,7 is calculated assuming priority streams I7 = {2, 4, 5}. Rank 1 priority
flow is q2 + q5 plus the possible effect of the major road right-turn stream 3. Rank 2
stream is q4 .
3.5.8 Movement capacity under two Rank 2 streams

In a four-leg intersection (Fig. 2.2, page 26) minor road through streams 8 and 11 have
Rank 3. They must give way to Rank 2 streams 1 and 4 (major road left turns). Stream
1 must give way to Rank 1 streams 5 and 6, and stream 4 must give way to streams
5 and 6. The priority streams of minor streams 8 and 11 are I8 = {1, 2, 4, 5, 6} and
I11 = {1, 2, 3, 4, 5}.
The impedance factor f4 of major-road left-turn stream 4 is given above. The impedance
factor of stream 1 is likewise
1 − ρ1 1− q1
Cp,1
f1 = = . (3.95)
e−q1 tf,1 /3600 e−q1 tf,1 /3600
It is the proportion of capacity available due to major stream queuing. Because the
queuing processes of streams 1 and 4 are independent, it can be assumed that the
combined impedance factor is the product
f1∧4 = f1 f4 . (3.96)
This is indeed the case, as will be shown next.

If the other Rank 2 stream is ignored, the proportion of time available for gap acceptance
in streams 1 and 4 is

q1 tf,1 q1
fa,1 = 1 + 1− , q 1 < C1
3600 C1
(3.97)
q4 tf,4 q4
fa,4 = 1 + 1− , q4 < C4 .
3600 C4
These factors are probabilities that at a random time instant gap acceptance is not
restricted by queuing vehicles in the respective stream. Because the queuing processes
are independent, the probability that at a random time instant gap acceptance is not
restricted by queuing vehicles in either stream is the product
fa,1∧4 = fa,1 fa,4 . (3.98)
Because this is also the proportion of vehicles in Rank 1 streams arriving during gap
acceptance periods, the number of vehicles observed in these streams during the gap
acceptance periods is
qa,2 = fa,1∧4 q2
qa,3 = fa,1∧4 q3
(3.99)
qa,5 = fa,1∧4 q5
qa,6 = fa,1∧4 q6 .
Of the “free headways” in stream 1 (4) only headways of vehicles arriving during the
non-queuing periods of stream 4 (1) can be considered for gap acceptance. Thus,
qa,1 = fa,4 (1 − ρ1 )q1

(3.100)
qa,4 = fa,1 (1 − ρ4 )q4 .
Streams 2, 3, 5, and 6 are assumed to be Poisson processes with parameters λ2 =

q2 /3600, λ3 = q3 /3600, λ5 = q5 /3600, and λ6 = q6 /3600. During the gap acceptance
periods streams 1 and 4 are assumed to have shifted exponential headway distributions
with scale parameters λ1 = q1 /3600 and λ4 = q4 /3600 and location parameters equal
to the follow-up times tf,1 and tf,4 .
During the time periods available for gap acceptance Rank 2 streams are not blocked
by Rank 1 vehicles, so that all streams 1, 2, 4, and 5 can be considered independent.
The capacity of streams 8 and 11 can be obtained using equation (3.46) as
qM e(−qM tc,8 +q1 tf,1 +q4 tf,4 )/3600

Cm,8 = (1 − ρ1 )(1 − ρ4 )
1 − eqM tf,8 /3600
(1 − ρ1 )(1 − ρ4 ) qM e−qM tc,8 /3600
= × (3.101)
e−q1 tf,1 −q4 tf,4 1 − eqM tf,8 /3600
qM e−qM tc,8 /3600
= f1∧4 ,
1 − eqM tf,8 /3600
where qM = q1 + q2 + q4 + q5 + q6 . Likewise, the movement capacity of minor stream

11 is
qM e−qM tc,11 /3600
Cm,11 = f1∧4 , (3.102)
1 − eqM tf,11 /3600
where qM = q1 + q2 + q3 + q4 + q5 . If ρ1 ≥ 1 or ρ4 ≥ 1, Rank 3 capacities are zero.
In the calculation of qM the major-road through flows are adjusted for the possible
effects of nonconflicting right-turn streams (stream 3 on stream 8 and stream 6 on
stream 11). However, in the estimation of Rank 2 capacities this adjustment is not
made, but the effect of major road right-turn streams is considered directly.
3.6 Movement capacity of Rank 4 streams

Rank 4 streams 7 and 10 (left turns from minor road) in four-legged intersections must
give way to streams of three higher ranks. A Rank 4 vehicle can enter the intersection
when there is an acceptable gap in the priority streams and no conflicting Rank 2 or
Rank 3 queues. These queues are, however, not independent. The queues of a Rank 3
stream can discharge when there is a gap larger than the critical gap in Rank 1 and Rank
2 priority streams, and conflicting Rank 2 queues are empty. No analytical equations
for Rank 4 stream capacity estimates have been presented (Troutbeck & Brilon 1997).
For Rank 4 stream 7 (10) the conflicting Rank 3 stream is the opposing through stream
11 (8). Conflicting Rank 2 streams are the major-road left-turn streams 1, 4 and the
opposing minor road right turning stream 12 (9). The latter stream has no effect on
the Rank 3 queues in stream 11 (8). If the major road has separate lanes for entering
streams 7 (10) and 12 (9), the streams do not have a conflict.
Because of the mutual dependency of queues, conflicting Rank 2 and Rank 3 queues
occur more simultaneously than would be the case, if the queuing processes were
independent. Ignorance of this dependency would give too low capacity estimates.
Brilon & Großmann (1991) have developed by simulation a corrected probability:
p0,1∧4∧i √
pz,i = 0.65p0,1∧4∧i − + 0.6 p0,1∧4∧i , (3.103)
p0,1∧4∧i + 3
where pz,i is the probability of empty queues in major road left-turn streams 1, 4, and
minor road through stream i, and p0,1∧4∧i = p0,1 p0,4 p0,i (see Grossmann 1991, Kyte
et al. 1996). This equation is also used in HCM2000 (Transportation Research Board
2000). As Figure 3.23 displays, the corrected probability (thick curve) is larger than
the simple product p0,1∧4∧i (diagonal line). The movement capacity of streams 7 and
10 is calculated as
Cm,7 = pz,11 p0,12 Cp,7
(3.104)
Cm,10 = pz,8 p0,9 Cp,10 .
The German HBS2001 (FGSV 2001) uses a different approach:

1
pz,i = . (3.105)
1 − p0,1 p0,4 1 − p0,i
1+ +
p0,1 p0,4 p0,i
The queues of all conflicting priority streams are considered as a single queue. M/M/1
queueing systems are assumed (Brilon, Wu & Bondzio 1997).
Equation (3.105) can also be expressed as
p0,1∧4∧i
pz,i = , (3.106)
p0,1∧4 + p0,i − p0,1∧4∧i
0.9
z, i
Idle server probability for dependent streams, p
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Idle server probability for independent streams, p
0, 1 ∧ 4 ∧ i
Figure 3.23: Corrected probability of empty queues in streams 1, 4, and i (8 or 11) (Brilon &
Großmann 1991)
which can be interpreted as the probability of an idle server in all streams 1, 4, and i
under the condition that a server is known to be idle either in both left-turn streams 1 and
4 or in stream i. All service processes are assumed to be independent. The HCM2000
equations (3.104) are used to obtain the movement capacities of streams 7 and 10.
Figure 3.24 displays the German impedance factor against different values of p0,1∧4
and p0,i . When there is no traffic in major-road left-turn streams (p0,1∧4 = 1), the
probability of a queue-free state in all priority streams is p0,i , as expected. As Figure
3.25 displays, the HCM2000 impedance factor is not consistent under these conditions.
The same observation can be made, when there is no traffic in stream i; i.e., p0,i =
1. Consequently, equation (3.103) does not fulfill the necessary marginal conditions
(Brilon, Wu & Bondzio 1997).
Brilon & Wu (2002) described these approaches as pragmatic simplifications, which are
only of an approximative nature. They chose to propose an even simpler approximation
(see section 3.10), which is easier to use in practical applications. However, below is
described a theoretically more solid approach.
The potential capacity Cp,k of a Rank 4 stream k is given by equation (3.12), where
the major flow rate qM is the sum of conflicting higher priority streams. Following
equation (3.89) this capacity should be adjusted for the utilization ratio of the server
and the modified headway distributions of higher priority streams.
First it is necessary to define a utilization factor of a service system with two streams (i
and j ) of units (vehicles). The average arrival rates are λi and λj , and average service
times S̄i and S̄j . The utilization factor is the product of average arrival rate and the
average service time. The average arrival rate of units is λi + λj and the average service
1 1.0
0.9 0.9
Idle server probability for streams 1, 4, and i; pz, i
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Idle server probability for major road left turns, p0, 1 ∧ 4
Figure 3.24: German impedance factor pz,i with different values of p0,1∧4 and p0,i (right)
(FGSV 2001)
1 1.0
0.9
0.9
0.8
0.8
0.7
0.7 0.6
0.6 0.5
0.4
0.5
0.3
0.4
0.2
0.3
0.1
0.2
0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Figure 3.25: HCM2000 impedance factor pz,i with different values of p0,1∧4 and p0,i (right)
time is the weighted average

λi S̄i + λj S̄j
S= . (3.107)
λi + λj
The utilization factor of the service system is
ρi,j = (λi + λj )S̄ = λi S̄i + λj S̄j = ρi + ρj , (3.108)
which is the sum of the utilization factors of the component streams.

The impedance factor pz,i is now derived following the queuing model in Figure 3.16,
in which Rank 3 service is interrupted while a Rank 2 vehicle is being served. Because
only one server can be busy at one time, the utility factors are additive.
Let us now assume that stream i has Rank 2 and stream j has Rank 3. The arrival
rates are qi = 3600λi and qj = 3600λj . Rank 2 stream i has follow-up time tf,i and
average service time S̄i = 3600/Cm,i . The utilization factor is ρi = λi S̄i . Rank 3
stream j has follow-up time tf,j and movement capacity Cm,j . The utilization factor ρj
is different from the demand-to-capacity (D/C) ratio ρj = qj /Cm,j . Because stream
j vehicles are served only during the fraction 1 − ρi of time, the average service time
is S̄j = 3600(1 − ρi )/Cm,j and the utilization factor ρj = λj S̄j = (1 − ρi )ρj . This
gives an interesting interpretation for the D/C ratio of a Rank 3 stream:
ρj
ρj = . (3.109)
1 − ρi
The D/C ratio of a Rank 3 stream j gives the probability that at a random time instant
a stream j vehicle is being served (waiting for a gap or a follow-up time) under the
condition that there are no Rank 2 vehicles in the service system.
The utilization factor of the total service system is
ρi,j = ρi + ρj = ρi + (1 − ρi )ρj = ρi + ρj − ρi ρj , (3.110)
which is the probability that there is a vehicle in the service system. The probability of
an empty system is now obtained directly as

1 − ρi,j = 1 − ρi + (1 − ρi )ρj = (1 − ρi )(1 − ρj ). (3.111)
The queuing processes are not independent, but the effect of stream i on the queuing
process of stream j is carried in the D/C ratio ρj , as demonstrated above (3.109). For
a four-leg intersection this indicates that
pz,i = p0,1∧4 p0,i = p0,1 p0,4 p0,i . (3.112)
This is displayed in figure 3.26 and as the diagonal line in figure 3.23. It should, however,
be observed that the model discussed here includes also a capacity adjustment for the
headway distribution of Rank 2 and Rank 3 streams, which adjustment is not included
in HCM2000. In addition, equation (3.69) gives higher capacities and lower utilization
factors than HCM2000 for Rank 3 streams, which increases the Rank 4 capacity.
The discussion above gives an important result: Capacities of Rank 2, 3, and 4 streams
are calculated in a hierarchical order. The capacities of higher priority streams are used
as input in the capacity estimation of lower priority flows. These capacity estimates
contain all the information necessary about the hierarchy of streams. When the capacit-
ies of major streams are known, their relative priorities can be ignored in the capacity
estimation of a minor flow. If qi is the flow rate of a stream, which is in a set Ik of
1 1.0
0.9 0.9
0.8 0.8
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Figure 3.26: Linear impedance factor pz,i with different values of p0,1∧4 and p0,i (right)
priority streams (Ranks 2 and 3) of minor stream k, the movement capacity of stream
k can be expressed as
"
i∈I p0,i
Cm,k = − qk t /3600 Cp,k , (3.113)
e i∈Ik i f,i
where Cp,k is the potential capacity of stream k assuming that all major streams are
independent, p0,i = 1 − ρi , and ρi is the D/C ratio of stream i. This equation can be
used to estimate the movement capacity of both Rank 3 and Rank 4 streams.
The set Ik includes only priority streams of Rank 2 and Rank 3. If a Rank 1 stream j
with minimum headway tp,j is included in Ik , the headways in stream j are assumed
to follow Tanner’s distribution. This is equivalent to assuming a conflicting “Rank 0”
stream with no traffic. Accordingly, ρj = qj tp,j /3600.
Figures 3.27 and 3.28 display the movement capacities of equation (3.113) compared
with Monte Carlo simulations. Figures 3.29 and 3.30 show that equation (3.113) gives
higher capacities than HCM2000. No adjustments for conflicting flows, as in HCM2000
Exhibit 17-4, have been made.
The movement capacities of streams 7 and 10 are
p0,1 p0,4 p0,11 p0,12 Cp,7
Cm,7 = (3.114)
e−(q1 tf,1 +q4 tf,4 +q11 tf,11 +q12 tf,12 )/3600
p0,1 p0,4 p0,8 p0,9 Cp,10
Cm,10 = , (3.115)
e 1 tf,1 +q4 tf,4 +q8 tf,8 +q9 tf,9 )/3600
−(q
where Cp,k is the potential capacity of stream k (7 or 10) assuming that all major
streams are independent and have negative exponential headway distributions with
scale parameter λi = qi /3600. The possible effect of nonconflicting major-road right-
turn streams (3 and 6) should be included in the major-road through-stream (2 and 5)
flow rates.
1200
1000
Rank 4 stream capacity (pc/h) q =0→

II
800
600
qII=300→
400
qII=600→
200
0
0 500 1000 1500
Figure 3.27: Movement capacity (3.113) of a Rank 4 stream when Rank 3 flow rate is qIII =
300 veh/h compared with Monte Carlo simulations (markers)
1200
1000
800
600
qII=0→
400
q =300→
II
200
0 qII=600→
0 500 1000 1500
600 veh/h compared with Monte Carlo simulations (markers)
1200
1000
q =0→
II
800
q =0→
II
600
q =300→
II
400
qII=300→
qII=600→
200
qII=600→
0
0 500 1000 1500
300 veh/h (blue curves) compared with HCM2000 (magenta curves)
1200
1000
800
600
qII=0→
400 qII=0→
qII=300→
200
qII=300→
0 q =600→
II
0 500 1000 1500
600 veh/h (blue curves) compared with HCM2000 (magenta curves)
3.7 Shared lanes

On shared lanes multiple streams are combined. If an approach in a four-leg intersection
has only one lane, all three streams (left, through, and right) share the same lane.
Hawkes (1966) described a minor-road lane shared by left-turn and right-turn vehicles
as an embedded Markov chain (see page 118 below). Fisk & Tan (1989) have modeled
a lane shared by two movements as an M/M/1 queuing system with two customer types.
The third approach, which is followed here, was presented by Harders (1968). Its is
simple and widely applied.
If the movement capacity of stream i is Cm,i , the average departure headway (service
time) from a continuous queue is
1
S̄q,i = . (3.116)
Cm,i
The average service time on a lane sharing n streams is

n n
pi
S̄q = pi S̄q,i = , (3.117)
i=1 i=1
Cm,i
where pi is the proportion of stream i on the lane

qi qi
pi = n = , (3.118)
i=1 qi q
and q is the total flow rate on the shared lane. The shared lane capacity is (Harders
1968, Catchpole & Plank 1986)
1 1 q
C= = p = n , (3.119)
i=1 ρi
S̄q n i
i=1
Cm,i
where ρi = qi /Cm,i . If a shared lane includes Rank 1 streams (major road through
and/or right movements) the movement capacity Cm,i of a Rank 1 stream i is equal to
the saturation flow rate si .
The discussion above considers the capacity of a lane shared by several movements.
Mathematically these movements are vehicles with different gap acceptance properties.
Accordingly, it is possible to extend equation (3.119) to estimate the capacity of lanes
shared by any types of heterogeneous vehicles (passenger cars, trucks, RV’s) and drivers
(men, women, older drivers) (Troutbeck & Brilon 1997). The capacity of a stream
having a heterogeneous mixture of vehicle types is the harmonic mean of capacities for
each vehicle type, weighted by the frequency of the vehicle type in the subject stream
(Catchpole & Plank 1986). That is, the capacity is the inverse of average service times.
In fact, this approach is valid for all traffic systems where traffic flows of different
capacities are using the same service facility (Kyte et al. 1996).
If Rank 1 (major road through and right turn) vehicles share a lane with a left turn
movement, the Rank 1 vehicles may be delayed because of left turning vehicles waiting
for an acceptable gap. These delayed Rank 1 vehicles increase the time required for
queue discharge on the shared lane. The impedance effect of a shared lane can be
estimated following a method proposed by Harders (1968).
Let us assume that a lane is shared by left-turn and through movements. The D/C
ratio of the left-turn movement (ρL ) is obtained following equation (3.65). It is the
fraction of time required for the queue discharge of left-turn vehicles. On a shared lane
this fraction of time is increased by the through vehicles joining the queue. Let (ρa )
designate the fraction of time required for the discharge of through vehicles. Undelayed
through vehicles do not form a queue.
The fraction of time required by queue discharge on a shared lane is
ρ S = ρ L + ρa . (3.120)
Because through vehicles are assumed to arrive following a Poisson process, the fraction
of delayed through vehicles is ρS . The fraction of time required by the discharge of
delayed through vehicles is
qT
ρa = ρS = ρ S ρT , (3.121)
sT
where qT , sT , and ρT are the arrival rate, saturation flow rate and D/C ratio of the
through stream, respectively. The factor ρS is now obtained as (Harders 1968)
ρL
ρS = ρL + ρS ρT = . (3.122)
1 − ρT
If a lane is shared also by a right turning stream having D/C ratio ρR , the total utilization
factor of Rank 1 streams is the sum ρT + ρR , which gives
ρL
ρS = . (3.123)
1 − (ρT + ρR )
Since the 1994 edition (Transportation Research Board 1994) HCM has suggested the
use of ρS in stead of ρL (L = 1, 4) to take into account the impedance effect of queue
formation on a shared lane.
It should be noted that the fraction of time ρS includes all headways less than the
follow-up time in the left-turn stream, but not in the through or right-turn streams.
Consequently, the headway distribution of the left-turn stream during the time intervals
available for Rank 3 and Rank 4 gap acceptance is shifted exponential. The headway
distributions of the Rank 1 streams are, however, unmodified (i.e., negative exponential).
3.8 Pedestrian effects

Pedestrians can cross a road in groups. A follow-up time needs to be considered only
at very high pedestrian volumes, when pedestrian enter the crosswalk in several rows.
The follow-up time is then applied to rows of pedestrians.
The problems regarding pedestrians in the hierarchy of traffic streams have been dis-
cussed above (page 28). No rank was assigned to pedestrian flows, because they break
the hierarchical structure of vehicular-stream priorities. Because pedestrians do not
have an absolute priority, but drivers and pedestrians “negotiate” the priority, the gap
acceptance theory is of a limited use.
Below a simple method is presented to estimate the impedance effects of pedestrian (and
cyclist) flows. The method is similar to the methods in the HCM2000 (Transportation
Research Board 2000) and in the Additive Conflict Flows methodology described in
section 3.10.
When a pedestrian or a group of pedestrians crosses a lane of width wl meters at speed
vp m/s, the lane is blocked for time interval
wl
tpb = . (3.124)
vp
Lane width can be adjusted for the length of pedestrian groups.

If pedestrians cross a road in groups of average size ng , the average number of pedestrian
groups per hour is
qp
qg = , (3.125)
ng
where qp is the hourly flow rate of pedestrians. Furthermore, if vehicles in minor stream
k give priority to a fraction pk,j of pedestrian groups in pedestrian stream j , the fraction
of time a lane is blocked by pedestrians can be obtained as
qg tpb
ρj = pk,j . (3.126)
3600
It is assumed that pedestrians and vehicles “negotiate” the priority. If a group of

pedestrians is given priority, their crossing time is unavailable for gap acceptance. If a
group of pedestrians is not given priority, it is assumed that they walk across the road
together with the next group of pedestrians that has obtained priority.
At high pedestrian flow rates the pedestrian impedance factor p0,j = 1 − ρj should be
adjusted for follow-up times between rows of pedestrians crossing a road. Otherwise
the pedestrian impedance factor is applied in the same way as the impedance factors of
vehicular streams.
3.9 Passenger-car equivalencies

Equation (3.119) can be used to estimate the capacity of a stream consisting of different
classes of vehicles with different gap-acceptance properties. Each vehicle class is
considered as a separate stream. Brilon (1988b) used this approach to confirm the
German passenger-car equivalencies presented by Jessen (1968).
The effect of passenger-car equivalencies are not evaluated in this research. It is sug-
gested that the same adjustment factors are used as for the signalized intersections
(Luttinen & Nevala 2002).
3.10 Excursion: Additive Conflict Flows methodology

Brilon & Wu (2001, 2002) have suggested the Additive Conflict Flows (ACF) procedure
as a third capacity analysis method, besides the empirical regression technique and the
gap acceptance procedure. They proposed that the theoretical background of ACF is
easier to understand than the theory of gap acceptance. ACF can also better analyze
pedestrian and cyclist flows as well as the limited priority effects. The draft chapter for
unsignalized intersection capacity analysis is based on theACF (Brilon & Miltner 2002).
The estimation of ACF parameters has been considered easier than the estimation of
gap acceptance parameters. The major advantage is that the capacity of a stream can
be calculated as a function of traffic volumes of other streams in a single step. In gap
acceptance theory the capacities of higher priority flows are required as parameters for
Rank 3 and 4 stream capacity calculations.
The capacity of stream k is the maximum discharge flow rate (qmax,k ) multiplied by the
proportion (p0 ) of time available for queue discharge:
3600
Ck = qmax,k p0 = p0 , (3.127)
tq,k
where tq,k is the length of time that the conflict area is occupied by a discharging stream
k vehicle. It is similar to the follow-up time tf,k in the gap acceptance models. The
probability p0 that the conflict area is not occupied by another stream is the product
p0 = p0,q p0,a , (3.128)
where p0,q is the probability of no queuing or discharging major stream vehicles and
p0,a is the probability of no approaching major stream vehicles.
The probability that a conflict area is not blocked by queuing or discharging stream i
vehicles is p0,q,i = 1 − ρi = 1 − λi tq,i . As in signalized intersections, the D/C ratio
(ρA ) of major streams in a conflict area (A) is calculated as the sum of the degrees of
saturation of conflicting major streams, so that
qi tq,i
p0,q,A = 1 − ρA = 1 − ρi = 1 − λi tq,i = 1 − . (3.129)
i∈A i∈A i∈A
3600
If a minor stream crosses several conflict areas, the probability that no area is blocked
by queuing or discharging vehicles is the product
!
∗
p0,q = p0,q,j , (3.130)
j
where the multiplication is over all conflict areas. Conflict areas are assumed to operate
independently, acknowledging that the assumption is not always realistic.
The probability p0,a,i that a conflict area is not blocked by an approaching stream i
vehicle is expressed as the probability that a gap is larger than a time period ta,i ; i.e.,

p0,a,i = P Ti > ta,i = Ri ta,i , (3.131)
where the time period ta,i is the time period that the conflict area is blocked by an
approaching vehicle in major movement i. It is comparable to the shortest acceptable
headway to,i = tc,i −tf,i /2, but contrary to the gap acceptance models, the time period ta,i
is related to major stream, not minor stream. For exponential headways the probability
is
p0,a,i = e−ta,i qi /3600 , (3.132)
The probability that conflict area A is not blocked by approaching vehicles is
!
p0,a,A = p0,a,i = e− i∈A ta,i qi /3600 . (3.133)
i∈A
Because all conflict periods have been excluded, the headways in all priority streams
can be assumed independent. If a minor stream crosses several conflict areas, the
probability that no area is blocked by approaching vehicles is the product
!
∗
p0,a = p0,a,j = e− i ta,i qi /3600 , (3.134)
j
where j runs over all conflict groups and i runs over all major streams. None of the
major streams should be counted twice.
The capacity of a minor stream k involved in more than one conflict group is
3600 ∗ ∗
Ck = p p
tq,k 0,q 0,a
 
(3.135)
3600 !  qi tq,i
 e− i∈Ik qi ta,i /3600 ,
= 1−
tq,k j i∈j
3600
where j runs over all conflict areas, and Ik is the set of all conflicting priority streams
of stream k.
Probabilities of priorities for pedestrians are used to adjust the capacity for pedestrian
movements. Probability of pedestrian priority gives the fraction of pedestrians con-
sidered occupying the conflict area in the capacity analysis. The same approach can be
used to adjust the capacity for limited priority.
In conclusion: The capacity of a minor stream k is the product of maximum flow rate
qmax,k and the time available for queue discharge. This time is obtained in two steps.
First, busy periods are excluded. Secondly, during the remaining time the priority
stream is considered as a superposition (sum) of all conflict flows (hence the name
“ACF”), and the time periods having lags less than ta,i are excluded. The parameters
ta,i are related to major streams, not minor streams. If a minor stream crosses several
conflict areas, these are assumed independent, but each major stream is used only once
in the calculations.
The proportion of occupancy (ρA ) of queueing/discharging movements across a conflict
area (A) is based on discharge headways (tq,i , i ∈ A), not on movement capacities as in
the gap acceptance theory. Consequently, the capacity of a minor flow can be calculated
in a single step. Rank 1 vehicles are also included in the queuing process of a service
area. This assumption reduces the capacity estimate. In fact, if all major streams are
assumed to have Rank 2 and the minor stream is assumed to have Rank 3, ACF and
HCM2000 give very similar results, as a comparison of Figures 3.31 and 3.22 indicates.
(Parameters used in Figure 3.31 are described below.)
1200
← qII=0
1000
Rank 3 stream capacity (veh/h)
← ACF
800
← q =300
600 II
400
← qII=600
200
← q =900
II
0
0 500 1000 1500
Total priority flow rate (veh/h)
Figure 3.31: Capacity of a Rank 3 stream, when Rank 2 flow rate is qII veh/h, according to
HCM2000 (dashed curves) and ACF (solid curve)
Figure 3.32 displays a queuing model for the ACF procedure. The model is microscopic
for major streams and macroscopic for minor streams. The discussion above (Sections
3.5.3 and 3.5.5) about the time available for gap acceptance and the major stream
headway distributions during these time periods is relevant here also.
Queue discharge rate

qmax,k
Busy period
Time
tq,i tq,i tq,i ta,i tq,i ta,i tq,i
1 2 3 4 5
Figure 3.32: ACF queuing model for major stream i and minor stream k
Parameters ta,i and tq,i are estimated using statistical procedures based on goodness of fit
between observed and estimated capacities. Because the hierarchy of major streams is
ignored, the estimates of tq,i may be sensitive to the distribution of the total priority flow
between different priority streams. The estimation procedure can adjust the estimates
of ta,i for the fact that an isolated major stream vehicle blocks a minor stream for time
period ta,i + tq,i , not for ta,i .
Figures 3.31 and 3.33 display comparisons of HCM2000 and ACF capacities for Rank 3
and Rank 4 movements. The critical gap was tc = 5 s and follow-up time tf = 3 s in the
HCM method for all movements. In the ACF method all movements had tq = tf = 3 s
and ta = tc − tf /2 = 3.5 s. Because the interactions between major streams are reduced
to a simple queuing process, the distribution of total priority flow among different
priority streams does not affect the ACF capacity estimate, when tq and ta are the
same for all movements. Consequently, the Rank 3 capacity according to ACF is
approximately equal to the HCM2000 capacity when all major flows are assigned to a
Rank 2 movement.
1200
1000
Rank 4 stream capacity (veh/h)
800
qII=0→
ACF →
600
400
qII=300→
200
qII=600→
0
0 500 1000 1500
Total priority flow rate (veh/h)
Figure 3.33: Capacity of a Rank 4 stream, when Rank 3 flow rate is qIII = 300 veh/h, according
to HCM2000 (dashed curves) and ACF (solid curve)
DELAYS AT UNSIGNALIZED INTERSECTIONS
4 DELAYS AT UNSIGNALIZED INTERSECTIONS

4.1 Definitions of delay
Delay indicates an excess time consumed in a transportation facility in comparison
to a reference value, which Kimber et al. (1986) have called a comparator. It is a
performance measure experienced by road users. Although it has been shown that
the users’ perception of quality of service may be difficult to measure (Pécheux,
Pietrucha & Jovanis 2000), delay is a widely used service measure for intersections
(see Transportation Research Board 2000, FGSV 2001). Delay is also a fundamental
parameter in the economic analysis of highway investments.
In capacity analysis a continuous queue at a subject lane was assumed. The details of
the queuing process were insignificant. Because delay is the time difference between
a realized and an ideal operation time, an analysis of the queuing process is necessary
in delay models.
Queuing processes at unsignalized intersections have been described above in sections
2.4 and 3.5.2. Definitions of delays experienced in these processes are described below.
For a more detailed analysis the reader is referred to Leutzbach & Köhler (1974),
McDonald, Hounsell & Kimber (1984), Kimber et al. (1986), Lapierre & Steierwald
(1987), and Akçelik (1998).
In delay models driver behavior and gap-acceptance theory must be described in terms
of queuing theory. A vehicle at a stop or yield line is in service, vehicles waiting for
service behind the first vehicle are in queue, and vehicles approaching the intersection
upstream of the queue are arriving vehicles (Fig. 4.1). Queuing vehicles and the
vehicle in service are in the system. In queuing theory a distinction is made between
the mean waiting time in queue and the mean waiting time in the system (Gross &
Harris 1998). Acceleration, deceleration, and the physical dimensions of vehicles
are ignored. In traffic flow theory the physical dimensions and other properties of
vehicles and transportation facilities must be considered. Consequently, delay is more
appropriate as a performance measure than waiting time.
In system
Arriving In queue In queue In service
Figure 4.1: Vehicles at an intersection approach
According to HCM2000 (Transportation Research Board 2000) total delay is the sum
of all components of delay:
• Control delay is measured by comparison with the uncontrolled condition

• Traffic delay results from vehicular interaction causing drivers to reduce their
speed below the free-flow speed
• Geometric delay is caused by geometric features requiring vehicles to reduce
their speed in negotiating a facility
• Incident delay is the time loss due to an incident, compared with the no-incident
condition.
Total delay is the difference between the travel time actually experienced and the refer-
ence travel time (comparator) that would result during base conditions, in the absence
of control, traffic, geometric, or incident delay. In the analysis of unsignalized inter-
sections delays caused by vehicular interaction (traffic delays) are included in control
delays. The effects of incidents, such as road maintenance, should be analyzed separ-
ately. Consequently, the most important delay components at unsignalized intersections
are control and geometric delays.
Figure 4.2 displays two space-time trajectories of an isolated vehicle at a yield-
controlled intersection. The first trajectory describes a vehicle entering the intersection
without any vehicular interaction. The vehicle approaches the intersection at approach
cruise speed (va ), decelerates (A–C) until it reaches the negotiation speed (vn ), proceeds
with that speed some distance (C–F1 ), and accelerates (F1 –G1 ) to exit cruise speed
(ve ). The details of the trajectory within the intersection are ignored (see McDonald
et al. 1984, Kimber et al. 1986, Akçelik 1998).
Distance
G1 G2
ve ve
B E 1 F1 F2
Yield line D2 H1 E2 H2
C
va A: Start of deceleration
B: Virtual arrival time to yield line
C: Decision to stop or to proceed at negotiation speed
D: Stopping
E: Crossing yield line
F: Acceleration to exit cruise speed
A G: Proceed at exit cruise speed
H: Virtual departure time
Time
Figure 4.2: Trajectories of an isolated vehicle at a yield-controlled intersection
In the comparator trajectory, vehicles enter the intersection at approach cruise speed
(va ) and depart at exit cruise speed (ve ), with an instantaneous change of speed at yield
line. The time lost due to deceleration, negotiation speed, and acceleration is the time
interval between virtual arrival and departure times T (B, H1 ). Because this excess time
is based on the geometry of the intersection, not on any traffic interactions or traffic
control, the lost time is called geometric delay
Wg(1) = T (B, H1 ). (4.1)
It is the sum of deceleration delay Wd(1) = T (B, C), negotiation delay Wn(1) = T (C, F1 ),
and acceleration delay Wa(1) = T (F1 , H1 ). Accordingly, the geometric delay of vehicle
i is
Wg(i) = Wd(i) + Wn(i) + Wa(i) (4.2)
(Kyte et al. 1996). It includes the delay caused by speed reduction in order to check
for possible conflicts, even if there are none. In this discussion the time lost due to
checking is included in the negotiation delay.

Kimber et al. (1986) call Wg the measurable geometric delay. Pure geometric delay is
the delay of an isolated vehicle knowing that it is safe to drive through an intersection
immediately on arrival. It is a theoretical, under normal operating conditions non-
measurable delay.
If a vehicle has to stop, it decelerates and stops (A–D2 ), waits for the next antiblock
(D2 –E2 ), accelerates to negotiation speed (E2 –F2 ), proceeds, and accelerates to the exit
cruise speed (F2 –G2 ). At stop-controlled intersections minor road vehicles have to stop,
even if the major stream lag is safe for entry.
Total delay is the difference between virtual arrival and departure times
Wt(2) = T (B, H2 ). (4.3)
For a vehicle that waits at the yield line, but does not queue, the total delay is the sum
of deceleration delay, stop delay, negotiation delay, and acceleration delay:
Wt(i) = Wd(i) + Ws(i) + Wn(i) + Wa(i) . (4.4)
Total delay can also be defined as the travel time between two points upstream and
downstream of the intersection, unaffected by it, subtracted by the travel time of the
comparator:
Wt(2) = T (A, G2 ) − [T (A, B) + T (H2 , G2 )] . (4.5)
For a vehicle coming to a complete stop deceleration and acceleration delays are larger
than these components in the geometric delay. If no negotiation delay is assumed for
a stopping vehicle (Wn(i) = 0), all delays are caused by vehicular interactions, and
geometry does not cause any additional delay.
In Figure 4.2 it is assumed that geometry does not cause any extra delay to a stopping
vehicle. However, compared to a vehicle able to cross the intersection without any
vehicular interaction, the extra delay, “net control delay”, is T (H1 , H2 ). If a turning
vehicle first accelerates to an exit negotiation speed and it can accelerate to exit cruise
speed only after exiting the intersection, geometric delay cannot be ignored, but delay
due to geometry is much less than the geometric delay Wg (without control delay). This
indicates a dependency between control delay and geometric delay, and suggests two
alternative definitions:
1. A control-oriented definition of control delay measures all delays T (B, H2 ) due

to traffic signs and vehicular interactions. As total delays increase, the impact
of geometric conditions decreases. Geometric delay is the extra delay due to
geometry. It is the difference between total delay and control delay:
Wg = Wt − W, (4.6)
According to this definition, geometric delay is a function of control delay. In
Figure 4.2 a stopped vehicle (2) can accelerate from stop to exit cruise speed
without any additional delay due to geometry (negotiation speed). The control-
oriented model assigns no geometric delay to this vehicle.
2. A comparator-oriented definition uses geometric delay as a comparator, so that
control delay is the difference between total delay and geometric delay
W = Wt − W g . (4.7)
This is the actual excess delay T (H1 , H2 ) caused by traffic signs and vehicular
interactions.
The definition of control delay in HCM2000, as presented above, suggests a comparator-

oriented definition. It is also compatible with the general definition of delay given in the
beginning of this section. According to Troutbeck (2000) the control delay comprises
the time spent queuing, the service time, and deceleration and acceleration delays
at speeds below negotiation speed. In the discussion below the comparator-oriented
definition of control delay is assumed.
In queuing models vehicles are assumed to arrive with instantaneous deceleration, and
all stopping vehicles are assumed to stop at the stop/yield line and stack vertically. The
arrival time to the system is the earliest time instant (E1 in Figure 4.2) that the vehicle
would be able to cross the stop/yield line in case of no vehicular interaction. The queue
discipline is first-in-first-out (FIFO); i.e., vehicles depart in the same order as they have
arrived.
Figure 4.3 displays vehicle trajectories assuming no geometric delay, so that control
delay is equivalent to total delay. Deceleration and acceleration rates are finite and
queues have a physical length.
Distance
Control delay W (1)
Wd(1) Ws(1) Wa(1)
Block
Yield line
tf(1) tf(2) tf(3) tf(4)
Waiting time in queue WQ(3) Service time S (3)

1 2 3 4 5
Time
Figure 4.3: Vehicle trajectories for a through movement at a yield-controlled intersection with
no geometric delay
Vehicular delays at an unsignalized intersection can be described in terms of waiting

times in queuing systems as follows: A vehicle enters the system (queue or service)
at its virtual arrival time. This is the virtual arrival time under prevailing geometric
conditions (E1 in Figure 4.2). It is the first time instant that the vehicle can cross the
yield line in case of an antiblock. Waiting time in queue (WQ ) begins when a vehicle
enters a queue and ends when the vehicle enters the yield line. The virtual arrival time
from the queue to yield line is assumed, because it is the time instant that a vehicle
would cross the yield line in case of an antiblock. When a vehicle enters the yield line,
its service time (S) begins (see section 3.5.2). The waiting time in system is the sum
of waiting time in queue and service time: WS = WQ + S. Average control delay can
now be expressed as
W̄ = W̄Q + S̄ − tf + W̄a
(4.8)
= W̄S − tf + W̄a .
If a vehicle can cross the yield line without stopping (vehicles 2 and 4), its acceler-
ation delay is smaller than the acceleration delay of a stopped vehicle.1 When W̄a
1 Greenshields, Shapiro & Ericksen (1947) made a distinction between functional and chronotropic
is defined as the average acceleration delay in excess of negotiation and acceleration

delays, T (E1 , H1 ), due to geometric conditions, equation (4.8) is compatible with the
comparator-oriented definition of control delay.
The follow-up times in Figure 4.3 have different lengths. This is caused by the early
acceleration of vehicles accepting a lag. At signalized intersections the longer discharge
headways for the first few discharging vehicles are a well know phenomenon (see King
& Wilkinson 1976, Briggs 1977, Teply & Jones 1991, Niittymäki & Pursula 1997). If
a stream has to observe a stop sign, the vehicles cannot benefit from early acceleration.
In mathematical models all follow-up times of a stream under given traffic conditions
are usually assumed to be constant. This approach has some empirical support from
the measurements by Koivisto (1999).
The discussion here applies to delays of minor stream vehicles. Concerning disturb-
ances in a Rank 1 stream the reader is referred to section 3.7 (see also Bonneson &
Fitts 1999).
Figure 4.4 displays an input-output diagram of a minor stream (see also Daganzo 1997).
The vertical distance between the curves indicates the number of vehicles in the system
at a given time, assuming that vehicles form a vertical stack at the stop line (see Fig.
2.13). The horizontal distance between the curves gives the waiting time in system of
a given vehicle. The area between the curves is equal to the cumulative delay during
the observation period.
4.2 Service times

4.2.1 Definition of service time
In queuing systems service time is the time that a customer spends at a counter. At
unsignalized intersections the service time of a vehicle begins when it reaches the
yield/stop line. The entry time is the virtual entry time; i.e., the first time instant that
the vehicle could cross the stop line in case on an antiblock (see Figure 4.3). Service
time ends when the follow-up time has passed since the vehicle has crossed the stop
line. Accordingly, service time is the sum of a waiting time at the yield/stop line during
a block and the follow-up time, as described in Section 3.5.2.
Vehicles can enter service either from a queue or without queuing. Queuing vehicles
enter service immediately when the service of the vehicle ahead has been completed.
It is evident that the remaining lag is not shorter than tc − tf . Isolated (non-queuing)
vehicles arrive to an empty system at a random time instant, but after the follow-up time
has passed since the previous vehicle had crossed the stop line. The arrival process of
vehicles that find an empty system is not independent of the arrival process of vehicles
in the major stream (Daganzo 1977). Because the expected state of the major stream at
the beginning of the service time is different for queuing and isolated vehicles, they have
different service time distributions. The service time distribution of all minor stream
vehicles is a mixture of service time distributions for queuing and isolated vehicles.
The basic properties of service times for both arrival types is discussed next. In order to
describe the dynamics behind the service processes, the equations for average service
decelerations. Functional deceleration is necessary to adjust speed for intersection geometry or to obey
a stop sign. A vehicle may accelerate as soon as the specific deceleration or stopping function has been
fulfilled. Chronotropic deceleration (“chronoceleration”) occurs when motion beyond a specific point
cannot be continued until after a specific time instant, such as the beginning of an antiblock or end of the
follow-up time. A driver may save time by an early deceleration so that the vehicle can enter an intersection
at the beginning of an antiblock or at the end of a follow-up time at a higher speed than possible with a
later deceleration. See the trajectory of vehicle 4 in Figure 4.3.
40
35
30
Number of vehicles
25
20
15
10
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (min)
Figure 4.4: Cumulative arrivals (thick curve) and departures (thin curve) of a minor stream
(major flow 800 veh/h, minor flow 400 veh/h, tc = 5 s, tf = 3 s, exponential headway distribu-
tions)
times are derived. In order to keep the mathematics simple, the second moments of the
service time distributions are presented with bibliographic references only.
4.2.2 Average service time of queuing vehicles

Under saturated conditions a minor stream vehicle enters service immediately after the
service of the previous vehicle has ended. This is demonstrated by vehicle 2 in Figure
3.17. The vehicle enters a yield/stop line (service) exactly tf after the previous vehicle
has crossed the stop line. The remaining lag is not shorter than tc − tf . If the lag is
greater than or equal to tc , the vehicle can cross the yield/stop line immediately. The
probability of no waiting at a yield/stop line is
Rϒ (tc )
pq,0 = P{ ϒ ≥ tc | ϒ ≥ tc − tf } = . (4.9)
Rϒ (tc − tf )
When the major stream headways are exponentially distributed the probability is
Rϒ (tc )
pq,0 = = e−λtf . (4.10)
Rϒ (tc − tf )
For vehicles discharging from the yield/stop line immediately upon entry the service
time is equal to the follow-up time tf .
The time interval from the beginning of an antiblock to the beginning of the next
antiblock can be called a cycle (c). The average service time of vehicles discharging
from a queue is the average cycle length divided by the average number of discharging
vehicles (Dc ) in a cycle:
E[c]
S̄q = . (4.11)
E[Dc ]
The average cycle length is the sum of a headway T ≥ tc and the sum of consecutive
headways shorter than the critical gap. If major stream headways are exponentially
distributed, a headway T ≥ tc follows shifted exponential distribution with mean
1
E[ T | T ≥ tc ] = + tc . (4.12)
λ
In order to estimate the expected sum of gaps in a block, we need to derive the expecta-
d
tion of a sum of a random number of i.i.d. random variables. Let X = { T | T < Tc } be
a headway shorter than the critical gap. The sum YN = X1 +· · ·+XN of N consecutive
headways shorter than the critical gap has expectation
% N & % % n &&

E[YN ] = E Xi = E E Xi | N = n , (4.13)
i=1 i=1
where X1 , . . . , Xn is a random sample of variate X, and the inner expectation has been
conditioned2 on the event N = n. Because Xi and N are independent, the conditional
expectation is
% n & % n &

E Xi | N = n = E Xi = n E[Xi ] = n E[X] . (4.14)
i=1 i=1
This is the expected value of the sum assuming that the number (n) of consecutive
headways shorter than the critical gap is known. However, as the number (N ) of
headways is a random variable, the expectation of YN is (Goodman 1988)
E[YN ] = E[N E[X]] = E[N] E[X] . (4.15)
Thus, the expected value of the sum is the expected number of headways shorter than
the critical gap multiplied by their expected length. Ross (1996) has given the result
(4.15) using moment generating functions.
A headway is shorter than the critical gap with probability
pg = P{T < tc } = F(tc ) , (4.16)
and the probability of n consecutive headways shorter than the critical gaps follows the
geometric distribution
pn = pgn (1 − pg ). (4.17)
The expected number of consecutive headways Ti < tc is obtained from equation (2.11)
as
pg F(tc )
E[N ] = = = eλtc − 1. (4.18)
1 − pg R(tc )
The expected length of a headway shorter than the critical gap is
E[X] = E[ T | T < tc ]
tc
= t f( t | t < tc ) dt
0
tc
1
= t f(t) dt
F(tc ) 0 (4.19)

1 −λt 1 tc
= e −t −
1 − e−λtc λ 0
1 tc
= − λt .
λ e c −1
2 For conditional expectations see e.g. Grimmett & Welsh (1986).
The expected cycle length can now be expressed as
E[c] = E[ T | T ≥ tc ] + E[N ] E[ T | T < tc ]

1 (4.20)
= −λt .
λe c
The number of departures during a headway T ≥ tc follows the geometric distribution

P{ Dc = d | T ≥ tc } = pq,0
d
1 − pq,0 . (4.21)
The expectation is obtained following equation (2.9) as

1
E[Dc ] = . (4.22)
1 − pq,0
For exponential headways pq,0 = e−λtf and E[Dc ] = (1 − e−λtf )−1 .

The average service time of vehicles discharging from a queue (arriving while server
was busy) is now obtained as
E[c] 1 − e−λtf
S̄q = = . (4.23)
E[Dc ] λe−λtc
This is the inverse of capacity equation (3.12), as expected. Consequently, this approach
can be used as an alternative derivation of the capacity equation. For a more rigorous
derivation of S̄q see Kremser (1962), who also derived the second moment of the service
time as
' ( 2eλtc
E Sq2 = 2 eλtc − λtc 1 − e−λtf − λtf e−λtf . (4.24)
λ
The average waiting time at yield/stop line for vehicles discharging from a queue is
1 − e−λtf
W̄q = S̄q − tf = − tf . (4.25)
λe−λtc
After each vehicle there is a time period tf during which the next vehicle is moving to
the yield/stop line and no vehicles are waiting at the stop line. This time interval is part
of the service time of the first vehicle and the queuing time of the second vehicle.
4.2.3 Average service time of isolated vehicles

An isolated (non-queuing) minor stream vehicle (vehicles 1 and 3 in Figure 3.17) arrives,
when the server is idle, and enters service (yield/stop line) immediately. The vehicle
can arrive after tf has passed since the last vehicle in the preceding queue has crossed
the yield/stop line. No major stream vehicle can arrive until tc − tf has passed since the
server became idle. However, it is possible that a major stream vehicle arrives before
the isolated minor stream vehicle reaches the stop line.
If a subject vehicle is blocked by higher priority vehicles, as vehicle 1 in Figure 3.17,
it must wait for the next antiblock. Once the antiblock starts, the vehicle crosses the
yield/stop line immediately. However, the server is busy until the follow-up time has
passed. Thus, the service time, S̄e , of a vehicle arriving to an empty system is the sum of
average waiting time at stop line We and the follow-up time tf (Yeo & Weesakul 1964).
An isolated vehicle enters an intersection at a random time and can enter the yield/stop
line without queuing. If the vehicle enters during an antiblock, it can depart immedi-
ately. Let us assume that, when a non-queuing minor stream vehicle arrives, the major
stream is in steady state (see, however, Daganzo 1977). The probability of no wait is
equal to the probability of a lag greater than or equal to the critical gap:
pe,0 = P{ϒ ≥ tc } = Rϒ (tc ) . (4.26)
This is equal to the “transparency”; i.e., the percentage of time a driver would consider
it safe to merge or cross the major stream (Herman & Weiss 1961, Weiss 1963, Allan
1968).
The vehicle has to stop and wait if the lag available is shorter that the critical gap. The
probability of wait is
pe,l = P{ϒ < tc } = Fϒ (tc ) . (4.27)
If headways are independent and identically distributed, the probability that the next
headway (gap) cannot be accepted is
pg = P{T < tc } = F(tc ) , (4.28)
and the probability that a driver has to wait for one lag and n − 1 gaps (n vehicles) is
pn = pe,l pgn−1 (1 − pg ). (4.29)
If major stream headways follow the negative exponential distribution, lag and gap
distributions are equal:
Fϒ (tc ) = F(tc ) = 1 − e−λtc , (4.30)
and p = pe,l = pg , so that pn = pn (1 − p). This is a geometric distribution with
expectation obtained from equation (2.11) as follows:
p
E[N ] = = eλtc − 1. (4.31)
1−p
The expected length of a headway shorter than the critical gap is given in equation
(4.19) as
1 tc
E[ T | T < tc ] = − λt . (4.32)
λ e −1c
Following equation 4.15 the average waiting time during a block for a vehicle arriving
to an empty system can be expressed as the product of the expectations (Greenshields
& Weida 1952):
eλtc − 1
W̄e = E[N] E[ T | T < tc ] = − tc . (4.33)
λ
The first term in the equation is the average service time of vehicles departing from
a queue with follow-up time equal to critical gap.3 W̄e is the lower limit for minor
stream average delay (Transportation Research Board 1997). As the major flow (λ)
approaches zero, the first part in the equation approaches tc , and W̄e approaches zero.
Because Adams (1936) was the first to present the result, it is also called the Adams’
delay. He solved the problem in terms of the average waiting time of pedestrians
3 When follow-up time (t ) is equal to critical gap (t ) the potential capacity under exponential major
f c
stream arrivals is obtained from equation (3.12) as follows:
λe−λtc λ
Cp = = λt .
1 − e−λtc e c −1
See Grabe (1954).
attempting to cross a traffic stream. Because follow-up times are ignored, average
service time approaches zero and capacity tends to infinity at very low major flow rates.
The distribution of this waiting time has been studied by Garwood (1940), Tanner
(1951), and Mayne (1954). Tanner (1951) and McNeil & Weiss (1974) have derived
the variance of this waiting time
e2λtc − 2λtc eλtc − 1
σe2 = (4.34)
λ2
as well as the mean and variance of the waiting time for isolated minor stream vehicles
at a yield controlled intersection, when the critical lag is shorter than the critical gap.
Figure 4.5 demonstrates that standard deviation σe is slightly higher than the Adams’
delay, which indicates a larger variation than exponentially distributed waiting times
would have. Cowan (1984) has generalized the Adams’ formula for M3 distributed
major stream headways.
14
12
10
Waiting time (s)
0
0 500 1000 1500
Major flow rate (veh/h)
Figure 4.5: Mean (W̄e , solid line) and standard deviation (σe , dashed line) of waiting time for
isolated vehicles, when major stream headway distribution is exponential, tc = 5 s, and tf = 3 s
When a vehicle has found an acceptable lag or gap, it enters the intersection. If there is
another vehicle behind the subject vehicle, it can enter the intersection after the follow-
up time has passed. Until that time, the server is busy (Fig. 3.17). Accordingly, the
average service time of isolated vehicles is (Kremser 1962)
eλtc − 1
S̄e = W̄e + tf = − (tc − tf ). (4.35)
λ
At very low major flow rates the average service time of an isolated vehicle (S̄e ) ap-
proaches the follow-up time (tf ), as Figure 4.6 demonstrates. As major flow rate
increases, the average service time for isolated vehicles increases slightly more steeply
than the average service time of queuing vehicles. (In some papers W̄e has been used
in stead of S̄e , which explains differences in the results.)
The second moment of the service time has been derived by Kremser (1962) as
λtc λtc
2 e −1 e
E Se = 2 − tc + tf − tc + tf2 − tc . (4.36)
λ λ
16
14
12
Average service time (s)

10
0
0 500 1000 1500
Major flow rate (veh/h)
Figure 4.6: Average service time of isolated vehicles (S̄e , solid line) compared with the average
service time of queuing vehicles (S̄q , dashed line) when major stream headway distribution is
exponential, tc = 5 s, and tf = 3 s
As stated above, this discussion was based on the assumption that, when an isolated
(non-queuing) minor stream vehicle arrives, the major stream is in steady state. Kremser
(1964) and Daganzo (1977) have demonstrated that this assumption is not correct. An
isolated minor stream vehicle can arrive only when the server is idle. When the server
becomes idle, there must be a lag ϒ ≥ tc − tf . Consequently, the arrival process
of isolated minor stream vehicles and the arrival process in the major stream are not
independent.
Let us assume that the last vehicle in a minor-stream (i) queue departed at time 0 so that
the busy period ended at time tf . Thus, the arrival time (τ ) of an isolated minor stream
vehicle can not be earlier than tf , and the arrival time (τk ) of the next major stream (k)
vehicle can not be earlier than tc . The lag found by the isolated minor stream vehicle
is (Daganzo 1977)

tc − τ + X, if tf ≤ τ < tc
ϒ(τ ) = (4.37)
X, if tc ≤ τ,
where X follows the negative exponential distribution with scale parameter λk (see
Figure 4.7). If tf = tc , a non-queuing vehicle cannot enter before the critical gap has
passed since the previous departure, and the lag is exponentially distributed.
If the arrival time of the minor stream vehicle is τ ≥ tc , the lag is exponentially
distributed:
f1 (υ) = λk e−λk υ . (4.38)
If the minor stream vehicle arrives before the critical gap of the previous vehicle has
passed, the lag follows the shifted exponential distribution with location parameter
tc − τ ; i.e.,
f2 (υ, τ ) = λk e−λk (υ+τ −tc ) . (4.39)
A given lag υ can realize with different arrival times. If the lag is υ ≥ tc − tf , the minor
stream vehicle can have arrived at any time in the interval tf ≤ τ < tc . However, a lag
tc − τ X
τ
0 tf tc τk Time
τ − tf ϒ(τ )
Figure 4.7: Lag for an isolated minor stream vehicle arriving before the critical gap has passed
since the departure of the preceding vehicle
υ < tc − tf is only possible, if τ > tc − υ. The density function of the lag distribution
is obtained as

t
P{τ > tc } f1 (υ) + tcc−υ f(τ ) f2 (υ, τ ) dτ, if υ < tc − tf
fϒ (υ) = t (4.40)
P{τ > tc } f1 (υ) + tfc f(τ ) f2 (υ, τ ) dτ, if υ ≥ tc − tf ,
where f(τ ) is the density function of the minor stream arrival times.
When the minor stream arrival process is Poisson with flow rate λi , the probability
density function of minor stream arrival time (given that it is larger than the follow-up
time) is
f(τ ) = λi e−λi (τ −tf ) . (4.41)
The probability that a non-queuing vehicle arrives after the critical gap has passed since
the last departure is
∞
P{τ > tc } = f(τ ) dτ = e−λi (tc −tf ) . (4.42)
tc
The density function of the lag distribution can now be expressed as (Daganzo 1977)

e−λi (tc −tf ) λk e−λk υ + tc λi e−λi (τ −tf ) λk e−λk (υ+τ −tc ) dτ, if υ < tc − tf
tc −υ
fϒ (υ) =
e−λi (tc −tf ) λ e−λk υ + tc λ e−λi (τ −tf ) λ e−λk (υ+τ −tc ) dτ, if υ ≥ tc − tf ,
k tf i k
(4.43)
which gives (Daganzo 1977)


 λ e−λi (tc −tf ) −λk υ
 k
 λk e + λ i e λi υ , if υ < tc − tf
λi + λk
fϒ (υ) = (4.44)

 λk e−λk υ −λi (tc −tf )

 λk e + λi e λk (tc −tf )
, if υ ≥ tc − tf .
λ i + λk
When tc = tf , equation (4.44) reduces to the negative exponential lag density (2.15).
This indicates that the assumption of steady state at time τ is correct only when the
critical gap is equal to the follow-up time (Daganzo 1977, Kremser 1964).
The probability of no wait is
pe,0 = P{ϒ > tc }

∞
= fϒ (υ) dυ
tc (4.45)
e −λ (tc −tf )
−λk tc
= λk e i + λi eλk (tc −tf ) .
λi + λk
Daganzo (1977) has pointed out that this probability is always larger than the probability
(4.26) based on exponential lags. Models based on equation 4.26 tend to predict longer
delays.
The probability of not accepting a lag is
pe,l = 1 − pe,0
e−λk tc −λi (tc −tf ) (4.46)
=1− λk e + λi eλk (tc −tf ) .
λi + λk
The average service time for an isolated vehicle was derived by Daganzo (1977) as

S̄e = pe,l −Lϒ (0) − (eλk tc + 1)LT (0) + tf , (4.47)
where Lϒ and LT are the Laplace transforms of rejected lags and gaps, respectively.
The first derivatives of these transforms at parameter value zero are (Daganzo 1977)

λk e−λi (tc −tf ) 1 λi (tc −tf ) 1 −λk tc 1 1
Lϒ (0) = tf − tc + e − +e tc − −
pe,l (λi + λk ) λi λi λk λk
(λi +λk )(tc −tf )
λi e
+ 2
e−λk tc (λk tc + 1) − e−λk (tc −tf ) [λk (tc − tf ) + 1]
λk
(4.48a)
tc e−λk tc 1
LT (0) = − . (4.48b)
1−e kc−λ t λk

Pöschl (1983) has presented the second moment E Se2 of the service time. The equation
is, however, not reproduced here, because it is even more complicated than the equation
for average service time. According to Daganzo (1977) the magnitude of error when
assuming an exponentially distributed lag density (2.15) instead of the correct density
function (4.44) is not likely to be large for most reasonable values of parameters.
4.3 Delays under steady-state conditions

4.3.1 Queuing as a stochastic process
Let us assume that the average arrival rate of a minor stream is λ. Vehicles cross the
stop line following the gap acceptance procedure. Vehicles waiting for entry form
a vertical stack at the stop line and are served following the first-come-first-served
principle. Under steady state conditions the arrival process and the service process stay
unchanged, the average arrival rate does not exceed capacity, and queues have reached
an equilibrium state.
The state of the process at time τ is described by the number N (τ ) of vehicles in the
system. When a minor stream vehicle arrives, the state of the process increases by one
unit. When a minor stream vehicle departs the intersection (tf after crossing the stop
line), the state of the process decreases by one unit. It is assumed that only one vehicle
can arrive or depart simultaneously.
Cooper (1981) has presented an important theorem, which can be applied in the queuing
process under discussion:
In any queuing system for which the realizations of the state process are step
functions with only unit jumps (positive or negative), the equilibrium state
distribution just prior to arrival epochs is the same as that just following
departure epochs. In addition, when the input is Poisson this distribution
is the same as the outside observers distribution.
Consequently, if minor stream arrivals are assumed to be Poisson, it is only necessary to

describe the queuing process at times of departure from service (see also Tanner 1962).
This approach will be followed here.
Let δn be the service departure time of vehicle n. The time interval between two
consecutive service departure times is Tn = δn − δn−1 . If the number of vehicles in the
system after departure n is Nn , the number of vehicles in the system immediately after
the next departure is (Neuts 1989)
Nn+1 = (Nn − 1)+ + A(δn , δn+1 ) = (Nn − 1)+ + An+1 , for n ≥ 1, (4.49)
where An+1 is the number of arrivals between departures n and n + 1. The number of
vehicles in the system after the next departure depends only on the number of vehicles
in the system after current departure and the number of arrivals between the current
and the next departures. This type of process is called an embedded Markov chain
(see Kendall 1951, Kendall 1953, Ross 1996). At transition points (times of departure
from service) the process behaves like a discrete-time Markov chain having a one-step
transition probability matrix
 
p00 p01 p02 ···
p10 p11 p12 · · ·
 
 · · ·
P =  0 p21 p22 , (4.50)
 0 0 p32 · · ·
 
.. .. .. ..
. . . .
where pij is the probability that between two consecutive departures the number of
vehicles in the system changes from i to j . Because P is a stochastic matrix, its rows
sum to one:
∞

pij = 1. (4.51)
j =0
The transition probabilities can be derived from equation (4.49). If i ∈ {1, 2, 3, . . . }

a vehicle (n) arrives to stop line (service) immediately when the previous vehicle
(n − 1) has departed service. The transition interval between departures n − 1 and
n is the service time of vehicle n. The probability pij of transition from state i to
j ∈ {i − 1, i, i + 1, . . . } is the probability αj −i+1 of j − i + 1 arrivals during the
service time. Because the arrival process is Poisson, it is not necessary to know the
backward waiting time (see Figure 2.6 on page 32) at the time of service completion.
If i = 0 a vehicle (n) arrives to stop line (service) some time after the previous vehicle
(n − 1) has departed service. The transition interval between departures n − 1 and
n is the idle time of the server plus the service time of vehicle n. However, vehicle
n + 1 cannot arrive before vehicle n. The probability p0j of transition from state 0 to
j ∈ {0, 1, 2, . . . } is the probability βj of j arrivals during the service time of vehicle n
that has arrived during an idle period. The transition probability matrix can be expressed
as  
β 0 β1 β2 · · ·
α0 α1 α2 · · ·
 
 
P =  0 α0 α1 · · · . (4.52)
 0 0 α0 · · ·
 
.. .. .. . .
. . . .
The transition probabilities are independent on n and depend on i and j only through
their difference. The rows sum to one:
∞

αk = 1
k=0
∞
(4.53)

βk = 1.
k=0
Let fq (u) and fe (u) be the probability density functions of service time distributions
for vehicles arriving to a queue and to an empty system, respectively. Because minor
stream arrivals are Poisson, the transition probabilities are
∞
(λu)k −λu
αk = e fq (u) du
0 k!
∞ (4.54)
(λu)k −λu
βk = e fe (u) du.
0 k!
These are probabilities of k arrivals during a service period. The process is equivalent
to an M/G2/1 queuing system. If fq (u) = fe (u), the transition probability matrix
describes an M/G/1 queuing system (see Neuts 1989).
4.3.2 M/G/1 model

If the arrival pattern of vehicles approaching an intersection in one lane is assumed
to follow the Poisson process, and vehicles are served by a single server having some
general i.i.d. service pattern, the system can be described as an M/G/1 queuing process.4
Because all service times are assumed to be independent and identically distributed,
αi = βi . The transition probabilities are
∞
(λu)k −λu
αk = e fS (u) du (4.55)
0 k!
where fS (u) is the probability density function of service times. The one-step transition
probability matrix is  
α0 α1 α2 · · ·
α0 α1 α2 · · ·
 
 0 α0 α1 · · ·
P= . (4.56)
 0 0 α0 · · ·
 
.. .. .. . .
. . . .
Let us define
πj = lim P{Nk = j } , j ∈ {0, 1, 2, . . . } (4.57)
k→∞
as the stationary (equilibrium) probability that in the long run the system is found in
state j . These probabilities can be expressed as
j +1

πj = π0 p0j + πi pij
i=1
(4.58)
j +1

= π0 αj + πi αj −i+1 , j ∈ {0, 1, 2, . . . } .
i=1
4 In Kendall’s (1953) notation M/G/1 stands for Markovian (Poisson) arrival process / General service
time distribution / 1 server.
The process can reach state j , if it first reaches some other state i from which it moves
to state j . The probabilities sum to unity
∞

πj = 1. (4.59)
j =0
This can be used as a normalization equation in further analysis.

The average waiting times can be derived either from the stationary probabilities (4.58)
which have probability-generating function
∞
(z − 1)LS (λ − λz)
Gπ (z) = πj z j = (1 − ρ), (4.60)
j =0
z − LS (λ − λz)
where LS (·) is the Laplace transform of the service-time distribution function, or dir-
ectly from equation (4.49) using mean values (Cooper 1981). The average waiting time
in queue is given by the Pollaczek-Khintchine formula (Pollaczek 1930, Khintchine
1960, Kendall 1951)5 :

ρ S̄ σS2 ρ S̄ 1 + CS2
W̄G,Q = 1+ 2 = , (4.61)
2(1 − ρ) S̄ 2(1 − ρ)
where S̄ is the mean and σS2 is the variance of service times, ρ = λS̄ is the server utiliz-
ation factor, λ is the arrival rate, and CS = σS /S̄ is the coefficient of variation of service
times. If S̄ is assumed to be equal to inverse capacity, the utilization factor is equal
to the D/C ratio. Because under low D/C conditions this assumption underestimates
ρ, it underestimates delay also. The average waiting time in the system, including the
service time, is (Gross & Harris 1998)
W̄G,S = W̄G,Q + S̄
# $
ρ 1 + CS2
= S̄ 1 +
2(1 − ρ) (4.62)

ρκ
= S̄ 1 + ,
1−ρ
where κ = (1 + CS2 )/2 is the randomness constant (Transportation Research Board

1997). It indicates how rapidly
the average delay increases with the D/C ratio (Troutbeck
1986). Because σS2 = E S 2 − S̄ 2 , the randomness constant is

E S2
κ= . (4.63)
2S̄ 2
Observing that ρ/S̄ = λ, the waiting times can be expressed as

λ E S2
W̄G,Q = (4.64a)
2(1 − ρ)

λ E S2
W̄G,S = S̄ + . (4.64b)
2(1 − ρ)
5 Khintchine presented the result in 1932 in a paper in Russian, independently of Pollaczek (Cooper
1981).
Little’s (1961) formula

L̄ = λW̄ (4.65)
gives the average number of vehicles in system or in queue, depending on the type of
average waiting time (W̄ ) used in the equation. Queue length distributions have been
studied by Heidemann (1991) and Wu (1994). Little’s formula (4.65) can also be used
to estimate average delay, when the average queue length is known. Daganzo (1977),
however, warns about some pitfalls in this technique, especially in connection with the
embedding technique.
In order to apply equation 4.62, the mean (S̄) and variance (σS2 ) of service times must
be estimated. It is usually suggested that the average service time of queuing vehicles
(S̄q = C −1 ) is used (Transportation Research Board 1997), although it may slightly
underestimate service times and delays (see Fig. 4.6).
The variance σS2 depends on the service time distribution. If all service times are
assumed constant, σS2 = 0, κ = 0.5, and the average waiting time in queue is
ρ S̄ ρ2
W̄D,Q = = . (4.66)
2(1 − ρ) 2λ(1 − ρ)
This queuing process is called M/D/1, where D stands for deterministic service times.
The assumption that all vehicles wait an equal time at yield/stop line is, of course,
unrealistic. The M/D/1 model has found better applications in the delay estimation of
signalized intersections (Webster 1958).
If both interarrival time (headway) and service time distributions are assumed to be
negative exponential, the queuing process is called M/M/1. In the negative exponential
distribution the standard deviation is equal to the mean (σS = S̄), so that CS = 1 and
κ = 1. The waiting times are obtained from the Pollaczek-Khintchine formula (4.61)
as
ρ S̄
W̄M,Q = (4.67a)
1−ρ
S̄
W̄M,S = . (4.67b)
1−ρ
According to Kimber & Hollis (1979) and Fisk & Tan (1989) the M/M/1 model will
produce a good approximation to a M/G/1 model having similar average service times.
The variation in service times doubles the waiting time in a queue over that in an M/D/1
system (W̄M,Q = 2W̄D,Q ). If S̄ is the inverse of capacity (C), the waiting time in system
can be expressed as the inverse of reserve capacity
3600
W̄M,S = . (4.68)
C−q
The relationship between delay and reserve capacity has been discussed e.g. by Brilon
(1988b) and Tracz, Chodur & Gondek (1990). This relationship demonstrates the
central role of a good capacity estimate in delay estimation. At high degrees of saturation
small errors in capacity estimates will result in significant errors in delay estimates
(Kyte, Dixon & Basavaraju 2003).
Because service times cannot be shorter than the follow-up time, a possible model
for service time distribution is the shifted exponential distribution, which has mean
S̄ = tf +θ −1 and standard deviation σS = θ −1 = S̄ −tf (Luttinen 1996). The coefficient
of variation is CS = 1 − tf /S̄. This indicates that the shifted exponential service time
model gives smaller delays than the M/M/1 model. Shifted exponential service time
distribution with location parameter equal to the follow-up time is equivalent to the
assumption of negative exponential waiting time distribution at yield/stop line.
The randomness constant is obtained as

tf tf
κ =1+ −1 . (4.69)
S̄ 2S̄
The average waiting time in the system can now be calculated as

tf
S̄ + ρtf −1
2S̄
W̄M ,S =
1−ρ (4.70)

ρtf tf
= W̄M,S + −1 .
1 − ρ 2S̄
Because tf /(2S̄) < 1, the waiting time W̄M ,S is less than W̄M,S .
The M/G/1 models discussed so far have ignored that vehicles arriving to an empty
system have different average service times (S̄e ) than queuing vehicles (S̄q ). In addition,
the variance of service times is based on arbitrary assumptions, not on solid theory.
In an M/G/1 queuing process the proportion of time the server is busy is equal to
the utilization factor ρ = λS̄. The distribution of service times is a mixture of two
distributions. The proportion ρ of service times are from the distribution fq for queuing
vehicles (see Section 3.5.3), and the proportion 1 − ρ are from the distribution fe for
isolated vehicles:
fS (s) = ρfq (s) + (1 − ρ)fe (s) . (4.71)
The average service time can be estimated as (SNRA 1995b, Pöschl 1983)
S̄ = ρ S̄q + (1 − ρ)S̄e (4.72)
The service quotients for queuing (ρq ) and non-queuing vehicles (ρe ) are
ρq = λS̄q (4.73a)
ρe = λS̄e . (4.73b)
The D/C ratio and the average service time are now obtained in terms of component
flows as (Pöschl 1983)
ρe
ρ = λS̄ = (4.74)
1 + ρ e − ρq
S̄e
S̄ = (4.75)
1 + ρ e − ρq
The probability that a randomly arriving vehicle finds the system empty is (Yeo 1962)
1 − ρq
p0 = 1 − ρ = . (4.76)
1 + ρ e − ρq
The second moment of the mixed service time distribution is

' (
E S 2 = ρ E Sq2 + (1 − ρ) E Se2
' (
ρe E Sq2 + (1 − ρq ) E Se2 (4.77)
= .
1 + ρ e − ρq
The average waiting time in queue is obtained from equation (4.64a) as

λ E S2
W̄G,Q =
2(1 − ρ)
' ( (4.78)
λ
= ρe E Sq2 + (1 − ρq ) E Se2 .
2(1 − ρq )
The average waiting time in system is obtained by adding the average service time to
W̄G,q :
W̄G,S = S̄ + W̄G,q
S̄e λ ' ( (4.79)
= + ρe E Sq2 + (1 − ρq ) E Se2 .
1 + ρ e − ρq 2(1 − ρq )
4.3.3 M/G2/1 model

It is possible to apply the M/G/1 model with a service time distribution that takes into
account the dependency between the major stream arrivals and the arrivals of non-
queuing minor stream arrivals. However, the M/G/1 model assumes that consecutive
service times are independent (Gross & Harris 1998); i.e., a sequence of service times
for queuing and non-queuing vehicles follows geometric distribution (cf. Cooper 1981).
The service times distributions behind transition probabilities αi and βi (4.55) are equal.
In assigning service times to customers the M/G/1 model does not make any distinction
between vehicles actually queuing or not queuing.
Yeo (1962) has presented a modified model, M/G2/1, which assigns service time Sq
to queuing customers and service time Se to customers arriving to an empty system.
The one-step transition probability matrix is given as in equation 4.52, and transition
probabilities follow equation 4.54 (see Neuts 1989). The average waiting time in queue
is (Yeo 1962) ' (
λ ρe E Sq2 + (1 − ρq ) E Se2
W̄G2,Q = . (4.80)
2(1 − ρq )(1 + ρe − ρq )
This differs from equation (4.78) by factor (1 + ρe − ρq )−1 . When average service
times for queuing and non-queuing customers are equal, the service quotients are also
equal (ρe = ρq ), and M/G2/1 model reduces to an M/G/1 model.
The average waiting time in system can be expressed as (Pöschl 1983)
W̄G2,S = S̄ + W̄G2,Q
λ ' (
S̄e + ρe E Sq2 + (1 − ρq ) E Se2 (4.81)
2(1 − ρq )
= .
1 + ρ e − ρq
Yeo & Weesakul (1964) have used the M/G2/1 model for the analysis of delays at
unsignalized intersections.
In an M/G/1 queuing system the distribution of queue lengths at service completion
times is equivalent to the queue size distribution at random times (see footnote 10 on
page 85.) Heidemann (1991) has shown that the M/G2/1 system also has this property.6
He also demonstrated that Little’s formula (4.65) can be applied in M/G2/1 systems.
6 Daganzo (1977) warns against assuming the equivalency between queue length distributions at a
random time and at the arrival time of a major stream vehicle (see Evans, Herman & Weiss 1964).
Lertworawanich & Elefteriadou (2003) have estimated storage (queue) lengths of left-
turn lanes. They found that, compared to the M/G2/1 model, the M/M/1 model provided
good estimates, except at near-capacity conditions.
4.3.4 Tanner’s delay model

Tanner (1962) described major stream as departures from an M/D/1 queuing system
with minimum headway tp (see page 41) and average arrival rate λ1 , and assumed a
Poisson process with average arrival rate λ2 for minor stream arrivals. Minor stream
vehicles cannot enter the intersection at shorter intervals than the follow-up time (tf )
and not within the critical gap (tc ) after the previous major-stream vehicle.
A block starts with a major-stream vehicle more than tc behind the previous major-
stream vehicle, and ends a time tc after a vehicle being separated from the next major-
stream vehicle by more than tc . The model is similar to the extension principle in
traffic signals (Fig. 2.12). Because tc > tp , the lag following a block is exponentially
distributed. The lag distribution is the same for both queuing and isolated vehicles.
Tanner modeled the queuing process at an unsignalized intersection as an embedded
Markov chain (see page 118) with two types of regeneration points:
1. The instants at which minor-stream vehicles cross the stop line
2. The ends of blocks in the major stream, if not included in the first type
These definitions are based on the description of gap acceptance, which is slightly
different from the description used in this report. See footnote 6 on page 44.
The state of the system at a regeneration point of type 1 is the number of minor
stream vehicles waiting after one has crossed the stop line. The state of the sys-
tem at regeneration point of type 2 is x. The transition probability from state i to j
(i = x, 0, 1, 2, . . . ; j = x, 0, 1, 2, . . . ) is pij . The system can now be described by a
transition matrix
 
pxx px0 px1 px2 · · ·
p0x p00 p01 p02 · · ·
 
 0 p10 p11 p12 · · ·
 
P= 0 0 p21 p22 · · · . (4.82)
 
 0 0 0 p32 · · · 

.. .. .. .. ..
. . . . .
The results of the study, however, indicated that the average queue length just after
service was the same as at random instant. Consequently, type 2 regeneration points
can be considered redundant, and matrix 4.82 can be reduced to matrix 4.50 with
appropriate adjustments to the transition probabilities. The passing of a major stream
vehicle has no effect on minor stream queues, when no minor stream vehicles are
waiting at stop line or arrive during the block.
The probability pij of transition from i ∈ {1, 2, 3, . . . } to j ∈ {i − 1, i, i + 1, . . . } is
the probability of j − i + 1 arrivals during a service time, which is either the follow-up
time tf or a lag ϒ1 shorter than tf plus a block (Tb ):
pij = P{tf ≤ ϒ1 , A(tf ) = j − i + 1}+P{ϒ1 < tf , A(ϒ1 + Tb ) = j − i + 1} . (4.83)

The probability of k arrivals during a transition interval is
αk = P{tf ≤ ϒ1 , A(tf ) = k} + P{ϒ1 < tf , A(ϒ1 + Tb ) = k}

tf ∞
(λ2 tf )k −λ2 tf [λ2 (υ + y)]k −λ2 (υ+y)
= e−λ1 tf e + λ1 e−λ1 υ e fb (y) dy dυ
k! 0 0 k!
k ∈ {0, 1, 2, . . . } , (4.84)
where fb (y) is the probability density function of block lengths (Tb ). Tanner (1953)
has given its moment generating function, from which the expected block length can
be obtained as
eλ1 (tc −tp ) 1
E[Tb ] = − . (4.85)
λ1 (1 − λ1 tp ) λ1
If a departing vehicle leaves the system empty (i = 0), Tanner’s (1962) model (4.82)
considers the possibility of transition to state x as:
p0x = P{ϒ1 < ϒ2 , A(Tb ) = 0}

∞
λ1 (4.86)
= e−λ2 y fb (y) dy.
λ1 + λ2 0
The transition takes place, if a major-stream vehicle arrives before a minor-stream

vehicle and no minor-stream vehicles arrive during the ensuing block y (Fig. 4.8).
Departure
tf
Block Time
Transition points
Arrival
Figure 4.8: Transition 0 → x in Tanner’s (1962) model
The state 0 stays unchanged (Fig. 4.9), if one minor-stream vehicle arrives a) during the
follow-up time in antiblock, b) after follow-up time but before the next major-stream
vehicle, c) during a lag ϒ1 < tf or adjacent block, or d) during a block that has started
after the follow-up time:
p00 = P{tf ≤ ϒ1 , A(tf ) = 1} + P{tf ≤ ϒ2 < ϒ1 }

+ P{ϒ1 < tf , A(ϒ1 + Tb ) = 1} + P{tf ≤ ϒ1 < ϒ2 , A(Tb ) = 1} . (4.87)
This becomes (Tanner 1962)
λ2
p00 = e−λ1 tf λ2 tf e−λ2 tf + e−(λ1 +λ2 )tf
λ1 + λ2
tf ∞
+ λ1 e−λ1 υ λ2 (υ + y)e−λ2 (υ+y) fb (y) dy dυ
0 0
∞
λ1 −(λ1 +λ2 )tf
+ e λ2 y e−λ2 y fb (y) dy. (4.88)
λ1 + λ2 0
By collecting terms we obtain

∞
e−(λ1 +λ2 )tf −λ2 y
p00 = λ2 (1 + λ1 tf + λ2 tf ) + λ1 λ2 y e fb (y) dy
λ1 + λ 2 0
tf ∞
+ λ1 e−λ1 υ λ2 (υ + y)e−λ2 (υ+y) fb (y) dy dυ. (4.89)
0 0
a) b) c) ϒ1 d) Departure
tf
Block
Time
Arrival
Figure 4.9: Possible ways for transition 0 → 0 in Tanner’s (1962) model
The state of the system shifts from state 0 to state j ∈ {1, 2, 3, . . . }, if j + 1 vehicles
arrive a) during a follow-up time in antiblock, or b) during a lag ϒ1 < tf or adjacent
block, or c) during a block initiated by a major-stream vehicle arriving later than tf after
the previous transition point but before the next minor-stream vehicle:
p0j = P{tf ≤ ϒ1 , A(tf ) = j + 1} + P{ϒ1 < tf , A(ϒ1 + Tb ) = j + 1}

+ P{tf ≤ ϒ1 < ϒ2 , A(Tb ) = j + 1} , j ∈ {1, 2, 3, . . . } . (4.90)
This can be expressed as (Tanner 1962)

tf ∞
(λ2 tf )j +1 −(λ1 +λ2 )tf −λ1 υ [λ2 (υ + y)]j +1 −λ2 (υ+y)
p0j = e + λ1 e e fb (y) dy dυ
(j + 1)! 0 0 (j + 1)!
∞
λ1 −(λ1 +λ2 )tf (λ2 y)j +1 −λ2 y
+ e e fb (y) dy, j ∈ {1, 2, 3, . . . } . (4.91)
λ1 + λ 2 0 (j + 1)!
Transition probabilities p0j are analogous with the probability p00 in cases a, c, and d.
Case b does, however, not have an analogy, because service time is zero and it does not
allow any queue accumulation.
If the system is empty when a block ends, the system remains in state x, if a major-
stream vehicle arrives before a minor-stream vehicle and no minor-stream vehicles
arrive during the ensuing block; i.e., pxx = p0x . A state x is followed by state 0 if the
next vehicle is from the minor stream or if the next vehicle is from the major stream
and exactly one minor-stream vehicle arrives during the block (Tanner 1962):
px0 = P{ϒ2 < ϒ1 } + P{ϒ1 < ϒ2 , A(Tb ) = 1}

∞
λ2 λ1 (4.92)
= + λ2 y e−λ2 y fb (y) dy.
λ1 + λ2 λ1 + λ2 0
Similarly (Tanner 1962)
pxj = P{ϒ1 < ϒ2 , A(Tb ) = j + 1}

∞
λ1 (λ2 y)j +1 −λ2 y (4.93)
= e fb (y) dy, j ∈ {1, 2, . . . } .
λ1 + λ2 0 (j + 1)!
Now all elements in the transition probability matrix P (4.82) have been defined.
The regeneration point of type 1 in this model is the time instant that a minor stream
vehicle crosses the stop line. The follow-up time is not included in waiting time. The
average waiting time in system is obtained by adding tf to the Tanner’s delay (W̄2 ), as
suggested by Heidemann (1991). The equation is

E Tb2 λ2
+ a1 e−λ1 tf (eλ1 tf − λ1 tf − 1)
2a1 λ1
W̄S = W̄2 + tf = + tf , (4.94)
1 − λ2 a1 (1 − e−λ1 tf )
where
eλ1 (tc −tp )
a1 = E[Tb ] + λ−1
1 = . (4.95)
λ1 (1 − λ1 tp )
If the major flow is Poisson (tp = 0), the waiting time becomes
λ1 eλ1 tf (eλ1 tc − λ1 tc − 1) + λ2 eλ1 tc (eλ1 tf − λ1 tf − 1)

W̄S = W̄2 + tf = + tf . (4.96)
λ1 [λ1 eλ1 tf − λ2 eλ1 tc (eλ1 tf − 1)]
Dunne & Buckley (1972) observed that for an isolated vehicle the average waiting time
during a block (Adams’ delay) according to equation 4.94 is

E Tb2
W̄e = lim W̄2 = . (4.97)
λ2 →0 2a1
Capacity (3.24) can be expressed as

λ1 1 − λ1 tp e−λ1 (tc −tp )
Cp =
1 − e−λ1 tf (4.98)
−1
= a1 (1 − e−λ1 tf ) ,
so that the demand-to-capacity ratio of the minor stream is

λ2
ρ= = λ2 a1 (1 − e−λ1 tf ). (4.99)
Cp
Delay equation (4.94) can now be rewritten as (Dunne & Buckley 1972)
W̄e + a2 ρ
W̄2 = , (4.100)
1−ρ
where
eλ1 tf − λ1 tf − 1
a2 =
λ1 (eλ1 tf − 1)
1 tf
= −
λ1 eλ1 tf − 1
The waiting time in system can be expressed as
W̄S = W̄2 + tf

a 2 − tf ρ (4.101)
= S̄e 1 + 1 + .
S̄e 1−ρ
This has the structure of the Pollaczek-Khintchine formula (4.61), assuming that the
randomness constant is given as
a 2 − tf
κ =1+
S̄e
(4.102)
1 1 tf
=1+ −
S̄e λ1 1 − e−λ1 tf
and S̄e = W̄e + tf is the average service time of isolated vehicles (see Section 4.2.3).
Thus, the average delay equation (4.94) of Tanner (1962) can be expressed in terms of
an M/G/1 queuing system. Pöschl (1983) has demonstrated numerically that Tanner’s
model is compatible with the M/G/1 model. See also Heidemann (1991). Also when
major-stream headways follow Cowan’s (1975, 1984, 1987) M3 model (see Section
2.3.6), the delay equation can be expressed in the form of the Pollaczek-Khintchine
formula (Troutbeck 1990, Troutbeck & Walsh 1994).
The discussion above has demonstrated that the theoretical queuing models have a close
connection with gap acceptance theory. Troutbeck & Walsh (1994) have remarked that
with proper service-time distributions queuing theory gives exactly the same estimates
as gap acceptance theory. They preferred gap acceptance theory, because it is based
on documented driver behavior. However, in this report queuing theory is preferred
mainly for three reasons: i) Queuing theory allows the application of sophisticated
capacity models without leading to very complex delay models. ii) Queuing models use
capacity as a parameter, so that delay estimates are compatible with capacity estimates.
iii) Queuing models are easier to analyze under transient (time-dependent) conditions
than gap acceptance models. This is the theme of the next section.
4.4 Delays under transient conditions

4.4.1 Definition of a transient state
The models discussed above assume steady-state conditions (see section 3.1). It is
assumed that all conditions remain similar, with only random variation, during the
analysis period, and that earlier conditions (such as a peak period) do not have any
significant effect on the analysis period. However, the conditions during an analysis
period are often transient, because
• flow rates, or other conditions, are changing,
• the system cannot reach equilibrium because flow rates exceed capacity, and/or
• preceding conditions still have an effect on the performance of the system.
During a transient period the performance measures of the system are time dependent.
The parameters, such as flow rates, of the models are usually given as averages over the
analysis period. Because delay is not a linear function of flow rates, but increases more
steeply at high flow rates, the use of average parameter values under transient conditions
make the results biased. The analysis should be performed for several shorter analysis
periods, describing flow rates as a step function, and with a proper consideration of
initial queues (Kimber & Hollis 1978, Fisk & Tan 1989).
When conditions are stable, but have changed recently, it takes some time before the
effect of earlier conditions becomes negligible. Steady state conditions are obtained
only after a transient phase. For example, after a peak period, when flow rates have
decreased, it takes some time before the queues have reached equilibrium. Also during
a peak period it takes some time before the queues reach an equilibrium state. The
problem of an initial transient period is well known in simulation theory (e.g., Bratley,
Fox & Schrage 1987, Kleijnen 1987).
Morse (1958) has estimated that the “relaxation time” of a queuing √ system with one
√
exponential channel and finite storage capacity is approximately ( C − q)−2 (see
also Kleinrock 1976). As flow rates approach capacity, relaxation time increases to
infinity. If flow rates exceed capacity, queues increase constantly, and no equilibrium
is reached.
It is apparent that oversaturated systems are in a transient state. For degrees of saturation
below but near unity the relaxation times are likely to span over observation periods.
Consequently, transient analysis is called for, if conditions change during the analysis
period and/or if the D/C ratio is near or above unity.
4.4.2 Oversaturated conditions

Overflow delay
If the average arrival rate exceeds capacity, the intersection is oversaturated. The
expected number of arrivals during a time period τ is A(τ ) = qτ . The average departure
rate is equal to the capacity C, so that D(τ ) = C(τ ) = Cτ , assuming a constant capacity
during the observation period.
If the queue length in the beginning of the observation period is L(0), the queue at
the end of the observation period is L(τ ) = L(0) + A(τ ) − D(τ ).7 Because under
oversaturated conditions A(τ ) > D(τ ), the system is in a nonequilibrium state, and
the queue length increases constantly.
A fluid analogy model for unsignalized intersections was described in section 2.5.
Figure 4.10 displays a fluid-analogy model for a case where demand flow rate instant-
aneously increases above the capacity at the beginning of the observation period. The
area between A(τ ) and C(τ ) curves is the overflow delay (Gazis & Potts 1965); i.e., the
excess waiting time due to oversaturation. The fluid analogy model ignores the service
times and the stochastic component of delay (Fig. 4.11). When queues are long, the
magnitude of of discontinuities is small relative to the average values (Kleinrock 1976).
To estimate the overflow delay it is only necessary to know the initial queue L(0), arrival
flow rate q(t), capacity C, and the length of the observation period τ . A special strength
of the fluid analogy overflow model is that a similar approach can be applied to estimate
overflow delay for different kinds of facilities, such as freeways (May & Keller 1967)
and signalized intersections (Gazis & Potts 1965, Luttinen & Nevala 2002).
The overflow queue is
Lo (τ ) = L(0) + A(τ ) − C(τ ), (4.103)
assuming that the overflow condition continues over the whole time interval (cf. Catling
1977):
∀u ∈ (0, τ ] : L(0) + A(u) − C(u) > 0. (4.104)
If the overflow period starts in the beginning of the analysis period and there is no initial
queue, then Lo (τ ) = A(τ )−C(τ ). For a constant arrival rate and capacity the overflow
7 Because at oversaturated conditions server is always busy, the departure rate from the system is equal
to the rate of customers moving from the queue to the service. The number of customers in the system is
N(τ ) = L(τ ) + 1.
Number of vehicles τ)
(
,A
als
r riv
A
C (τ )
ty,
aci
Cap
Time
Figure 4.10: Fluid analogy model for overflow delay
60
50
40
Number of vehicles
30
20
10
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
Time (min)
Figure 4.11: Cumulative arrivals (thick curve) and departures (thin curve) at an oversaturated
minor approach (major flow 800 veh/h, minor flow 700 veh/h, tc = 5 s, tf = 3 s, exponential
headway distributions)
queue is Lo (τ ) = τ (q − C) = Cτ (ρ − 1). If the overflow queue at time instant τ1 is

Lo (τ1 ), the overflow queue at τ2 > τ1 is
Lo (τ2 ) = Lo (τ1 ) + A(τ2 ) − A(τ1 ) − C(τ2 ) + C(τ1 ) = Lo (τ1 ) + C(ρ − 1)(τ2 − τ1 ).

(4.105)
Under condition (4.104) the equation holds also for ρ < 1.
The cumulative overflow delay (Figure 4.12) during interval (0, τ ] is the area between
the cumulative arrival curve and the capacity curve (Gazis & Potts 1965, May & Keller
1967): τ
τ2 Cτ 2
Wo (τ ) = Lo (u) du = (q − C) = (ρ − 1). (4.106)
0 2 2
This delay includes the overflow delay accumulated until τ , that is the overflow delay
0PQ of vehicles C(τ ) and the overflow delay PQR of vehicles A(τ )−C(τ ) accumulated
up to time τ . If there is an initial overflow queue L(0) > 0, the cumulative delay is
τ
Cτ 2
Wo (τ ) = Lo (u) du = L(0)τ + (ρ − 1). (4.107)
0 2
Number of vehicles
als
r riv
A
R
A(t)
ty
aci
Cap
P
C(t)
Q
t Time
0
Figure 4.12: Overflow delay based on queue sampling
The delay includes delays suffered by vehicles in the initial overflow queue, and ex-
cludes all delays suffered after time τ (Kimber & Hollis 1979). The initial queue
captures all relevant information about the history of the system (Heidemann 2002).
Consequently, the cumulative delays based on queue sampling are additive: τ1 <
τ2 < τ3 : Wo (τ1 , τ3 ) = Wo (τ1 , τ2 ) + Wo (τ2 , τ3 ). Because the delay is based on the
measurement of queue lengths, this approach is called the queue-sampling method.
Delay estimation based on the measurement of individual vehicle delays is called the
path-trace method (Rouphail & Akçelik 1992).
The average queue-sampling overflow delay per vehicle (Fig. 4.13) is obtained by
dividing the cumulative overflow delay by the number of departures Cτ (Neuburger
1971, Hurdle 1984):
Wo (τ ) L(0) τ
W̄o (τ ) = = + (ρ − 1) . (4.108)
Cτ C 2
The method overestimates the average overflow delay, because it includes also the delay
PQR. The average deterministic waiting time in system under oversaturated conditions
is obtained as the sum of average service time and the average overflow delay (Troutbeck
2000)
1 1 L(0) τ
W̄D,S (τ ) = + W̄o (τ ) = + + (ρ − 1) . (4.109)
C C C 2
As the length of the oversaturation period increases, the curve becomes steeper, and in
the limit becomes vertical, as analysis period approaches infinity.
15
10
Average delay (min)
0
0.8 0.9 1 1.1 1.2 1.3 1.4 1.5
Degree of saturation, ρ
Figure 4.13: Average overflow delay for oversaturation periods of length 15 and 60 minutes
Peak period delay

At oversaturated conditions the queues increase constantly, and the system does not
reach equilibrium. In the extreme, after infinite time, there will be an infinite queue.
The analysis of equilibrium conditions is apparently implausible, when the system is
oversaturated. There will be available neither infinite time for queue accumulation nor
infinite space for the queues. Severe oversaturation has also an impact on traffic demand,
as some traffic redistributes into the surrounding network an avoids the oversaturated
intersection (Kimber, Marlow & Hollis 1977). Consequently, oversaturation is always
a peaked phenomenon lasting only for a limited length of time (Neuburger 1971, Yagar
1977).
The discussion above provides some tools for the analysis of oversaturation during a
peak period. In order to estimate the delay during a peak period we need to know
the maximum flow rate and the shape of the demand pattern. May & Keller (1967)
suggested triangular or trapezoid shaped demand patterns (Fig. 4.14). Kimber & Hollis
(1978) defined a low-definition (rectangular) and a high-definition (piecewise constant)
pattern.
Let us assume a rectangular demand pattern with peak flow qP > C lasting for time τP ,
after which the flow rate decreases to q < C. The time (0, τP ] is the peak period. There
is no initial queue, L(0) = 0, in the system. During the time interval (0, τP ) the overflow
queue increases at rate C(ρP − 1) and reaches its maximum Lo (τP ) = CτP (ρP − 1),
where ρP = qP /C > 1 is the D/C ratio during the peak period (Fig. 4.15). After τP
the queue length decays at rate C(ρ − 1), where ρ < 1 is the D/C ratio after the peak
Flow rate
a b c d
Capacity
Time
Figure 4.14: Triangular (a), trapezoidal (b), rectangular (c), and piecewise constant (d) demand
patterns
period (Kimber & Hollis 1978). The overflow queue is

Cτ (ρP − 1), if 0 < τ ≤ τP

Lo (τ ) = (4.110)
Lo (τP ) − C(τ − τP )(1 − ρ), if τP < τ ≤ τo
The overflow period ends at τo , when the overflow queue vanishes:
L(τo ) = Lo (τP ) − C(τo − τP )(1 − ρ) = 0. (4.111)
The length of the overflow period is determined by the length of the peak period and
the D/C ratio during and after the peak period (see Kimber & Hollis 1978):
Lo (τP ) (ρP − ρ)τP
τ o = τP + = . (4.112)
C(1 − ρ) 1−ρ
This result can also be obtained more directly by equating the number of overflow
vehicles to the capacity during time interval (0, τo ] (Rouphail & Akçelik 1992):
qP τP + q(τo − τP ) = Cτo . (4.113)
Number of vehicles
a city
als Cap
riv
Ar
τP τo Time
Figure 4.15: Overflow delay due to a rectangular peak flow pattern
The cumulative overflow delay is the area of the triangle in Figure 4.15 (cf. Kimber &
Hollis 1978):
τo
CτP2 (ρP − 1)(ρP − ρ)
Wo (τo ) = Lo (u) du = . (4.114)
0 2(1 − ρ)
Since the queue increases and decreases in a linear fashion the average queue length
during time interval (0, τo ] is half the maximum queue length of Lo (τP ) (Neuburger
1971), and the delay is
τo
Wo (τo ) = Lo (τP ). (4.115)
2
Because the delay includes the area of the total triangle, the results for both queue-
sampling and path-trace methods agree, and the bias in the average overflow delay
disappears.
The average overflow delay per vehicle is obtained by dividing the cumulative overflow
delay with the number Cτo of vehicles experiencing overflow delay:
Wo (τo ) Lo (τP ) τP
W̄o = = = (ρP − 1). (4.116)
Cτo 2C 2
The same result can also be derived using the Little’s formula (4.65), which states
that the average queue length is equal to the product of average waiting time and the
arrival rate (Little 1961). Because during the interval (0, τo ] the average queue length
is L̄o (τp )/2, and the average arrival rate to the overflow queue is equal to the departure
rate (capacity), we obtain
Lo (τP ) τP
Wo = = (ρP − 1). (4.117)
2C 2
As equation (4.108) shows, this delay is equal to the average queue-sampling delay
during time interval (0, τP ].
The average overflow delay does not depend on the D/C ratio after the peak. Because
the average overflow queue length Lo (τP )/2 does not change with ρ, the delay per
vehicle does not increase, although ρ increases. The time period of high average
delays, however, becomes longer.
Equation (4.116) gives the average overflow delay per vehicle during the overflow
period (0, τo ]. Kimber & Hollis (1978) have attributed the excess delay to the vehicles
A(τP ) = CρP τP of the peak itself:
Wo (τo ) τP (ρP − 1)(ρP − ρ)

W̄o = = . (4.118)
CρP τP 2ρP (1 − ρ)
One should be careful, if this average delay is used to estimate the cumulative overflow
delay.
Using the queue-sampling method the cumulative delays over consecutive time periods
can be summed. The calculations should consider the initial queue at the start of the
analysis period. Rouphail & Akçelik (1992) have discussed the delay estimates based
on the path-trace method.
4.4.3 Coordinate-transformation method

Oversaturation is usually a peaked phenomenon lasting only for a limited length of
time. For a peak period of limited length the average delay can not increase to infinity.
Consequently, the steady-state approach breaks down at high degrees of saturation
(Rouphail, Tarko & Li 1997).
At D/C ratios considerably above unity the queues are very large and the effects of
random variation can be ignored (Yagar 1977). The average overflow delay approaches
the deterministic overflow delay Wo presented above. On the other hand, for D/C ratios
considerably below unity the probability of overflow due to random fluctuations is very
low, and the steady-state results appear to be plausible. When the D/C ratio is near
unity the average delay is above the fluid analogy model but below the steady-state
model (see Hurdle 1984).
The stochastic nature of a peak period has been studied analytically by de Smit (1971),
Newell (1971) and others. The fluid approximation suggests some analogy with flood-
ing theory (see Prabhu 1965, Gani 1969). Haight (1963) used a transition matrix
(Markov chain) for queue length probabilities. The most commonly used approach is,
however, based on a combination of a steady-state model and a deterministic oversat-
uration model.
The average delay curve of the steady-state model approaches asymptotically the ver-
tical line ρ = 1. If it is assumed that the average time-dependent delay curve should
approach the deterministic overflow curve, the new delay curve can be obtained by
a coordinate transformation suggested by Kimber et al. (1977) and Catling (1977).
Originally the method was applied in the Transyt program (Robertson 1969) by P. D.
Whiting.
The horizontal difference between the time-dependent waiting-time curve and the de-
terministic waiting-time curve is ρD − ρ (Fig. 4.16). If the time-dependent curve
approaches asymptotically the deterministic curve in the same way as the steady-state
curve approaches asymptotically the vertical line ρ = 1, then the following relation
holds (Kimber & Hollis 1979):
1 − ρn = ρD − ρ, (4.119)
where ρn and ρD are the degrees of saturation giving the same steady-state and determ-
inistic average waiting times, respectively, as the transformed average waiting time W̄T
for D/C ratio ρ.
Average waiting time
Time-dependent
Steady-state
Deterministic
1
C
0
0 ρn 1 ρ ρD v/c ratio
Figure 4.16: Coordinate transformation for waiting times in the system
The average deterministic waiting time in system is (Kimber & Hollis 1979)
1 L(0) τ
W̄D,S = + + (ρD − 1). (4.120)
C C 2
Because this is equal to the transformed waiting time W̄T,S for D/C ratio ρ, we obtain

2 1 + L(0)
ρD = W̄T,S − + 1. (4.121)
τ C
Equation (4.119) can now be expressed as

2 1 + L(0)
W̄T,S − + ρn − ρ = 0 (4.122)
τ C
The equivalent D/C ratio (ρn ) for the steady-state model can be obtained from a steady-
state waiting-time equation. This equation should describe the waiting time in system,
including the service time.
If the steady-state waiting time is estimated using the M/G/1 model (4.62), ρn can be
expressed as
C W̄T,S − 1
ρn = . (4.123)
C W̄T,S − 1 + κ
The time dependent average waiting time in the system can be solved from the quadratic
equation

2C 2 2[L(0) + 2 − κ] 2[L(0) + 1]
W̄ − ρ − 1 + C W̄T,S +(1−κ) ρ + −1 = 0
τ T,S Cτ Cτ
(4.124)
as (Catling 1977, Kimber & Hollis 1979)
/
W̄T,S = a1 + a12 + a2 , (4.125)
where
τ L(0) + 2 − κ
a1 = (ρ − 1) +
4 2C
τ 2[L(0) + 1]
a2 = 1 − (1 − κ) ρ + .
2C Cτ
The negative branch of the solution has no physical significance (Kimber & Hollis
1979).
In an M/M/1 system with no initial queue κ = 1, L(0) = 0, and
1
ρn = 1 − . (4.126)
C W̄T,S
The time-dependent average waiting time in the system is (see also Kimber et al. 1977,
Troutbeck 2000)
0 2
τ 1 τ 1 τ
W̄T,S = (ρ − 1) + + (ρ − 1) + + . (4.127)
4 2C 4 2C 2C
It can also be obtained directly from equation (4.125) as

/
W̄T,S = a1 + a12 − a2 , (4.128)
where
τ 1
a1 = (ρ − 1) +
4 2C
τ
a2 = .
2C
Because ρ − 1 = C −1 (q − C), the average waiting time can be expressed also in terms
of capacity and reserve capacity (see Brilon 1991, FGSV 2001).
If the assumption (4.119) of equivalent differences between D/C ratios is replaced by

an assumption of equivalent proportions (Heidemann 2002)
ρ
= ρn , (4.129)
ρD
the equation for an M/M/1 system with no initial queue is

2 1 1
W̄T,S − +1 1− − ρ = 0. (4.130)
τ C C W̄T,S
The time-dependent average waiting time in the system is obtained as the larger root
of the equation above, which gives
# 1 $
1 τ 8ρ
W̄T,S = + ρ − 1 + (ρ − 1)2 + . (4.131)
C 4 Cτ
This formula was suggested by Akçelik & Troutbeck (1991), and it has been adopted
in the HCM2000 (Transportation Research Board 2000) with a slight modification.
The German method (FGSV 2001) uses a coordinate-transformation approach with
parameters based on simulation and field measurements (Brilon et al. 1994). Instead
of D/C ratio the reserve capacity is used as a parameter in the model.
As Hurdle (1984) remarks, the coordinate-transformation method is not a result of any
detailed analysis of queue behavior. The only justifications for the method is that it
provides a smooth transition from steady-state analysis to time-dependent analysis in a
way that satisfies the intuitive ideas of what ought to happen. Also, the method does not
consider the length of the overflow-queue discharge process following the peak period.
Heidemann (2002) has derived exact formulas for means and variances of queue lengths
and waiting times in M/G/1 systems under transient conditions. The means and vari-
ances were obtained by numerical methods. They were shown to depend on the shape
of the service time distribution, not only on its first and second moments. However,
the coordinate-transformation approximations presented by Kimber & Hollis (1979)
and Akçelik & Troutbeck (1991) were found to agree fairly well with the exact results.
The method of Kimber & Hollis (1979) performed slightly better than the method of
Akçelik & Troutbeck (1991).
Equation (4.131) gives the time-dependent average waiting time in the system. Fol-
lowing equation 4.8 the average time-dependent control delay can be expressed as
W̄T = W̄T,S − tf + W̄a , (4.132)
where W̄a is the average acceleration delay.

PERFORMANCE ANALYSIS METHODOLOGIES
5 PERFORMANCE ANALYSIS METHODOLOGIES

5.1 Performance measures
Capacity describes the maximum performance of a traffic facility. Capacity is, however,
not a sufficient measure for the analysis of an intersection, because operating conditions
may be unreasonably low for traffic demand below, but near capacity. The first edition
of Highway Capacity Manual (Bureau of Public Roads 1950) addressed this issue by
defining a practical capacity.1 It is “the maximum number of vehicles that can pass
a given point on a roadway or in a designated lane during one hour without the traffic
density being so great as to cause unreasonable delay, hazard, or restriction to the
drivers’ freedom to maneuver under the prevailing roadway and traffic conditions.”
The second edition (Highway Research Board 1965) developed this idea further by
introducing levels of service. Practical capacity was replaced by service volumes rep-
resenting specific volumes related to operating conditions collectively termed ‘level of
service’ (LOS). A service volume was “the maximum number of vehicles that can pass
over a given section of a lane or roadway in one direction on multilane highways (or in
both directions on a two- or three-lane highway) during a specified time period while
operating conditions are maintained corresponding to the selected or specified level of
service.” No specific analysis procedure was presented for unsignalized intersections.
The manual suggested that capacities and service volumes were estimated by the sig-
nalized intersection criteria; i.e., load factor. Signal split was to be calculated on the
basis of the relative volumes on intersecting roads and inversely on the basis of their
relative widths.
The third edition of HCM (Transportation Research Board 1985) presented an analysis
procedure based on the German methodology. Reserve capacity was used as a level-
of-service criterion. This approach is still used in Finland. In later updates of HCM
the “average total delay” was adopted as a service measure (Transportation Research
Board 1994). Total delay was defined as “the total elapsed time from when a vehicle
stops at the end of the queue until the vehicle departs from the stop line.”
The fourth edition of HCM (Transportation Research Board 2000) uses control delay as
a service measure for both signalized and unsignalized intersections (Table 5.1). While
reserve capacity is a practical engineering measure, delay is a measure experienced
directly by road users. Control delay includes initial deceleration delay, queue move-
up time, stopped delay, and final acceleration delay. The same service measure is also
used in the German manual HBS 20001 (FGSV 2001), as displayed in Table 5.1.
Table 5.1: Average control delay as a LOS criterion in HCM2000 (Transportation Research
Board 2000) and HBS 2001 (FGSV 2001)
LOS HCM 2000 HBS 2001
A ≤ 10 ≤ 10
B ≤ 15 ≤ 20
C ≤ 25 ≤ 30
D ≤ 35 ≤ 45
E ≤ 50 > 45
F > 50 v/c > 1
1 This concept is still in use. In the Australian guidelines (Austroads 1988) “practical absorption
capacity” is defined as 80 % of the theoretical “absorption capacity”. The Australian Sidra model (Akçelik
1998), however, applies the HCM level-of-service criteria.
Kyte et al. (1996) suggested both delay and queue length as LOS criteria. However,
according to the German manual (FGSV 2001) queue length cannot be considered as
a level-of-service criterion but a factor in intersection design.
The discussion below describes the analysis methodologies used in the United States
(HCM), Denmark (DanKap), Sweden (Capcal 2), and Finland (Capcal) to estimate the
capacity and average delay at unsignalized intersections and roundabouts. The German
Additive Conflict Flows methodology has been described in Section 3.10. For further
information about current German capacity analysis procedures the reader is referred
to HBS 2001 (FGSV 2001). It is assumed that the reader is familiar with the theory
presented above.
5.2 U.S. methodology

HCM2000 (Transportation Research Board 2000) Chapter 17 describes analysis
methods for two-way stop-controlled (TWSC) intersections, all-way stop-controlled
(AWSC) intersections and roundabouts. The manual suggests that with appropriate
changes in key parameters an analyst can apply the TWSC method to yield-controlled
intersections.
The analysis procedures are extensively described in the manual. Background informa-
tion is presented in Contractor’s Final Report (Kyte et al. 1996). The roundabout meth-
odology follows the suggestions of Troutbeck (1997b). The outline of the HCM2000
method is displayed in Figure 5.1.
Collect input
Identify streams Estimate tc,i and tf,i
Compute potential capacity
Compute movement capacity
Compute queue lengths
Compute control delays
Determine levels of service
Figure 5.1: HCM2000 analysis methodology for unsignalized intersections (Transportation

Research Board 2000)
HCM2000 assumes Poisson arrivals and constant critical gaps and follow-up times. It
allows an analysis of short peak periods. Because the theoretical basis of HCM2000
has been extensively discussed above, only a short overview will be given here.
Traffic streams at an unsignalized intersection are classified following Figures 2.2 and
2.3. Pedestrian flows along the major road have Rank 1 and those across the major
road have Rank 2. Conflicting flow rates are the flow rates of conflicting streams, with
a some exceptions:
• If a minor stream merging to an N -lane road and is in conflict with a major stream
i in one lane only, the conflicting flow rate is qi /N .
• One half of the right-turn movement exiting before conflict area is included in
major stream
• The effect of left-turn major-stream flows is twice their number
• For a minor-road left-turn movement only half of the opposing through and right-
turn flow rates are included as conflicting flow rate.
At roundabouts the vehicles passing in front of the entering vehicle are counted as the
major flow. At most well-designed roundabouts the exiting traffic can be ignored.
Base values for critical gaps and follow-up times are displayed in Table 5.2. Adjust-
ments for critical gap are made to account for the presence of heavy vehicles, approach
grade, T-intersections, and two-stage gap acceptance. The follow-up time is adjusted
for the presence of heavy vehicles.
Table 5.2: Base values for critical gaps and follow-up times for an unsignalized intersection
with a two or four-lane major road in HCM2000 (Transportation Research Board 2000)
Critical gap (s) Follow-up
Movement 2-lane 4-lane time (s)
Major left 4.1 4.1 2.2
Minor right 6.2 6.9 3.3
Minor through 6.5 6.5 4.0
Minor left 7.1 7.5 3.5
For one-lane roundabouts HCM2000 uses the critical gaps and follow-up times sugges-
ted by Troutbeck (1997b), as displayed in Table 5.3. An upper bound and a lower bound
for the capacity is estimated. The analysis procedure can be applied if the circulating
flow does not exceed 1,200 veh/h. No analysis procedure is presented for multi-lane
roundabouts.
Table 5.3: Critical gaps and follow-up times for roundabouts in HCM2000 (Transportation
Research Board 2000)
Critical gap (s) Follow-up time (s)
Upper bound 4.1 2.6
Lower bound 4.6 3.1
The potential capacity (3.12) is calculated assuming a Poisson arrival process on major
streams. The impedance factor of a Rank 2 stream i is the probability p0,i = 1 − ρi
that the system is empty. See discussion in Section 3.5.5. The impedance factor for
multiple conflicting Rank 2 streams is obtained by multiplication. For Rank 4 streams
the impedance factor to account for conflicting Rank 2 and Rank 3 queues is obtained
using the adjustment (3.103) of Brilon & Großmann (1991).
The effect of a pedestrian stream j is estimated by a pedestrian blockage factor
qg tpb
ρj = , (5.1)
3600
where qg is the number of pedestrian groups per hour, and tpb is the average duration
of time the lane is blocked by a pedestrian group (see Section 3.8). The pedestrian
impedance factor is obtained as p0,j = 1 − ρj . This factor is combined with other
impedance factors (p0,i ) by multiplication.
The estimation of shared lane capacity in HCM2000 is described in Section 3.7. The
capacity estimation of a flared approach is described at the end of this section.
Although HCM2000 suggests the use of simulation models and field data to assess the
effects of upstream signals, the manual presents a rather detailed procedure for the task.
As platoons from a signalized intersections arrive at an unsignalized intersection, four
flow regimes result:
1. No platoons
2. Platoon from the left only
3. Platoon from the right only
4. Platoons from both directions.
A minor stream is considered to be blocked if a platoon is traveling in a conflicting

stream. During the unblocked period the major stream is assumed to be random with
flow rate 
 qM − sM (1 − pM,u ) , if q > s (1 − p )
M M M,u
qM,u = pM,u (5.2)

0, otherwise,
where qM is the total flow rate of the major stream, sM is the saturation flow rate of the
major stream, and pu,j is the proportion of time that the minor stream is unblocked by
major stream platoons. The capacity of the minor stream j is
Cj = pu,j Cu,j , (5.3)
where Cu,j is the capacity assuming a continuous random flow rate qM,u in the major
stream.
If a wide median allows minor stream vehicles to cross the major road in two phases, the
capacity is analyzed in two stages. The two-stage gap acceptance analysis procedure
in HCM2000 follows the method developed by Brilon et al. (1996) and Brilon & Wu
(1999). See also Luttinen (1998).
The 95th-percentile queue length of a minor stream j is estimated as
 2 
3
3 2
3600
×
qj
q 3
 j 4 qj Cj Cj  Cj
L95,j = 900τ  −1+ −1 + 
 Cj Cj 150τ  3600
(5.4)
# 0 $
τ Cj 2 24ρj
= ρj − 1 + ρj − 1 + ,
4 τ Cj
where τ is the length of the analysis period. The queue length can also be expressed in
terms of reserve capacity (κj = Cj − qj ) as
# 1 $
τ 24qj
L95,j = −κj + κj +2
. (5.5)
4 τ
The service measure of unsignalized intersections in HCM2000 is the average control

delay. The definition of control delay in HCM2000 has been discussed in section 4.1.
The formula (4.131) suggested by Akçelik & Troutbeck (1991) is used, but with an
additional value of five seconds to “account for the deceleration of vehicles from free-
flow speed to the speed of vehicles in queue and the acceleration of vehicles from the
stop line to free-flow speed.”2 The delay equation is
# 0 $
3600 8ρ j
W̄j = + 900τ ρj − 1 + (ρj − 1)2 + + 5. (5.6)
Cj τ Cj
The level-of-service criteria are presented in Table 5.1.

HCM2000 considers a flared minor-road approach as an approach with an additional
short lane. The traffic in the flared lane consists of a right turn movement having flow
rate qR and a left+through movement having flow rate qLT . The capacity of a flared
lane is estimated by interpolation between the capacity of two separate lanes and the
capacity CSH of a shared lane.
The capacity of a shared lane is calculated as described above. The capacity of a
separate-lane condition is calculated using equation3

qLT qR
CS = min CR 1 + , CLT 1 + , (5.7)
qR qLT
where · is the floor function (Spanier & Oldham 1987). For a movement k ∈ {R, LT}
considered to operate in a separate lane the average queue length is estimated using the
Little’s formula (4.65) as
W̄k qk
L̄k = , (5.8)
3600
where W̄k is the average control delay of movement k assuming a separate lane. The
required storage space (vehicles) for the approach to operate effectively as separate
lanes is estimated as
Lr = max L̄k + 1, (5.9)
k∈{R,LT}
where · is the nearest integer (rounding) function (Spanier & Oldham 1987). The
capacity of a flared lane is obtained by interpolation between the shared lane capacity
(CSH ) and the capacity (CS ) assuming two separate lanes:

(C − C ) Ls + C , if L < L
S SH SH s r
C= Lr (5.10)

CS , if Ls ≥ Lr ,
2 This description indicates a possible overlap in control delay and geometric delay. See discussion in
Section 4.1. HCM2000 does not subtract the follow-up time (see equation (4.132).
3 This equation and some other details of the method are described in a correction sheet (TRB Committee
A3A10, Highway Capacity and Quality of Service 2003). It contains also some corrections for the
upstream-signal effects procedure.
where Ls is the actual storage capacity of the flare.

If the required storage does not exceed the actual storage, capacity (5.10) and delay are
calculated assuming separate lanes. The average control delay is calculated as
qLT W̄LT + qR W̄R

W̄ = , (Ls ≥ Lr ) (5.11)
qLT + qR
where W̄LT and W̄R are the average control delays of left+through and right-turn move-
ments, respectively, assuming separate lanes. The probability or frequency of storage
overflow are not considered. If the storage is less than required, the average control
delay is calculated using the capacity estimate in Equation 5.10 for Ls < Lr .
5.3 Danish methodology

DanKap (Vejdirektoratet 1999b, Vejdirektoratet 1999a) estimates potential capacity us-
ing the gap acceptance model (3.12) which assumes Poisson arrivals, as does HCM2000
(Transportation Research Board 2000). The impedance factors and the effect of shared
lanes are calculated using the same methods as in HCM2000. Traffic volume and
capacity are given in terms of passenger cars per analysis period, not as pc/h.
Separate critical gaps (see Tables 5.4 and 5.5) are defined for bicycles and other unre-
gistered vehicles. The critical gap used in the calculations is the weighted average of
critical gaps for bicycles and motorized vehicles. The follow-up time is 3.0 s for all
movements. No distinction is made even between yield and stop controlled approaches.
Table 5.4: Base values for critical gaps and follow-up times for an unsignalized intersection
with a two or four-lane major road in DanKap (Vejdirektoratet 1999b)
Passenger cars Bicycles time
Yield control Stop control and other
Movement 2-lane 4-lane 2-lane 4-lane unregistered (s)
Major right — — — — 2.5 3.0
Major left 5.5 6.0 — — 2.5 3.0
Minor right 5.5 5.5 6.5 6.5 2.5 3.0
Minor through 6.0 7.0 7.0 8.0 2.5 3.0
Minor left 7.0 8.0 8.0 9.0 2.5 3.0
Table 5.5: Base values for critical gaps and follow-up times for a roundabout in DanKap
(Vejdirektoratet 1999a)
Entry lanes PCs Bicycles time (s)
1 4.5 2.5 2.8
2 4.0 — 2.6
Traffic streams are classified following Figures 2.2 and 2.3. Conflicting flow rates are the
flow rates of conflicting streams. Major stream right-turn movements exiting before the
conflict area can be multiplied with a factor between zero and one. It is recommended
that these right-turn movements are ignored (Aagaard 1998). Conflicting bicycle flows
along the major road (movements 2 and 5) are included in the analysis. Pedestrian
flows are ignored. The passenger-car equivalents used to express vehicle counts as
equivalent number of passenger cars are displayed in Table 5.6.
Table 5.6: Passenger-car equivalents in DanKap (Vejdirektoratet 1999a)

Gradient Vehicle category
(% ) Motorcycles etc. PCs and vans Trucks and buses Semi-trailers etc.
+4 0.7 1.4 3.0 6.0
+2 0.6 1.2 2.0 3.0
0 0.5 1.0 1.5 2.0
-2 0.4 0.9 1.2 1.5
-4 0.3 0.8 1.0 1.2
The capacity of a roundabout entry is adjusted for pedestrians (fp ) and exiting traffic
(fe ) effects as
C = fp fe Cp . (5.12)
If the traffic flow exiting before the conflict area is less than 400 pc/τ the adjustment
factor is fe = 1. Otherwise it is fe = 0.9. These values are based on measurements at
only one roundabout, and therefore are very uncertain (Aagaard 1998).
In a one-lane entry of a roundabout the adjustment fp for pedestrians depends both
on the crossing pedestrian volume (ped/τ ) and the circulating volume (pcu/τ ) of both
motorized vehicles and bicycles (Table 5.7). If no pedestrians cross the entry, fp = 1.
These adjustments follow the results of German empirical research (Brilon, Wu &
Bondzio 1997). For a two-lane entry the adjustment is fp = 1.0.
Table 5.7: Adjustment factor fp for crossing pedestrian flows at a one-lane entry to a roundabout
according to DanKap (Vejdirektoratet 1999b)
Motorized veh + Pedestrians (ped/τ )
bicycle flow (pcu/τ ) 100 200 300 400
0 0.99 0.93 0.87 0.81
100 0.99 0.93 0.87 0.82
200 0.99 0.94 0.88 0.83
300 0.99 0.94 0.89 0.84
400 0.99 0.95 0.90 0.86
500 0.99 0.95 0.91 0.88
600 0.99 0.96 0.93 0.90
700 0.99 0.97 0.95 0.93
800 0.99 0.98 0.97 0.96
900 0.99 1.00 1.00 1.00
1000 1.00 1.00 1.00 1.00
The time-dependent delay formula (4.131) suggested by Akçelik & Troutbeck (1991)
is used in DanKap. But unlike in HCM2000, no additional component for deceleration
and acceleration delay is used. DanKap does not present any level-of-service criteria.
The 95 and 99-percentile queue lengths for lane j are evaluated recursively from equa-
tion (Aagaard 1998)

2La 100 − a 1/(La +1)
ρ̄j = + , (5.13)
τ Cj 100
where La is the queue length exceeded a percent of time in the analysis period (τ ), ρ̄j is
the average D/C ratio during the analysis period, and Cj is the capacity. This equation
stems from the research of Wu (1994).
5.4 Swedish methodology
5.4.1 Analysis procedure
The Swedish capacity analysis software Capcal implements the procedures of the capa-
city manual (Statens vägverk 1977). Capcal 2 procedures for unsignalized intersections
(SNRA 1995b) and roundabouts (SNRA 1995c) have been slightly modified from the
previous version. Capcal 3 is a MS-Windows version of the software with some en-
hancements. The software version referred to in this report is Capcal 2.10.
Figure 5.2 displays the main steps in Capcal 2 methodology. Mean service times are
used to estimate degrees of saturation, which are the major parameters in the estimation
of MOEs, such as capacity, delay and queue lengths. Roundabouts are treated as a series
of T-intersections. In TV131 roundabouts were analyzed as short merge areas.
Determine major flow
Estimate critical gap
Calculate mean service time
Adjust mean service time
Adjust for short lanes
Calculate MOEs
Figure 5.2: Capcal 2 analysis methodology for unsignalized intersections (SNRA 1995b)
The performance measures provided by the method are: capacity, delay, queue length,
and the proportion of stopping vehicles. The Swedish method does not present any
level-of-service criteria.
Although Capcal 2 is used in Finland, the Finnish guidelines (Tiensuunnittelutoimisto
1978) and software (Tiensuunnittelutoimisto 1987, Kehittämiskeskus 1991) still follow
the Swedish capacity manual TV131 (Statens vägverk 1977). Consequently, the dif-
ferences between Capcal 2 and the old method will be described. The new roundabout
capacity analysis methodology developed by Hagring (1998b) has been described in
section 3.4.
5.4.2 Major flows

For each minor stream the conflicting higher priority flows a determined. Major flow is
the sum of these higher priority flows. In crossing conflicts the whole crossing flow is
added to the major flow. In a merging conflict the flow rate of the merging major flow
rate is divided by the number of lanes in the exit, where the merging occurs. Major-
road right-turn vehicles are ignored if they do not have conflict with the minor stream.
Bicycles and pedestrians having a crossing conflict with the subject movement have an
equivalency value of 0.5. Bicycles having a merging conflict are ignored.
The major flow is adjusted for the possible effect of a wide median by an additive
adjustment
fM = −0.4 min {qR , cqL } , (5.14)
where qR is the major-road through flow rate from the right and qL is the major-road
through flow rate from the left. For a left-turn minor movement onto a two-lane exit
c = 0.5, otherwise c = 1. The effect of major-road left-turn movements is ignored.
This is a simplification of the method proposed by Tanner (1951).
Capcal 2 has a unique way of defining major flows at a roundabout. The vehicular
interactions can be described as three different cases (Hagring 1996b, Hagring 1998b):
1. Weaving occurs when the circulating flow leaves the roundabout at an exit up-
stream to the exit of the minor flow. The entire circulating flow is included in the
major flow.
2. Merging occurs when the circulating flow and the minor flow leave the roundabout
at the same exit. Only part of the circulating flow is included in the major flow,
based on the roundabout geometry.
3. No interaction occurs when the circulating flow leaves the roundabout down-
stream to the exit of the minor flow. The circulating flow is not included in the
major flow. It is possible that a minor flow has no major flow at all, despite there
being circulating vehicles.
These interactions are summarized in table 5.8. These principles have been derived
from literature and traffic engineering judgement (Bergh 1991).
Table 5.8: Interactions between major and minor stream at a roundabout according to Capcal 2
(Hagring 1996b)
Minor Major stream
stream Left side approach Opposite approach
Through Left turn Left turn
Right turn Merging No interaction Merging
Through Weaving Merging Weaving
Left turn Weaving Weaving Weaving
The new Swedish roundabout capacity analysis procedure developed by Hagring

(1998b) uses the conventional way of defining major flow: it includes the circulating
vehicles up to the entry point considered. At a two-lane roundabout also the vehicles
circulating in the inner lane are included in the major flow of the minor stream vehicles
in the curb lane. Different kinds of interactions are, however, described by different
(lane-specific) critical gaps.
5.4.3 Critical gaps

The base values for critical gaps (tc ) in Capcal 2 are displayed Table 5.9. Table 5.10
is from the old manual. Critical gaps increase with the level of maneuvering difficulty
and higher major road speed limits. Stop controlled approaches have larger critical
gaps than yield controlled. Follow-up times tf are 0.6tc (Nordqvist et al. 1973).
Table 5.9: Base values for critical gaps in Capcal 2 (SNRA 1995b)
Major road speed limit
50 km/h 70 km/h 90 km/h
Movement Yield Stop Yield Stop Yield Stop
Major left 4.8 4.8 5.7 5.7 6.7 6.7
Minor right 5.0 5.7 5.9 6.6 6.9 7.5
Minor through 5.1 5.8 6.0 6.7 7.0 7.6
Minor left 5.3 6.0 6.2 6.9 7.2 7.8
Table 5.10: Base values for critical gaps in TV131 (Statens vägverk 1977)
Major road speed limit
50 km/h 70 km/h 90 km/h
Movement Yield Stop Yield Stop Stop
Major left 5.0 — 5.8 — 6.5
Minor right 4.8 5.5 6.0 6.5 7.2
Minor through 5.2 5.8 6.0 6.5 7.0
Minor left 5.3 6.0 6.2 6.8 7.5
The critical gaps in Table 5.9 assume 10 % of heavy vehicles in the subject movement.
For other percentages the additive adjustment is fHV = pHV − 0.1, where pHV is the
proportion of heavy vehicles. That is, heavy vehicles are assigned one second longer
critical gaps than passenger cars (Hagring 1997). In the old manual the adjustment was
fHV = 2(pHV − 0.1); i.e., heavy vehicles were assigned two seconds larger critical
gaps.
If the major road has more than two lanes, critical gaps are increased, but if the inter-
section allows two-stage gap acceptance, critical gaps are reduced (Table 5.11). For
right-turn movements critical gaps are adjusted for curb radius r (meters) and angle α
(degrees):
r − 12 α − 90
fr = 1 − 1 + 1− . (5.15)
18 120
If the major road is one way, the critical gaps of minor road through and left-turn
movements are reduced by 0.5 s.
Table 5.11: Critical gap adjustment for wide median and number of major-road lanes (SNRA
1995b)
Two-stage Lanes
gap acceptance 2 3–4 > 4
No 0 0.3 0.6
Yes -0.5 0 0.3
TV131 had also an adjustment for the size of urban area or rural conditions (lower
critical gap in densely populated areas), but this correction has been omitted in Capcal 2.
The critical gaps for 90 km/h roads was altered as a correction for rural conditions
(SNRA 1995b). The old version of Capcal adjusted the critical gap for gradients, but
in Capcal 2 the adjustment is made to lane capacity.
For roundabouts specific critical gaps are defined for right-turn movements tc,r and for
crossing movements tc,c based on the width/length ratio of the weaving section:
ww w2
tc,r = 3.06 + 1.1 + 2.375 2w (5.16)
lw lw
ww w2
tc,c = 3.06 + 2.6 + 2.75 2w , (5.17)
lw lw
where ww is the width and lw is the length of the weaving section (SNRA 1995c).
Additive adjustments are applied to the base values.
Hagring (1996b) has found this method to give too short critical gaps. His equation for
critical gaps at a roundabout is
tc = 3.91 − 0.0278lw + 0.121ww + 0.592(i − 1), (5.18)
where i is the entry lane (1 for right lane, 2 for left lane). The values are about 0.5
seconds greater than the critical gaps estimated by Capcal. Later Hagring (1998b)
estimated critical gaps for two-lane roundabouts by making distinction between both
entry lanes and circulating lanes (Table 5.12). The follow-up time is tf = 2.4 s on both
entry lanes. See also Hagring (2000).
Table 5.12: Average critical gaps in two-lane roundabouts according to Hagring (1998b)
Entry Circulating lane
lane Outer Inner
Right 4.273 3.998
Left 4.615 4.403
5.4.4 Average service times

The most important measure in the Swedish methodology is the average service time
(x). Under queuing conditions the average service time is the reciprocal of capacity; i.e.,
S̄q = C −1 . Service times S̄q can be measured under moderate queuing conditions, while
the measurement of capacity requires long periods of continuous queues (Nordqvist
et al. 1973). This point has been later emphasized by Akçelik (1998).
The Swedish capacity manual (Statens vägverk 1977) recognizes three major stream
headway distributions (Bergh 1991):
1. Tanner’s M/D/1 headway model (2.44) is used when the major flow occupies
one lane only. This model is used as a major-stream headway model for major
road left-turn and minor road right-turn movements. The minimum headway is
tp = 1.8 s.
2. Exponential major-stream headway model is used for minor streams having con-
flicts with major streams on two lanes; i.e., minor-road through and left-turn
movements on a two-lane major road, and major-road left-turn movements on a
multi-lane road. In addition, the exponential headway model is used for minor
road right-turn movements to a multi-lane road.
3. Pearson type III (gamma) distribution (see Luttinen 1991, Luttinen 1996) is used
when traffic signals are within 300 m on the major road. It is also used for minor-
road through and left-turn movements, when the major road has four or more
lanes. The application of this model was influenced by a paper by Voigt (1973).
Since, eventually, no empirical data was found to support the Pearson III distribution,
Capcal 2 has replaced it with the exponential distribution (Anveden 1988, Bergh 1991).
Accordingly, Tanner’s distribution is used for major-road left turns and minor road right
turns on two-lane major roads. In all other cases the exponential headway distribution is
assumed. At roundabouts the headway distribution is a mixture of random (exponential)
headways and a platoon model leading to the following expression of average service
time for queuing vehicles (SNRA 1995c):
S̄q = (1 − pq )S̄1 + pq S̄2 (5.19)
where
1 − e−λtf
S̄1 = −λtc
(5.20)
λe
λ(tc −tf )

e − e0.6λ(tc −tf ) 1 − e−0.542λtc
S̄2 = (5.21)
0.146λ
pq = 0.29λe 0.25tc
. (5.22)
For vehicles departing from a queue the average service time is the reciprocal of the
capacity. When major-stream headways follow the Tanner’s M/D/1 model, the average
service time under queuing conditions is (SNRA 1995b)
1 − e−λtf
S̄q = , (5.23)
λ 1 − λtp e−λ(tc −tp )
which is the inverse of the corresponding capacity equation (3.24). For exponential
major-stream headways the average service time under queuing conditions is (SNRA
1995b)
1 − e−λtf
S̄q = , (5.24)
λe−λtc
which is the inverse of capacity equation (3.12).
For an isolated vehicle, arriving to stop line after the follow-up time has passed since
the last departure, the average service time S̄e is assumed to be either the follow-up
time (tf ) or the average waiting time (W̄e ) of an isolated vehicle (4.33), whichever is
larger (SNRA 1995b):
S̄e = max {tf , We } . (5.25)
Because waiting time (4.33) is used in stead of service time (4.35), it is necessary to
set follow-up time as a lower limit for service time in order to avoid capacity estimates
larger than tf−1 .
The average service time is defined as
S̄ = ρ S̄q + (1 − ρ)S̄e . (5.26)
Because the D/C ratio (ρ) is a function of the average service time, an iterative procedure
is required. The following estimate is used as an initial value
q S̄e
ρ̃ = λS̄e = . (5.27)
3600
The variance ratio of service times is estimated as
S̄q + S̄e
CS2 = . (5.28)
2S̄q
If the major road has wide shoulders, the interaction with minor-road right-turn vehicles
can be viewed as a merging procedure. The maximum merging rate of minor stream
vehicles is (1500 − qM )+ , where qM is the conflicting major flow rate. If S̄q or S̄e are
larger than the average service time of the merge, they are replaced with the latter.
The impedance effect of higher priority queues is estimated by multiplying the service
time with an adjustment factor (Hagring 1997)
1
fq = " m , (5.29)
j =1 (1 − ρj )
where ρj is the D/C ratio for secondary major stream j , and m is the number of
secondary higher priority movements, which have to yield to streams of higher rank.
5.4.5 Capacity
Capacity is defined as the stationary outflow from a lane having a continuous queue,
assuming that the flow rates of all other traffic streams are given (Nordqvist et al. 1973,
Anveden 1988). The capacity of a lane k sharing movements i ∈ I is calculated as
fC
Ck = qi , (5.30)
S̄q,i
i∈I
qk
where fC is the correction factor for lane capacity, S̄q,i is the service time under queuing

conditions of movement i, qi is the flow rate per movement i in lane k, and qk = i∈I qi
is the total flow rate in lane k. The correction factor fC includes corrections for bicycles,
lane widths, and gradients.
5.4.6 Delay
The components of delay in the Swedish methods are defined as (Nordqvist et al. 1973,
Anveden 1988)
• Total delay (Wt ) is the additional time consumption per vehicle above the time
consumption of traffic entering the intersection at the same speed as upstream of
the intersection and exiting at the same speed as traffic flows downstream of the
intersection.
• Geometric delay (Wg ) is the hypothetical total delay if there were no interactions
between different vehicle movements and no control devices (stop signs) causing
delay.
• Interaction delay (W ) is the delay caused by control devices and interactions

between different vehicle movements; i.e., W = Wt − Wg .
Total delay has a physical interpretation. Geometric and interaction delays do, however,
not correspond to any two different parts in a time-space trajectory of a vehicle (Anveden
1988).
Geometric delay has a deceleration and an acceleration component

(va − vn )2 (ve − vn )2
W̄g = W̄gd + W̄ga = + , (5.31)
2ad va 2aa ve
where va is approach speed, ve is exit speed, vn is the negotiation speed, and ad and
aa are deceleration and acceleration constants, respectively. Total delay is the sum of
interaction delay and geometric delay. However, it is assumed that at least some part of
the deceleration component in the geometric delay should be ignored, because decel-
eration to some extent takes place when a vehicle catches up a queue. For simplicity,
total delay is calculated as (SNRA 1995b)

W̄t = max W̄ , W̄gd + W̄ga . (5.32)
This indicates a control-oriented definition of the interaction delay. In the old manual
all acceleration delays were included in the geometric delay (Anveden 1988).
Total interaction delay is calculated as the sum of the average waiting time in queue
and the average service time
W̄ = W̄q + S̄. (5.33)
The follow-up time is not subtracted, because it may not be included in the service time
S̄e (5.25) of isolated vehicles. Acceleration delay due to control is not included.
Under degrees of saturation less than 0.8 the waiting time in queue is calculated using
the Pollaczek-Khintchine formula (4.61) with variance ratio given by equation 5.28.
For a major-stream through+left lane ρ in the numerator is replaced by the D/C ratio
of the left-turn movement only, while no change is done in the denominator. When
ρ > 1.4, the fluid-analogy model (4.116) is applied. For 0.8 ≤ ρ ≤ 1.4 the average
waiting time in queue is interpolated between the two cases.
In TV131 the P-K formula (4.61) was used for ρ ≤ 0.9. The method could not be
applied when the D/C ratio was ρ > 0.9. The variance ratio was estimated as a
function of ρ and calibrated by simulation and field measurements. As a result, the
estimate for the average waiting time was a function of ρ only (Hanson 1978).
5.4.7 Short lanes

A short lane can be blocked by a queue on the adjacent lane (a) extending beyond the
short lane. A long lane can be blocked by a queue extending beyond the storage space
Ls of an adjacent short lane. If a lane is blocked a new lane allocation is done by an
iterative procedure.
In an analysis procedure for short lanes Capcal 2 calculates the queue length for the
lane (m) having the higher D/C ratio (ρm = max {ρa , ρs }) of the two lanes. The
Pollaczek-Khintchine formula for average queue length in an M/G/1 system

1 + CS2 ρm
L̄m = (5.34)
2(1 − ρm )
is used for degrees of saturation ρ ≤ 0.8. At higher degrees of saturation the queue
length is estimated by an application of the Little’s formula (4.65)
L̄m = λm W̄m . (5.35)
In the old Swedish manual (Statens vägverk 1977) the probability of storage overflow
was calculated as Ls
L̄m
P Ls < L̄m = . (5.36)
L̄m + 1
In Capcal 2 the short lane is treated as an ordinary lane, if L̄m ≤ Ls . The probability
or frequency of blocking are not considered. If a lane is blocked, the average waiting
time in queue is taken to be the same as for the adjacent lane, which is considered to
be a slight overestimation. The utilization factor of the short lane is calculated as
Ls Ls
λs λa
ρs = 1 − − . (5.37)
λa + λs λa + λs
5.5 Finnish methodology

The Finnish guidelines (Tiensuunnittelutoimisto 1978) and the Finnish Capcal software
(Tiensuunnittelutoimisto 1987, Kehittämiskeskus 1991) follow the Swedish capacity
manual TV131 (Statens vägverk 1977), which was described above. The differences
between the Finnish method and the TV131 method are described below.
The Swedish policy favors speed limits 50, 70, 90, and 110 km/h, while in Finland 50,
60, 80, 100, and 120 km/h are most commonly used. The highest speed limit (120 km/h)
is posted on freeways only. To reflect this difference in speed limits the critical gap table
(5.13) has been adjusted for Finnish conditions. The adjustment factors for critical gap
and the follow-up time tf = 0.6tc follow the Swedish method.
Table 5.13: Base values for critical gaps in the Finnish guidelines (Tiensuunnittelutoimisto
1978)
Major road speed limit (km/h)
50 km/h 60 km/h 80 km/h 100 km/h
Movement Yield Stop Yield Stop Yield Stop Yield Stop
Major left 5.0 — 5.8 — 6.2 — 6.8 —
Minor right 4.8 5.5 5.4 6.0 6.4 6.9 7.0 7.5
Minor through 5.6 6.4 6.0 6.8 6.7 7.4 7.2 7.8
Minor left 5.9 6.7 6.4 7.2 7.3 8.1 8.0 8.8
The estimation of capacity, delay and other performance measures follows the Swedish
TV131 (Statens vägverk 1977) methodology. The level-of-service criteria (Table 5.14)
follow the criteria in HCM 1985 (Transportation Research Board 1985). The service
measure is “reserve capacity”, which is the inverse of average waiting time in system
in a M/M/1 queuing model. See section 4.3.2 for details.
Level of service can also be determined from Figure 5.3, which is based on Table 5.14
slightly modified. The D/C ratio (Table 5.15) is used as an alternative service measure.
Table 5.14: Level-of-service criteria in Finland (Pursula & Ristikartano 1988, Tie- ja
liikennetekniikka 2001)
Level of service Reserve capacity (veh/h)
A 400–
B 300–399
C 200–299
D 100–199
E 0– 99
F q>C
Based on field measurements Uusiheimala (1995) suggested new ranges for critical
gaps (Table 5.16). See also Niittymäki & Uusiheimala (1996). In particular, the
critical gaps for major road left turns and minor road right turns appeared to need some
Figure 5.3: Level of service according to the Finnish guidelines (Tie- ja liikennetekniikka 2001)
Table 5.15: Alternative quality-of-service criteria in Finland (Tie- ja liikennetekniikka 2001)

Quality of service D/C ratio
Good 0.00–0.50
Fair 0.50–0.70
Tolerable 0.70–0.85
Bad 0.85–1.00
Very bad 1.00–
adjustment (Pursula & Peltola 1998). The results were used to recommend new critical
gaps to be implemented in the Hutsim simulation software (Table 5.17). Some of these
adjustments were very significant, as a comparison of Tables 5.13 and 5.17 reveals.
Table 5.16: New ranges for critical gaps suggested by Uusiheimala (1995)
50 60 80 100
Crirical gap (s) 3–5.5 4–6 6–8 7–8
Figure 5.4 displays a chart for a rough assessment of a roundabout according to the
old Finnish guidelines (Tiehallitus 1992). The upper edge of the gray area can be
interpreted as the practical capacity of a one-lane roundabout. Pedestrian and bicycle
flows crossing an entry are added to the circulating flow. If the distance between an
exit and an entry is less than 15 meters, 30 % of exiting flow is added to the circulating
flow. The pce factor for trucks and vans is 1.5 and for semi-trailers 2.0.
These capacity estimates were found to be too small at low circulating flow rates
Table 5.17: Recommended critical gaps for simulation of unsignalized intersections (Pursula
et al. 1997)
50 km/h 60 km/h 80 km/h 100 km/h
Movement Yield Stop Yield Stop Yield Stop Yield Stop
Major left 4.0 — 4.5 — 5.5 — 6.0 —
Minor right 4.0 5.0 4.5 5.5 6.5 7.0 7.0 7.5
Minor through 5.0 6.0 5.5 6.5 7.0 7.0 7.0 7.5
Minor left 5.5 6.3 6.0 6.8 7.0 8.0 8.0 8.8
Figure 5.4: Operation of a roundabout according to the old Finnish guidelines (Tiehallitus 1992)
(Liimatainen 1997). The new guidelines (Tie- ja liikennetekniikka 2001) have a slightly
corrected chart (Figure 5.5). For a more detailed capacity estimation it is suggested
that a roundabout is analyzed as a series of T-intersections with critical gap tc = 4.8 s
(Liimatainen 1997). The value is slightly higher that the lower bound in HCM2000
(Table 5.3). Hagring & Niittymäki (1997) have demonstrated that in Denmark, Fin-
land, and Sweden there is a general trend of diminishing critical gaps with increasing
roundabout size.
The guidelines do not give specific capacity estimates. The empirical studies
(Jormalainen 1991, Kehittämiskeskus 1996) were not able to find congested condi-
tions at roundabouts under study. The results, however, suggested that the Finnish
version of Capcal (TV131 method) overestimated the capacity of roundabouts. The
reason for this was probably in the estimation of critical gaps (Kehittämiskeskus 1996).
This was also suggested by the Swedish studies of Hagring (1996b), as discussed in sec-
tion 5.4.3. The new Swedish capacity model developed by Hagring has given capacity
estimates very close to Finnish simulation results (Hagring & Niittymäki 1997, Pursula
et al. 1997).
Figure 5.5: Operation of a roundabout according to the current Finnish guidelines (Tie- ja
liikennetekniikka 2001)
SIMULATION STUDIES
6 SIMULATION STUDIES
6.1 Hutsim as a simulation tool
A Finnish object-oriented simulation software, Hutsim, was used in simulation experi-
ments. It has three components: Hutedi, Hutsim, and Hutsim-analyzer. Hutedi is a
simulation model editor. Hutsim is a microscopic traffic simulator. Hutsim-Analyzer
is a postprocessor.
Hutsim has been under development at the Helsinki University of Technology (HUT)
since 1989. The stochastic properties of traffic flow and the conflicts with other traffic
flows can be described and analyzed with sufficient accuracy. Hutsim was originally
developed for the analysis of signalized intersections, but it has been extended for the
analysis of unsignalized intersections, roundabouts (Fig. 6.1), small networks, and
highway sections as well. The version of Hutsim used in this research was 4.21.
Figure 6.1: Hutsim roundabout model
The building blocks of Hutsim are object-oriented programming and rule-based in-
teraction dynamics. The most important models to control the movement of vehicles
are:
1. Vehicle acceleration and deceleration models

2. Car-following model
3. Lane-changing model.
Kosonen (1999) has described each model in detail.

Lane segments are modeled as links or “pipes”, in which vehicles move. The main prop-
erties of “vehicle objects”, initialized at generation are identification number, vehicle
type, length, desired speed and destination. Traffic signals as well as stop and yield
signs are “obstacle signals” connected to pipes (Kosonen 1996).
For unsignalized intersections Hutsim applies a traffic-signal analogy approach: “red
signal” (stop) during blocks and “green signal” (go) during antiblocks. Stop sign gives
SIMULATION STUDIES
a red (stop) signal until a minor stream vehicle approaching the stop line has come to
a stop. “Stop” signal activates the deceleration model. “Go” signal has no effect on
vehicle movements. Yield and stop signs are both obstacles in vehicles’ path and “sight
objects”. They open the drivers’ sight to vehicles outside of their own driving path.
In Hutsim the vehicular interactions at an unsignalized intersection are slightly different
from the gap acceptance models. Safety lag υs in Hutsim is the minimum acceptable
lag (from the rear of a minor-stream vehicle to the front of a major-stream vehicle)
at conflict point, whereas critical gap is the minimum acceptable headway or lag that
a driver at a stop/yield line observes. A safety lag is used in Hutsim, because it is
possible to use the same parameter value for all vehicle types (Koivisto 1999). Safety
lags may follow a statistical distribution (Kosonen 1996). However, in this study, they
were constant.
The lanes and driving paths at an intersection are modeled as pipes. Stop/yield line is
typically at the end of a yielding pipe. Conflict point is usually at the end of a major
stream pipe, which may be connected to the same pipe as the yielding pipe. Figure 6.2)
presents a simple illustration of a yield-sign object, which controls two conflict points.
A more realistic simulation arrangement requires separate sign objects for all conflict
points.
Figure 6.2: The yield-sign object in Hutsim (Kosonen 1996). The “Critical Time Gap” in the
figure refers to “safety lag”.
When the distance (dy ) of a yielding vehicle to a conflict point is shorter than the
maximum sight distance ds , the yield sign indicates whether it is safe to proceed (“go”)
or not (“stop”). When a minor-stream vehicle approaches the intersection at speed vy
and the next major-stream vehicle is approaching with speed vM at distance dM from
the conflict point, the yield sign gives a stop signal if
# $
dM dy + ly vy2
< + υs AND < αmax OR dy > ds , (6.1)
vM vy 2dy
where ly is the length of the yielding vehicle.1 The conflict point is the collision point of
the rear bumper of a yielding vehicle and the front bumper of the major stream vehicle.
The yield sign will give a stop signal, if the time distance at conflict point, when a minor
1 Equation 6.1 is a slightly modified version of the equation of Kosonen (1996).
SIMULATION STUDIES
Figure 6.3: Simulation screen in Hutsim
stream vehicle has passed it, to the next approaching major stream vehicle is shorter
than the safety lag υs and the minor stream vehicle is able to stop without exceeding
the maximum deceleration rate. If multiple major-stream lanes (pipes) are connected
to one yield or stop sign, the status of the sign is determined by the first vehicle in all
pipes to enter the conflict point. As long as the distance dy to the yield sign is longer
than the maximum sight distance ds , the driver observes it as a stop signal. A short
sight distance forces the vehicle to slow down and prepare to stop. (Kosonen 1996)
Major-stream vehicles are assumed to drive at their current speed, possible future ac-
celerations or decelerations are not predicted. Acceleration is accounted for a stopped
or nearly stopped minor-stream vehicle, but its driving time to conflict point may be
overestimated for some slowly moving vehicles. Accordingly, the safety-lag estimate
used in the decision making may be inaccurate. In order to minimize the number of
inconsistent operations the conflict points were set to the stop lines.
Rule 6.1 considers driving times to a conflict point. In general, the operations of
simulated vehicles are based on the properties of the vehicle, environment, control,
and other vehicles stored in a data base, not on driver’s observations. Thus, Hutsim
cannot be used to analyze the gap acceptance process, when observation of major road
vehicles exiting before the conflict point is unclear. However, Hutsim can model the
modification of the headway distribution of through vehicles due to the deceleration of
right turning vehicles in a shared lane.
Critical gap in Hutsim is the sum three components:
1. Occupancy time (tM ) of a major stream vehicle at a conflict point
2. Occupancy (tm ) of a minor stream vehicle at the conflict point
3. Safety lag (υs ) to the arrival of the next priority vehicle to the conflict point
SIMULATION STUDIES
The minor stream vehicle can enter the conflict point immediately after the major stream
vehicle has passed it, no safety interval is defined, neither has the conflict point any
physical dimensions. A critical lag is the sum of components 2 and 3 above. For a given
safety lag the critical gap is larger than the critical lag. Although all vehicles obey the
same safety lag, the minimum acceptable gaps and lags may be different for different
vehicles, depending on vehicle dynamics and the initial speed of a minor-stream vehicle.
For a major-stream vehicle having length lv and driving at speed v, the occupancy time
at a conflict point is tM = lv /v. The stop or yield line is assumed to locate at the conflict
point. For a stopped minor-stream vehicle having the same length lv and acceleration
√
rate a, the occupancy time at a conflict point is tm = 2lv /a. The critical gap is
obtained as
tc = tM + tm + υs
1
lv 2lv (6.2)
= + + υs .
v a
In Hutsim passenger cars have length lv = 4 m and acceleration rate a = 1.6 The
difference between a critical gap and a safety lag is about 2.5 s, when v = 50 km/h and
about 2.4 when v = 100 km/h. If a minor-stream vehicle passes a conflict point at a
negotiation speed of 30 km/h, the difference tc − υs is approximately 0.8–0.6 seconds.
The vehicle-dynamics parameters are independent of traffic conditions. Accordingly,
follow-up times in Hutsim are not related to major stream flow rates or waiting times.
Hutsim has a more realistic model for vehicle dynamics than gap acceptance and
queuing models. Compared to signalized intersections, where the status of signals
follows explicitly defined rules, the modeling of unsignalized intersections is much
more complicated. In Hutsim the decision making processes of drivers at unsignalized
intersections are not more sophisticated than the mathematical gap acceptance models.
Hutsim has not been calibrated to model the variation in the behavior of a driver or
between drivers. Although safety lags between drivers may follow a statistical distri-
bution, Hutsim assumes consistent drivers. The criticism directed at gap acceptance
models is in many respects also valid for Hutsim, as well as for many other simulation
models (see Hagring 1996a).
6.2 Calibration of Hutsim for unsignalized intersections

6.2.1 Vehicle-dynamics parameters
Hutsim has been extensively calibrated and validated for Finnish signalized intersec-
tions (Niittymäki 1993, Niittymäki & Pursula 1994, Niittymäki 1998). The micro-
scopic and macroscopic properties of the system were measured, the Hutsim para-
meters were calibrated, and the macro-level results were compared with field meas-
urements. The results of Hutsim were considered very reliable. It can be assumed
that the vehicle-dynamics parameters are similar in both signalized and unsignalized
intersections. The calibration and application of Hutsim for roundabouts has been
reported by (Hagring 1997, Pursula et al. 1997).
During this study field data were collected to further calibrate Hutsim for the analysis of
unsignalized intersections, including roundabouts. The results of these measurements
have been reported by Koivisto (1999). Only a short overview will be given here.
In Hutsim the departure processes at unsignalized intersections are simulated using
the rules and parameters of vehicle dynamics (Table 6.1). Critical gaps and follow-
up times are not used as parameters. Because changing the safety lag has no effect
SIMULATION STUDIES
on vehicle dynamics parameters, Hutsim cannot be used to study, for example, the
possible correlation between critical gaps and follow-up times.
Table 6.1: Vehicle dynamics parameters for unsignalized intersections in Hutsim (Pursula
et al. 1997, Koivisto 1999)
Vehicle class
Passenger car City bus Tram Truck
Parameter Unit Normal Maximum
Acceleration m/s2 1.6 1.2 1.3 1.0–1.2
Deceleration –m/s2 1.9 3.2 1.3 1.3 1.2
All vehicle classes
Follower gap s 1.2
Starting response time s 0.9
The arrivals were generated following a shifted exponential headway distribution with a
minimum arrival-time headway tp = 0.5 seconds. The cumulative distribution function
of the headway distribution was

0, if t < 0.5 s
F(t) = −θ(t−0.5)
(6.3)
1−e , if t ≥ 0.5 s.
The random number seed in a Hutsim run was the same in all simulation runs. Thus,
it was possible to replicate the same traffic input with different control parameters.
However, it was not possible to analyze the variance of results with given control
parameters.
6.2.2 Follower gaps

Follower gap is the desired time interval (from back bumper to front bumper) between
two consecutive vehicles. When major-stream flow rate is low, follow-up time is the
the most important parameter to describe minor-stream capacity, and follower gap is
the most important parameter to describe the follow-up time. The follower gap in
Hutsim was calibrated both by direct follower-gap measurements and by verifying that
the resulting follow-up times are realistic.
Follow-up times were measured at eight roundabouts. The diameter (d) of the central
island ranged from 8 to 55 meters. The measured follow-up times decreased slightly
with increasing diameter (Koivisto 1999):
tf = 2.2506 − 0.0043d, R 2 = 0.31. (6.4)
When the follower gap in Hutsim was set to 1.2 s, the resulting follow-up times were
slightly higher than obtained in the field measurements. With follower gap 1.1 s Hutsim
slightly underestimated the follow-up times.—Compared to international guidelines the
measured follow-up times were very short.
Follow-up times were also measured at three T-intersections. The results were eval-
uated by movement and by queue position. For major-road left turns and minor-road
right turns the follow-up times were about 2.0–2.1 seconds. For minor-road left-turn
movements the follow-up time was about 2.25 seconds. These data were, however,
from one intersection only.
With follower gap set to 1.2 seconds Hutsim overestimated the follow-up times, except
for minor-road left turns, for which Hutsim produced good estimates. When the fol-
lower gap was set to 1.1 seconds, the follow-up times produced by Hutsim were close
SIMULATION STUDIES
to field measurements. The follow-up times at a minor-road left-turn were, however,

underestimated.
The follower gaps were also measured directly. The follower gap of 1.2 seconds was
found to better describe the field measurements, and it was used in the simulation runs. It
may underestimate capacities in some cases, but it is not likely to produce unrealistically
high capacities. Follower gap 1.2 seconds is the default value in Hutsim.
6.2.3 Driving speeds

Acceleration ceases when a vehicle has attained its target speed or the maximum speed
at current location, such as a roundabout. Target speeds have a discrete distribution.
Koivisto (1999) used the speed distributions in Table 6.2. In this study the average
target speed was 80 km/h with 20 % of drivers having target speed 70 km/h and another
20 % having target speed 90 km/h. At the intersection the speed limit was, however,
60 km/h. This arrangement allowed the possibility of an arrival rate higher than the
queue-discharge rate. For the major-road through stream this arrangement was not
necessary. The average target speed was 60 km/h with 20 % of drivers having target
speed 50 km/h and another 20 % having target speed 70 km/h. Figure 6.4 displays speed
as a function of distance when a stopped vehicle accelerates at rate 1.6 m/s2 .
Table 6.2: Speed distributions used by (Koivisto 1999)

Target Speed limit
speed (km/h) 50 km/h 80 km/h 100 km/h
50 60 %
60 40 %
70 10 %
80 40 %
90 50 % 60 %
100 30 %
110 10 %
Maximum speeds of turning and circulating vehicles were calibrated with field measure-
ments (Koivisto 1999). At roundabouts the driving speed v of passenger cars increases
with the diameter r of the central island:
v = 26.676 + 0.0862r. (6.5)
The speeds of vehicles turning right or left at a conventional unsignalized intersection

are similar to the speeds at small roundabouts. The negotiation speed of left turning
vehicles was set to 30 km/h and of minor-road through vehicles to 40 km/h.
6.2.4 Limited priority merge

Although not a calibration issue, a short discussion is presented here about the con-
ditions under which Hutsim simulation describes a limited priority merge (see page
60). Priority becomes limited if a merging vehicle cannot reach its target speed before
a major stream vehicle driving with its target speed has reached the follower gap tmin
behind the merged vehicle. When a merging vehicle passes the conflict point with
speed v0 and accelerates at constant acceleration rate a, the vehicle reaches its target
SIMULATION STUDIES
35
30
25
Speed (km/h)
20
15
10
0
0 5 10 15 20 25
Distance (m)
Figure 6.4: Speed as a function of distance with the default acceleration rate 1.6 m/s2 of
passenger cars in Hutsim
speed v after time ta and distance da :

v − v0
ta = (6.6)
a
v 2 − v02
da = . (6.7)
2a
Limited priority merge takes place when the major stream vehicle drives the distance
da in a time shorter than ta − υs + tmin , where υs is the safety lag, and the speed of the
major-stream vehicle is equal to the target speed (v) of the merging vehicle. In order
to avoid limited priority, the safety lag should be
da v − v0 v 2 − v02
υs ≥ ta − + tmin = − + tmin . (6.8)
v a 2av
Limited priority modifies the major stream headway distribution, which decreases ca-
pacity below the value expected under an unmodified headway distribution. This effect
is largest under reasonably high major-stream flow rates, when gap forcing may pro-
duce shock waves in the major stream. Figure 6.5 displays an estimate of minimum
safety lags required for absolute priority merge.
6.2.5 The effect of right-turn vehicles

If a turning vehicle decelerates from v1 to v2 at rate a, the deceleration takes time
v1 − v2
td = (6.9)
a
and distance
v12 − v22
dd = . (6.10)
2a
SIMULATION STUDIES
8
10
7
15
Minimum safety gap (s)

6 20
25
5
1
40 50 60 70 80 90 100 110
Target speed (km/h)
Figure 6.5: Minimum safety lag values in Hutsim to avoid gap forcing, when major and minor
stream vehicles are all passenger cars with equal target speeds, and speeds of merging vehicles
at conflict point range from 10 to 25 km/h
The delay due to deceleration for a right turning vehicle is

dd
Wd = td − . (6.11)
v1
If a turning vehicle decelerates from 100 km/h to 30 km/h with rate 1.9 m/s2 , the delay
of a right turning vehicle is 6.7 seconds. Because gap and lag estimation in Hutsim is
based on the assumption of constant speed, the delay in arrival time due to deceleration
is not correctly estimated and the lag extension cannot be utilized by minor-stream
vehicles. This decreases the simulated capacity. This is most likely the explanation for
the low simulated capacities obtained by Koivisto (1999).
Imperfect estimation of priority-vehicle deceleration may happen in the real world also,
but the effect is included in the critical gaps, if measured on field. If necessary, a specific
adjustment should be defined. However, the performance of basic models should be
evaluated without this effect.
It was considered necessary to perform the simulation studies with a limited number
of movements. Because of the problems described above more reliable results could
be obtained without any right-turn movements.
6.3 Simulated capacity of unsignalized intersections

6.3.1 Rank 2 capacity
For Rank 2 capacity analysis the stream arrangements displayed in Figure 6.6 were
analyzed. The minor-stream left-turn movement (7) was the most important movement
in the analysis. Because it has Rank 4, it can be analyzed with one, two, or three higher
rank conflicting movements. The arrangements included all two-stream combinations
used in the capacity analysis of Rank 3 and Rank 4 streams. The first three arrangements
were used in Rank 3 and Rank 4 analysis. The last three arrangements were used in
Rank 4 analysis only.
SIMULATION STUDIES
4 (R1)
2 (R1)
7 (R2) 7 (R2)
11 (R1)
4 (R2)
2 (R1)
7 (R2)
11 (R2) 11 (R2)
4 (R1)
2 (R1)
Figure 6.6: Simulated streams for Rank 2 capacity analysis
The safety lags in the Hutsim runs were 2, 3, and 4 seconds. The minor roads and
major-road left-turn streams had a yield-sign object with sight distance 200 meters.
A stop sign was simulated by setting a yield-sign with a sight distance of 1.5 meters.
In Hutsim a yield-sign object gives a stop signal to an approaching vehicle until it is
closer than the sight distance.
Capacity models were estimated using the least squares procedure. For the exponential
capacity model (3.12) the estimated parameters were the critical gap (tc ) and the follow-
up time (tf ). For the shifted exponential model (3.17) the minimum time headway (tp )
was also estimated. Critical gaps were estimated as initial critical gap (for a two-second
safety lag) and an increment corresponding to one-second increment in a safety lag.
Figure 6.7 displays capacity estimates for a stop-controlled minor-road left-turn stream
(7) crossing a major-road through stream (2). The capacity curves are approximately
SIMULATION STUDIES
linear, which suggests a shifted exponential capacity model. Shifted exponential dis-
tribution was the headway distribution used in the vehicle generation. Apparently,
vehicular interactions before the intersection did not change the headway distribution
significantly enough to be observed in the capacity models.
1200
1000
Minor stream capacity (pc/h)
800
600
400
200
0
0 200 400 600 800 1000 1200 1400 1600 1800
Major stream flow rate (pc/h)
Figure 6.7: Capacity of a stop-controlled minor-road left-turn stream crossing a major-road

through stream. Simulated (red triangles), exponential (solid blue) and shifted exponential
(dashed black) capacity curves with safety lags 2 (upper), 3 (middle), and 4 (lower) seconds.
The shifted exponential model fits well the simulated capacities. The follow-up time
3.7 s (Table 6.3) indicates a capacity of 980 pc/h, when the major stream is negligible.
Because the vehicle-dynamics parameters do not change with the safety lag, neither
changes the follow-up time. The critical gap is approximately two seconds larger than
the safety lag. The minimum major-stream headway (1.1 s) is approximately the same
as the minimum headway parameter (1.2 s) in the simulation model.
Table 6.3: Gap-acceptance parameters for a minor-road left-turn stream crossing a major-road
through stream
Stop sign Yield sign
Parameter Shifted Negative Shifted Negative Safety
exponential exponential exponential exponential lag (s)
Critical gap (tc ) 4.2 5.4 4.7 5.8 2
5.0 6.5 5.6 6.9 3
5.9 7.5 6.5 7.9 4
Follow-up time (tf ) 3.7 3.4 2.1 2.1
Min. headway (tp ) 1.1 0.0 0.9 0.0
The critical gaps and follow-up times obtained by fitting the shifted exponential model
describe driver behavior in the simulations better than the parameters for the negative
exponential model. However, the application of the parameters of the shifted exponen-
tial capacity model in a negative exponential capacity model (Figure 6.8) overestimates
SIMULATION STUDIES
the simulated capacity.
1000
900
800
700
600
500
400
300
200
100
0
0 200 400 600 800 1000 1200 1400 1600 1800

Parameters obtained by fitting the shifted exponential model to the simulation results.
At a yield-controlled intersection the follow-up time is lower and critical gaps are higher.
At low major-stream flow rates the capacity is higher, but approaches the capacity of a
stop-controlled stream at high major-stream flow rates (Figure 6.9). One explanation
for larger critical gaps at yield-controlled approaches may be that in Hutsim a slow
moving vehicle cannot always anticipate its acceleration while it estimates its arrival
time to the conflict point. Accordingly, Table 6.3 describes the relationships between
safety lags and critical gaps at stop and yield controlled approaches in Hutsim. It
should not be used to compare the effects of stop and yield signs at any intersection.
The critical-gap estimates for the exponential model are more than a second larger than
the critical gaps for the shifted exponential model. The follow-up time is slightly lower
(3.4 s). Because the exponential model has a stronger curvature than the simulated
curves, the model overestimates the simulated capacity at low and high major-stream
flow rates. At moderate conditions the model underestimates the simulated capacity.
Major-road left-turn vehicles (4) have to decelerate to the negotiation speed (30 km/h)
before entering the intersection. Because in Hutsim a minor-road left-turn vehicle
(7) cannot anticipate this deceleration, minor-stream vehicles cannot use all acceptable
lags. Accordingly, the critical gaps in the theoretical models must be increased (Table
6.4).
The shifted exponential capacity model gives again a better fit than the negative expo-
nential model (Fig. 6.10). The capacity under negligible major-flow is, of course, the
same as in the major-road through case (Fig. 6.7). The follow-up time estimates in both
models are similar as in the major-road through case (Table 6.3). The critical gaps are,
however, larger, especially for large safety-lag values.
SIMULATION STUDIES
1800
1600
1400

1200
1000
800
600
400
200
0
0 200 400 600 800 1000 1200 1400 1600 1800
Figure 6.9: Capacity of a yield-controlled minor-road left-turn stream crossing a major-road

Table 6.4: Gap-acceptance parameters for a minor-road left-turn stream crossing a major-road
left-turn stream
Stop sign Yield sign
Critical gap (tc ) 5.6 7.4 6.6 8.1 2
6.9 8.9 8.0 9.6 3
8.3 10.4 9.3 11.0 4
Follow-up time (tf ) 3.7 3.5 2.2 2.1
Min. headway (tp ) 1.3 0.0 1.0 0.0
At a yield-controlled intersection follow-up times are smaller and stream capacity is

higher, but it approaches the capacity of a stop-controlled stream at high major-stream
flow rates (Fig. 6.11).
A major-road left-turn vehicle (4) does not have to stop, if it finds an acceptable lag in
the major-road through stream (2). Consequently, the follow-up time and critical gaps
are short (Table 6.5), and the capacity under low major-stream flow rates is high (Fig.
6.12). The critical gaps in the shifted exponential capacity model are approximately
1.6 seconds larger than the safety lag. In the negative exponential model the critical
gaps are more than a second larger. The shifted exponential model has again a better
fit, but the negative exponential model fits the simulated capacities better than in the
case of a stop-controlled minor stream. The parameters are practically the same as for
a yield-controlled minor-road left-turn stream (Table 6.3). In essence, the simulation
models are the same.
SIMULATION STUDIES
1200
1000
800
600
400
200
0
0 200 400 600 800 1000 1200 1400 1600 1800

left-turn stream. Simulated (triangles), exponential (solid blue) and shifted exponential (dashed
black) capacity curves with safety lags 2 (upper), 3, and 4 (lower) seconds.
1800
1600
1400
1200
1000
800
600
400
200
0
0 200 400 600 800 1000 1200 1400 1600 1800
Figure 6.11: Capacity of a yield-controlled minor-road left-turn stream crossing a major-road

left-turn stream. Simulated (triangles), exponential (solid blue) and shifted exponential (dashed
SIMULATION STUDIES
Table 6.5: Gap-acceptance parameters for a major-road left-turn stream crossing a major-road
through stream
Parameter Shifted Negative Safety
exponential exponential lag (s)
Critical gap (tc ) 4.7 5.8 2
5.6 6.9 3
6.6 7.9 4
Follow-up time (tf ) 2.2 2.1
Min. headway (tp ) 0.8 0.0
1800
1600
1400
1200
1000
800
600
400
200
0
0 200 400 600 800 1000 1200 1400 1600 1800
Figure 6.12: Capacity of a major-road left-turn stream crossing a major-road through stream.
Simulated (triangles), exponential (solid blue) and shifted exponential (dashed black) capacity
curves with safety lags 2 (upper), 3, and 4 (lower) seconds.

The capacity of a Rank 3 stream 7 was analyzed with the stream arrangement of Figure
6.13. The safety lag of stream 4 was two seconds. Stream 7 had a (simulated) stop sign
and a three-second safety lag. A stop sign was simulated by setting a yield-sign with a
sight distance of 1.5 meters.
The simulated capacity of a minor-road left-turn stream (7) is displayed in Figure 6.14.
Simulated major-stream flow rates did not reach 1,800 pc/h, because at high Rank 1
flow rates the number of Rank 2 vehicles passing the intersection was restricted by
Rank 2 capacity.
The critical gaps and follow-up times used in the estimation of the theoretical (3.89) and
conventional (3.72) capacity models were obtained from the Rank 2 capacity analyses
based on the shifted exponential model presented above (Tables 6.3, 6.4, and 6.5).
Because the critical gaps of stream 7 were different for major streams 2 and 4, the
critical gap used in the analysis was the weighted average of these two critical gaps.
SIMULATION STUDIES
4 (R2)
2 (R1)
7 (R3)
700
600
500
400
300
200
100
0
200 400 600 800 1000 1200 1400 1600 1800
Figure 6.14: Rank 3 capacity of a stop-controlled minor-road left-turn stream (7) crossing
major-road through (2) and left-turn (4) streams with left-turn flow rates 300 (upper curves)
and 600 pc/h (lower curves). Simulated capacity (triangles), conventional capacity estimate
(solid magenta) and theoretical capacity estimates(solid blue). Dashed blue curve displays the
theoretical capacity estimate assuming a 1.8 second discharge headway for the Rank 1 stream.
The flow rates of the major movements were used as weighting factors.
The conventional capacity estimator is based on random arrivals and impedance factor
p0 = 1 − ρ, as used in both HCM2000 (Transportation Research Board 2000) and
Capcal 2 (SNRA 1995b). As figure 6.8 demonstrates, the negative exponential ca-
pacity model (3.12) overestimates potential capacity at high major-stream flow rates.
However, the impedance factor (3.76) in the conventional model gives too low capacity
estimates. As a result, the conventional model gives too low capacity estimates at low
major-stream flow rates, but at high flow rates the errors begin to cancel each other, and
the capacity estimates are acceptable.
The theoretical model gives higher capacity estimates than the simulation model. Better
estimates are obtained assuming a 1.8 s discharge headway for the Rank 1 stream.
SIMULATION STUDIES
Consequently, Rank 1 headways follow the Tanner model.

When a Rank 2 queue discharges, stopped vehicles accelerate to the negotiation speed.
Because yielding vehicles in Hutsim cannot anticipate the acceleration of higher-
priority vehicles, a Rank 3 vehicle may enter the intersection when the Rank 2 queue
starts to discharge. This causes a small positive bias in simulated capacities. This bias
is largest (about 20 pc/h) at saturated conditions.

The capacity analysis of a Rank 4 stream (7) was based on the stream arrangement in
Figure 6.15. Because a Rank 4 vehicle in Hutsim could not anticipate the deceleration
of a stopping Rank 3 vehicle, the resulting critical gaps were very high and capacit-
ies were very low. Better results were obtained in the analysis of a yield-controlled
intersection.
11 (R3)
4 (R2)
2 (R1)
7 (R4)
The capacity of a minor-road left-turn stream (7) crossing a minor-road through stream
(11) is similar to the capacity of crossing the major-road left-turn stream (4). The only
difference in the simulation model is a higher negotiation speed (40 km/h) for stream 11
than for stream 4 (30 km/h). Consequently, Hutsim gives lower critical gaps (Table 6.6)
and higher capacities for crossing the through stream (11). Again, this is not necessarily
a reflection of reality, but a property of the simulation model. When negotiation speed
is higher, the deceleration time is shorter, and the error in the estimated arrival time to
the conflict point is smaller.
Table 6.6: Gap-acceptance parameters for a yield-controlled minor-road left-turn stream cross-
ing a minor-road through stream
Parameter Shifted Negative Safety
exponential exponential lag (s)
Critical gap (tc ) 5.7 7.0 2
7.1 8.5 3
8.4 9.9 4
Follow-up time (tf ) 2.2 2.1
Min. headway (tp ) 0.9 0.0
The minor-road through stream (11) must yield to streams 2 and 4. Because the ne-
gotiation speed of stream 11 in the Hutsim model is higher (40 km/h) than of stream
7 (30 km/h), the follow-up time is shorter (see Table 6.7). This gives a higher capa-
city at low major-stream flow rates (Figure 6.16). The critical gaps are larger when
SIMULATION STUDIES
merging with the major-road left-turn stream, because the left-turn stream has a lower
negotiation speed (30 km/h) than the major-road through stream (60 km/h), and in Hut-
sim the minor-road vehicles cannot anticipate the extension of a gap or a lag due to a
deceleration.
Table 6.7: Gap-acceptance parameters for a yield-controlled minor-road through stream cross-
ing a major-road through and left-turn streams
Major through Major left
Critical gap (tc ) 4.9 6.0 6.9 8.3 2
5.8 7.0 8.1 9.6 3
6.8 8.1 9.3 10.9 4
Follow-up time (tf ) 2.0 1.9 2.0 1.9
Min. headway (tp ) 0.8 0.0 0.9 0.0
2000
1800
1600
1400
1200
1000
800
600
400
200
0
0 200 400 600 800 1000 1200 1400 1600 1800
Figure 6.16: Capacity of a yield-controlled minor-road through stream crossing a major-road

through stream. Simulated (triangles), exponential (solid blue) and shifted exponential (dashed
Simulated Rank 4 capacities and capacity estimates based on the conventional (3.104)
and theoretical (3.114) models are displayed in Figure 6.17. The safety lag for the
major-road left-turn stream (4) was 2.0 seconds, for the minor-road through stream
(11) it was 2.5 seconds, and for the minor-road left-turn stream (7) the safety lag was
3.0 seconds. Longer safety lags at lower priority streams were used in order to minimize
the number of “sneakers”. The critical gaps and follow-up times were obtained from
the shifted exponential capacity models in the Rank 2 analysis. For minor streams
crossing several priority streams the critical gaps were estimated as averages weighted
by the conflicting flow rates. For stream 11 the values were interpolated.
The simulated priority-stream flow rates were measured at exit points. At high flow rates
SIMULATION STUDIES
500
450
400

350
300
250
200
150
100
50
0
600 700 800 900 1000 1100 1200 1300 1400 1500 1600
Figure 6.17: Rank 4 stream capacity at a yield-controlled intersection with Rank 3 flow rate
300 pc/h and Rank 2 flow rates 300 (upper curves) and 600 pc/h (lower curves). Simulated
capacity (triangles), conventional capacity estimate (3.104, magenta) and theoretical capacity
estimate (3.114, blue). Dashed blue curves display the theoretical capacity estimates assuming
a 1.8 second discharge headway for the Rank 1 stream. Dotted black curves display the
conventional capacity estimates without the additional adjustment (3.103).
capacity may have restricted the number of departures of Rank 2 or Rank 3 streams.
Consequently, the total major-stream flow rate did not reach 1,800 pc/h, which was the
maximum expected traffic generation rate; i.e. the demand.
The capacity did not decrease to zero at high major-stream flow rates, because in Hutsim
some Rank 4 vehicles departed while a Rank 2 or Rank 3 queue started to discharge,
as discussed above. This factor increased the capacity at high major-stream flow rates
by approximately 25 pc/h.
The conventional model slightly underestimates the simulated capacity. As observed
in the Rank 3 analysis, the impedance factor (3.103) gives too low values, while the
exponential model (3.12) with realistic parameter values tends to overestimate the
simulated capacity (Fig. 6.8). The dotted curves display the conventional capacity
estimates without the additional adjustment (3.103), as in Capcal 2. The estimates are
too low. The adjustment (3.103) in HCM2000 improves the estimates considerably.
The theoretical model gives slightly too high capacity estimates, especially if the simu-
lation results are adjusted for the effect of “sneakers”. The assumption of a 1.8 second
discharge headway in the Rank 1 stream has a minor decreasing effect on the results.
6.4 Simulated capacity of roundabouts

Figure 6.18 displays potential capacity estimates for two roundabouts based on Hutsim
simulations. Two central island diameters (8 m and 38 m) were used. Safety lag was two
seconds. Capacity decreases almost linearly with increasing circulating flow. Larger
diameter (d) increases capacity by approximately 50 pc/h. At low circulating flow rates
SIMULATION STUDIES
the increase in capacity is higher. Linear regression equations are
Ĉp = 1436 − qc pc/h, d = 8m (6.12a)

Ĉp = 1484 − qc pc/h, d = 38 m (6.12b)
From the family of M3 capacity models (3.19) the shifted exponential (φ = 1) model
(3.17) gives the best fit, when minor stream capacity decreases linearly with increasing
major stream flow rate. Figure 6.18 displays these models fitted for the simulation
results. The critical gap was four seconds. The follow-up time was 2.5 s for the
smaller roundabouts and 2.3 s for the larger roundabout. The saturation flow rate for
tf = 2.5 is 1,440 pc/h, which is close to the capacity of the smaller roundabout under
very low circulating flow rates. For the larger roundabout the follow-up time tf = 2.3
gives a higher saturation flow rate (1,565 pc/h) than predicted by the regression line
(1,484 pc/h), but the simulation results indicate that the decrease in entry capacity is
steeper at low circulating flow rates. For the larger roundabout the minimum headway in
the circulating flow is tp = 1.8 s, as in the Swedish models. For the smaller roundabout
the results suggest a slightly larger (tp = 2.0 s) minimum headway.
1600
Simulated 8m
Shifted exp.
1400 Simulated 38m
Shifted exp.
1200
Entry capacity (pc/h)
1000
800
600
400
200
0
0 500 1000 1500
Circulating flow rate (pc/h)
Figure 6.18: Capacity of a one-lane roundabout (simulation results and shifted exponential
models) with critical gap tc = 4 s and central island diameters 8 m and 38 m
The simulation results indicate a nearly linear relationship between entry capacity
and circulating flow. Similar results have been obtained for two-lane roundabouts
(Hagring 1997, Pursula et al. 1997). Linear capacity models have been supported by
both empirical (Kimber & Coombe 1980) and theoretical (Troutbeck 2002) studies.
The linearity is supposed to result from decreasing critical gaps and limited priority
under high circulating flow rates. These phenomena are, however, not included in the
simulation model.
The shifted exponential model suggested by the linear relationship is the headway model
used in Hutsim. When the follower gap in Hutsim is 1.2 s, and circulating speeds are
SIMULATION STUDIES
about 30 km/h, the follower headway is approximately 1.8 s. Further research is needed
to evaluate the effect of headway distributions on roundabout capacity.
6.5 Simulated delays

For delay analysis the operation of an intersection was simulated for 80 minutes. The
first 10 minutes was an intialization interval, after which data was collected for 60
minutes (report interval). The final 10-minute interval was required by Hutsim ba-
sic reporting system for vehicles that were generated during the “report interval” but
departed the system after it.
Data was collected from detector output stored in a “delay file”. The detector was
located at the intersection exit point. Vehicles passing the detector during the report
interval were included in the analysis. Accordingly, the delay analysis followed the
path-trace method.
The analytical delay formulas are based on the queue-sampling method. At undersat-
urated conditions the results of path-trace and queue-sampling methods are similar. At
oversaturated conditions the methods give different delay estimates, as demonstrated
in Figure 4.12 on page 131. The queue-sampling overflow delay during a time period
(0, t] is the area 0QR, while the path-trace method gives overflow delay 0QP. In both
methods the average delay is obtained by dividing the total delay by the number of
departures, C(t). The queue-sampling method overestimates the average delay, be-
cause it includes a delay component PQR of additional vehicles, which is not included
in the path-trace method. On the other hand, the path-trace method does not include
the delays accumulated to the vehicles in the queue. These vehicles have started to
experience delays larger than the average delays of departed vehicles.
The delay of each vehicle was calculated from the Hutsim delay file. The difference
between the travel times from traffic generator to the intersection exit point under given
conditions and under free conditions without a stop sign was used as an estimate of
control delay. Because simulated minor-stream flow-rates were measured at exit point,
they could not exceed capacity even though vehicle generation rate (demand) may have
been higher.
The average control delay of a stop-controlled minor-road left-turn stream crossing
a major-road through stream is displayed in Figure 6.19. The delay estimates were
calculated according to the coordinate-transformation method using equations 4.131
and 4.132. The average acceleration delay was estimated as five seconds, following
the HCM2000. A lower value would have fit better the simulation results, but the
acceleration delays due to a “simulated stop sign” in Hutsim may have been too low. A
more detailed analysis of the acceleration delay would have required field data, which
was not available for this research.
The capacity estimates were based on the shifted exponential capacity model (3.17) with
parameters as in Table 6.3. The figure demonstrates that the coordinate transformation
method gives reliable delay estimates, when capacity estimates are accurate.
HCM2000 calculates capacity (3.12) assuming negative exponential headways. To
account for the effect of a stop sign, five seconds are added to the delay estimate of the
coordinate-transformation method (4.131). The follow-up time is not subtracted from
the waiting time in the system. The delay estimates are good, although they do not
follow the simulation results as closely as the blue curves, which had capacity estimates
based on the properties of the simulation model.
Figure 6.20 displays similar analysis for a yield-controlled approach. Because vehicles
SIMULATION STUDIES
50
45
40
35
Average control delay (s)
30
25
20
15
10
0
0 100 200 300 400 500 600 700
Minor stream flow rate (pc/h)
Figure 6.19: Average control delay of a stop-controlled minor-road left-turn stream cross-
ing a major-road through stream having flow rates 300 (lower curve), 600 (middle), and
900 pc/h (upper). Simulated delays (red triangles) and delay estimates based on the coordinate-
transformation method (shifted exponential capacity model, blue curves) and the HCM2000
method (dashed magenta curves).
accepting a lag do not stop, the average acceleration delay was calculated as

tf C tf C
W̄a = 5 1 − =5− . (6.13)
3600 720
It was assumed that the acceleration delay of a delayed vehicle is five seconds, and the
proportion of undelayed vehicles is the proportion of actual capacity (C) to maximum
capacity (tf−1 ).
The coordinate-transformation method again produces good delay estimates, when
capacity estimates are based on the shifted exponential headway distribution. The
delays are slightly overestimated, but the shape on the delay curves are very realistic.
Especially at low major-stream flow rates many delayed minor-stream vehicles can
cross the yield line without stopping. A better acceleration delay estimate would have
improved the delay estimates. It was, however, considered appropriate to keep the
estimation of average acceleration delay very simple. The difference between the
estimates would most likely have been in the range of 1–2 seconds, and the validation
would have been possible only against simulation results.
HCM2000 delays were too high at low minor-stream flow rates. If five seconds (the
adjustment for a stop sign) was subtracted, the delay estimates would have improved.
Because HCM2000 capacity estimates were too high, delay estimates were too low at
high flow rates.
It is evident that with good capacity estimates the coordinate-transformation method
gives reliable delay estimates. It was also demonstrated that the HCM2000 delay
estimation method (coordinate-transformation method) can be improved by subtracting
SIMULATION STUDIES
50
45
40
35
Average control delay (s)

30
25
20
15
10
0
0 100 200 300 400 500 600 700 800 900 1000
Minor stream flow rate (pc/h)
Figure 6.20: Average control delay of a yield-controlled minor-road left-turn stream cross-
ing a major-road through stream having flow rates 300 (lower curve), 600 (middle), and
900 pc/h (upper). Simulated delays (red triangles) and delay estimates based on the coordinate-
transformation method (shifted exponential capacity model, blue curves) and the HCM2000
method (dashed magenta curves).
the follow-up time from the waiting time in the system. A simple adjustment (6.13) for
the acceleration delay at yield-controlled intersections was shown to give reasonable
results.
CRITICAL GAPS AND FOLLOW-UP TIMES
7 CRITICAL GAPS AND FOLLOW-UP TIMES

7.1 Estimation methods
Critical gaps and follow-up times are measures of driver behavior at unsignalized in-
tersections. Gap acceptance theory usually assumes that drivers are consistent and
homogeneous. Gap acceptance is modeled by a step function (see Section 3.2.2),
which indicates a constant critical gap and follow-up time for each movement. This
assumption was used in the simulation studies also. In order to use the capacity and
delay models, critical gaps and follow-up times have to be estimated.1
It is not possible to measure critical gaps directly. It is only possible to observe a driver
either accepting or rejecting a gap. Under consistent behavior an accepted gap is an
upper bound for the critical gap of the driver. Likewise, the largest rejected gap is a
lower bound for the personal critical gap. If drivers are inconsistent, the estimation
becomes even more complicated. It is not possible to observe a minimum acceptable
gap/lag that may be different for different gaps/lags. Darzentas (1981) concludes that
the distribution of accepted gaps/lags cannot be estimated from field data. However,
probability distributions of gap acceptance created under the assumptions of consistent
and inconsistent behavior are identical.
The capacity analysis methods described above assume consistent and homogeneous
drivers. Gap acceptance can, accordingly, be described by a point estimate, a critical
gap. The critical gap studies in Finland have been based on the Raff’s method. Because
recent international studies have indicated that the maximum likelihood method gives
more accurate estimates, both of these methods have been applied in this research.
According to the Raff’s method (Raff 1950) the critical gap (tc ) is the time-gap that
has the property that the number of accepted gaps shorter than tc is the same as the
number of rejected gaps longer than tc . A driver can accept only one gap, but reject any
number of gaps. Raff observed that if all gaps were included in the analysis, rejected
gaps had a too strong weight in the analysis.2 Consequently, Raff considered only lags
so that no more than one observation was collected for each driver. The number of
observations was, however, greatly reduced. Consequentlt, Hewitt (1985) considered
methods using only lags inferior to methods using both lags and gaps. In this study
all gaps were included in the analysis so that the results were compatible with earlier
Finnish critical gap studies.
Maximum likelihood method (ML) has been extensively used in statistical parameter
estimation (Edwards 1974, Edwards 1992). A ML estimate has the highest likelihood of
being the parameter of a hypothetical distribution which has generated the observations.
Miller & Pretty (1968) suggested the ML method to estimate critical gaps. The method
has been further developed and demonstrated to be one of the best critical-gap estima-
tion methods (Hewitt 1985, Troutbeck 1992, Brilon, Koenig & Troutbeck 1997, Tian,
Troutbeck, Kyte, Brilon, Vandehey, Kittelson & Robinson 2000).
7.2 Stop vs. yield control

Following the terminology of Greenshields et al. (1947) a stop sign indicates a functional
deceleration and stop (see footnote on page 108). All vehicles must stop at stop line.
1 If a linear gap acceptance function (see Section 3.2.3) is used, the number of discharging vehicles
as a linear function of headway length has to be estimated. The shortest acceptable headway is to =
inf {t : α(t) > 0}.
2 Raff also observed that a headway included the time interval that a major-stream vehicle occupied
the conflict point, while it was unoccupied during the entire lag.
A yield sign, on the other hand, indicates a chronotropic deceleration. Minor stream
vehicles must decelerate and maybe stop, if a block prevents their entry.3
A vehicle facing a yield sign does not have to stop, if it arrives during an antiblock.
A vehicle arriving at the end of a block may decelerate early so that at the beginning
of the block it may cross the stop line at a higher speed than possible otherwise. The
vehicle may accept a shorter gap or lag than a stopping vehicle. Also the acceleration
delay will be shorter. In addition, vehicles crossing a yield line during an antiblock may
discharge with shorter follow-up times than vehicles having to stop at a stop line. The
difference between these two control methods is largest at low degrees of saturation and
low major stream flow rates, when the proportion of minor stream vehicles accepting
lags is largest. As the major stream flow rate and the D/C ratio increase, more minor-
stream vehicles have to stop at yield line and the difference between stop control and
yield control decreases.
Some early German studies indicated that the priority rule has no influence on critical
gaps and follow-up times (Brilon 1988a). In the current German guidelines (FGSV
2001) the follow-up times are different for yield and stop controlled intersections,
but critical gaps are the same. The HCM 2000 (Transportation Research Board 2000)
method is for stop-controlled intersections only. Capcal 2 (SNRA 1995b) gives different
critical gaps for yield and stop controlled intersections. Follow-up times are also
different (tf = 0.6tc ). DanKap (Vejdirektoratet 1999b) has control-specific critical
gaps, but the same follow-up times. Consequently, the guidelines that address both
yield and stop control have either critical gaps, follow-up times, or both control specific.
The models used in this report assume that critical gaps and follow-up times are inde-
pendent of major and minor stream flow rates. Yield controlled approaches are likely
to have shorter critical gaps and follow-up times than stop controlled approaches, as-
suming that other conditions are similar.
The simulation results indicated clearly that follow-up times are different at yield and
stop controlled intersections. Because the safety lag in Hutsim was given as a para-
meter, the simulation results could not be used to estimate critical gaps.
7.3 Field studies

7.3.1 Data collection
The critical gap data available at the Helsinki University of Technology Laboratory of
Transportation Engineering were supplemented by field measurements at eight intersec-
tions and nine roundabouts. Traffic operations were video recorded during peak hours.
The data were collected and analyzed by the staff of the Laboratory. All locations were
in the Southern Finland.
Intersections were chosen so that three different major-road speed limits were included:
50 km/h, 60 km/h, and 80 km/h. Three movements were studied:
1. Left turn from major road
2. Right turn from minor road
3. Left turn from minor road
Times, when a vehicle arrived to the stop-line and when a gap was accepted or rejected,
3 This makes also the estimation of acceleration delay difficult at yield-controlled intersections.
were written down. This material was adequate to get the accepted and largest rejected
gaps for the estimation of critical gap. The data contained mostly passenger cars.
Some of the not rejected headways were too long to give a realistic description of driver
behavior. Consequently, very large rejected headways were excluded from the analysis,
as described below.
7.3.2 Results
Although efforts were made to find intersections and time periods of high flow rates,
the major streams were dominated by long headways. Because accepted headways
were long, the information obtained about the upper bound of the critical gap was very
limited. It is also possible that drivers were willing to reject gaps that they would
have accepted under congested conditions, which may have resulted in too high lower
bounds for the critical gap. Accordingly, both lower and upper bounds may have been
overestimated.
The good supply of long headways appeared to distort especially the maximum likeli-
hood estimates. For comparison, maximum likelihood estimates were calculated also
for right-censored data. The maximum value for accepted gaps was set to 11 seconds
for roundabouts and for intersections with major-road speed limit 50 or 60 km/h, and to
17 seconds when the speed limit was 80 km/h. The results are displayed in Appendix
A. Raff’s (1950) method was not very sensitive to censoring.
For each movement both average and weighted average were calculated (see Appendix
A). The square root of sample size was used as the weighting factor. It became,
however, apparent that more data was needed, especially under high D/C ratios, in
order to obtain reliable critical-gap estimates. Until then the critical gaps presented by
Pursula et al. (1997) and displayed in Table 5.17 are suggested as base values. The
follow-up time estimate suggested by Jessen (1968)4 tf = 0.6tc can still be used as a
base value. However, the capacity estimates are more reliable, if field-measured critical
gaps and follow-up times are used.
For roundabouts the estimates in Table A.9 are in the range 4.0–4.7 s. No significant
correlation was found between the roundabout size and the critical gaps. However,
definite conclusions about the effect of roundabout geometry to critical gaps would
require more field data.
For typical Finnish roundabouts the critical gaps as suggested by Hagring (1996b) and
Capcal 2 (SNRA 1995c) are presented in Table 7.1 (see equations 5.18 and 5.16). The
small (10 m) roundabout does not have approach islands, which makes the weaving
section longer and the estimated critical gap shorter.
Table 7.1: Critical gaps for typical Finnish one-lane roundabouts according to equation (5.16)
of Capcal 2 (SNRA 1995c) and equation 5.18 of Hagring (1996b)
Central island Weaving section Weaving Critical gap, tc /s
diameter length section width Capcal 2
d/m lw /m ww /m Hagring Right turn Crossing
10 21 10.0 4.5 4.1 4.9
20 23 8.5 4.3 3.8 4.4
40 30 7.0 3.9 3.4 3.8
4 Later German studies, among others, have indicated that the direct correlation suggested by Jessen
(1968) may not exist (Brilon 1988a).
Because the Finnish observations presented in Table A.9 do not indicate a correlation
between critical gap and roundabout size, a single critical-gap value has been con-
sidered appropriate. The suggested base value for a critical gap in Finnish one-lane
roundabouts is 4.3 seconds, which is close to the estimates in Table A.9. It is also the
critical-gap estimate for a typical Finnish roundabout with central island radius of 20 m
(Table 7.1), as well as the critical gap at a two-lane roundabout between traffic flows in
outer circulating and entry lanes (Table 5.12), as estimated by Hagring (1998b). The
suggested value is also in the range of the upper (4.1 s) and lower (4.6 s) bounds (Table
5.3) in HCM2000 (Transportation Research Board 2000), but slightly higher than the
value 4.1 s suggested by the German HBS 2001 (FGSV 2001). For two-lane round-
abouts the critical gaps suggested in Table 7.2 follow the results of Hagring (1998b).
Table 7.2: Suggested critical gaps and follow-up times for Finnish roundabouts
One-lane roundabout Two lane roundabout
Entry Critical Follow-up Critical gap Follow-up
lane gap time Outside lane Inner lane time
Right 4.3 2.3–2.5 4.3 4.0 2.4
Left 4.6 4.4 2.4
For two-lane roundabouts the follow-up time 2.4 s was suggested by Hagring (1998b).
For one-lane roundabouts the follow-up time is calculated as
tf = 2.5 − 0.0067(d − 8), 8 ≤ d ≤ 40, (7.1)
where d = 2(rc − ww ) is the central island diameter. The resulting follow-up times are
similar to the result (2.4 s) of Hagring (1996b) but shorter than suggested by HCM2000
(2.6–3.1 s), DanKap (2.6–2.8 s), and HBS 2001 (2.9 s)5 , which indicates that Finnish
and Swedish roundabouts have higher capacity than the roundabouts in the U.S., Den-
mark, and Germany, when traffic flows are strongly dominated by one entry.
In the Australian guidelines (Austroads 1993) and in Sidra (Akçelik 1998) follow-up
times at one-lane roundabouts are related to both roundabout size (inscribed circle
diameter) and circulating flow rate. Follow-up times decrease with increasing diameter
and increasing circulating flow rate. At low circulating flow rates the follow-up time
ranges from 2.27 to 2.99 seconds, when the inscribed circle ranges from 80 to 20 meters
(Austroads 1993). When the circulating flow approaches 1,500 veh/h the follow-up
time can be as low as 1.7 seconds. In two-lane roundabouts with circulating flow rate
2,500 veh/h the follow-up time can be as low as 1.3 seconds. Sub-dominant streams
have larger follow-up times than dominant streams.
The follow-up times suggested in Table 7.2 are still larger than the follow-up times
obtained in field measurements (see equation 6.4). As the calibration studies indicated,
with follower gap 1.2 s Hutsim slightly overestimated follow-up times. The number
of field measurements was, however, limited. In the light of international results, more
field data are needed before lower follow-up times can be suggested.
The simulation results suggest that the minimum headway in circulating traffic is a
function of central-island diameter (d):
tp = 2.0 − 0.0067(d − 8), 8 ≤ d ≤ 40. (7.2)

5 New German research suggests 2.5 s as the follow-up time for a one-lane entry. (Thorsten Miltner at
the German–Dutch–Finnish Seminar on Traffic Engineering at 8–9 June 2004 in Delft.)
In large roundabouts tp is close to the value 1.8 s suggested by Hagring (1998b). In small
roundabouts tp is close to 2.0 s used in the Australian guidelines (Austroads 1993) and
in Sidra (Akçelik 1998). The value 2.1 s used in the German guidelines (FGSV 2001)
is slightly higher.
CONCLUSIONS AND RECOMMENDATIONS
8 CONCLUSIONS AND RECOMMENDATIONS

8.1 Limitations of simulation experiments
Simulation tests reported above have been used to test analytical models against models
with more advanced driver dynamics. The simulation software, Hutsim, has been
calibrated for Finnish unsignalized intersections, as described in Section 6.2. However,
before making suggestions about capacity analysis methodologies it is necessary to
briefly discuss the limitations of the simulation models.
The headway distribution used in the simulation models was shifted exponential. The
distribution at specific locations is modified by target speed distributions, vehicular
interactions, and roadway conditions. In the simulation models the distances from
traffic generators to the intersection was short (less than one kilometer), which limited
the effects of vehicular interactions.
In Hutsim safety lag is the key parameter in gap acceptance. Drivers were assumed to be
consistent and homogeneous. For a given safety lag at a yield-controlled intersection
the critical gap was larger at high major-stream flow rates, when most vehicles had
to stop at the yield line. However, studies on driver behavior have indicated that at
high major-stream flow rates the critical gap may be shorter, not longer, than at low
major-stream flow rates (see Section 2.4).
Minor-stream vehicles in Hutsim cannot anticipate the acceleration or deceleration of
major-stream vehicles. To some extent this may be realistic, but it is not validated.
With a given safely lag critical gaps increase when major-stream vehicles decelerate.
To minimize this effect all right-turn movements were eliminated from the simulation
models.
On the other hand, Rank 3 (or 4) vehicles may enter the intersection while a Rank 2 (or
3) queue starts to discharge, because the vehicles cannot anticipate the acceleration of
major-stream vehicles. This property of Hutsim was demonstrated and its effect was
discussed above. In the Rank 3 and Rank 4 cases analyzed the number of “sneakers”
was not higher than 20–25 pc/h.
The shifted exponential capacity model (3.17) provided an excellent fit for the simu-
lated Rank 2 capacities. These are, however, “tailor-made” capacity estimates for the
simulation model. The main objective of the simulation experiments was the analysis of
Rank3 and Rank 4 capacities. These results were consistent with the theory presented
in this report.
8.2 Unsignalized intersections

Gap acceptance is still the dominating theory in the capacity analysis of unsignalized
intersections. It is also the methodology suggested for the analysis of Finnish unsig-
nalized intersections. This section describes the methodology for three and four-leg
intersections. The capacity analysis of roundabouts is discussed in the next section.
The theory of capacity and delays at unsignalized intersections have been described
in Chapters 3 and 4. These chapters present the theoretical background required to
manually perform operational analysis of unsignalized intersections and to understand
and evaluate published methodologies and capacity analysis software. For Rank 3 and
Rank 4 capacity analysis a new theoretical approach has been developed.
In Chapter 5 three analysis methods were described: HCM2000, Capcal 2 and DanKap.
All of them have been implemented as computer software. Of these three HCM2000
is the most advanced. It has procedures for shared lanes, flared lanes, heavy vehicles,
pedestrians, two-stage gap acceptance, and upstream signals.

The simulation studies in Section 6.3 indicated that capacity estimates based on the shif-
ted exponential headway distribution give good approximations of potential capacity.
Figure 8.1 displays a comparison of capacity models based on shifted exponential and
Tanner’s headway models. The minimum headway estimate for the shifted exponential
models was about one second, which has been used in the figure. It is compared with
Tanner’s capacity (3.24) model with 1.8 second follower headways, as in the current
Swedish and Finnish models. It is evident that Tanner’s model is an appropriate model.
With realistic parameters (estimated from the shifted exponential model based on sim-
ulation results) Tanner’s model does not overestimate the simulated capacity as much
as the negative exponential model (see Fig. 6.8).
1600
4s
1400
1200
5s
Capacity, C/(pc/h)
1000
6s
800 7s
600
400
200
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000
Figure 8.1: Potential capacity for priority-stream headways following Tanner’s distribution
(solid blue) and the shifted exponential distribution (dashed red). Minimum headways are 1.8 s
for the Tanner model and 1 s for the shifted exponential model.
The simulation studies indicated that the new theoretical model for Rank 3 and Rank
4 headways gives good approximations of simulated capacities. The assumption of a
1.8 s minimum discharge headway (Tanner’s model) in Rank 1 streams gives slightly
better capacity estimates than the negative exponential model. For Rank 4 capacity the
conventional (HCM2000) estimates may be slightly better than the theoretical model.
The new theoretical model with a 1.8 second discharge headway for Rank 1 streams
is suggested for the analysis of Finnish unsignalized. Compared with the methods in
HCM2000 and Capcal 2 the new method has several advantages:
1. Potential capacity estimates are in better agreement with simulation results than
the estimates of HCM2000. For Rank 2 capacity analysis the method is the same
as in Capcal 2.
2. Rank 3 capacity estimates are in better agreement with the simulation results
than the estimates of HCM2000 and Capcal 2.
3. The Rank 4 capacity estimates of Capcal 2 are too low. The new estimates are sim-
ilar to the HCM2000 estimates. Compared wit simulated capacities HCM2000
estimates are slightly too low, while the new estimates are slightly too high.
4. The new method has a more solid theoretical foundation, and it is not—at least
not to the same extent as the HCM2000 method—based on cancellation of errors.
Consequently, it is suggested that the capacity of Finnish unsignalized intersections is

estimated following equation (3.113), where the follower headway of Rank 1 streams
is 1.8 seconds. The method is described in more detail in Chapter 3. Priority streams
are those streams or substreams having an actual conflict with the minor stream, as
discussed in Section 2.2. The critical gaps in Table 5.17 are used as base values.
However, better results can always be obtained with field-measured values.
The effects of heavy vehicles were not analyzed in this research. For signalized inter-
sections the adjustment factors of Capcal 2 (SNRA 1995a) were suggested (Luttinen
& Nevala 2002). The same adjustment factors are suggested for unsignalized intersec-
tions.
Control delays can be estimated using the coordinate-transformation method with equa-
tions 4.131 and 4.132, as described in Chapter 4. Acceleration delay is five seconds for
stop-controlled streams. For yield-controlled streams the simple estimate in equation
6.13 can be applied.
8.3 Roundabouts
The capacity of a roundabout entry can be estimated using the method (3.58) of Hagring
(1998b), but assuming a shifted exponential headway distribution (φ = 1) in the major
flow. The capacity equation for entry lane k is

3600e− i=1 γi (tc,i −tp ) ! qi
n n
Ck = n , (8.1)
1 − e− i=1 tf γi i=1
3600γi
where n is the number of circulating lanes (1 or 2), tf is the follow-up time in entry
lane k, tc,i is the critical gap in circulating lane i for the vehicle in
entry lane k, tp is
the minimum (platoon) headway in a circulating lane, and = ni=1 γi . The scale
parameters are estimated as
qi
γi = . (8.2)
3600 − qi tp
For a one-lane roundabout the equation simplifies to
qe−γ (tc −tp )

Ck = , (8.3)
1 − e−γ tf
the potential capacity (3.17) with major-stream headways following the shifted expo-
nential distribution.
The exiting flows are assumed to have no effect of capacity. However, if local conditions
indicate such an effect, the major flow in the outer circulating lane can be expressed as
q1 = pe qe + qc,1 , (8.4)
where qc,1 is the actual circulating flow, qe is the exiting flow rate, and pe is the
proportion of exiting flow included in the major flow. As a default pe = 0.
Base values for critical gaps and follow-up times can be obtained from Table 7.2 on
page 181. For one-lane roundabouts the follow-up time tf and the minimum headway
tp can be estimated as functions of the central island diameter (d):
tf = 2.5 − 0.0067(d − 8), 8 ≤ d ≤ 40 (8.5)

tp = 2.0 − 0.0067(d − 8), 8 ≤ d ≤ 40. (8.6)
Figure 8.2 displays capacity estimates for one-lane roundabouts. At low circulating
flow rates the capacity estimates are slightly higher than in the current guidelines (Fig.
5.5). When circulating flows are high, the new capacity estimates are lower than the
current estimates. Better accuracy can be obtained, if the parameters are estimated
using field data.
1600
10m
40m
1400
1200
1000
800
600
400
200
0
0 500 1000 1500
Figure 8.2: Capacity of Finnish one-lane roundabouts with central island diameters 10 m and
40 m
There was not enough field data to calibrate the Hutsim simulation model for two-
lane roundabouts. The methodology suggested is based on earlier simulation studies
(Hagring 1997, Pursula et al. 1997), Swedish research (Hagring 1998b, Hagring et al.
2003) as well as on generalizations of the results for one-lane roundabouts.
At two-lane roundabouts the vehicles circulating in the inner lane are included in the
major flow of the right entry lane. If the distribution of traffic between outer and inner
lanes is not known, it can be approximated following the results of Hagring et al. (2003).
As a default it is assumed that the outer lane is used by 55 % of left-turn vehicles, 95 %
of through vehicles, and 100 % of right-turn vehicles. Lane changes from the inner
lane to the outer lane are assumed to take place in the weaving area preceding the exit,
so that they can be ignored in the capacity analysis. Especially under high degrees of
saturation the realism of the lane allocation should be checked.
Figure 8.3 displays capacity estimates for a two-lane roundabout when 50 % or 90 % of
circulating traffic is in the outer lane. The entry capacity is expressed as the sum of the
capacities of both entry lanes. In practice, capacity should be calculated separately for
both entry lanes. In most cases the demand on the left entry lane is much lower than
on the right lane. New research results in Germany indicate that the increase in entry
capacity due to a second entry lane is only about about 14 percent.1
Figure 8.4 displays capacity, when both circulating and entering traffic have the same
lane distribution. If the capacity of lane 1 is C1 , the capacity of lane 2 is C2 , the
proportion of traffic on the lane 1 is p, the entry
−1capacity in Figure 8.3 is calculated as
−1
C = C1 + C2 and in Figure 8.4 as C = min p C1 , (1 − p) C2 . When arrival rate
is higher than C, demand exceeds capacity in one or both entry lanes. The increase in
capacity due to an additional entry lane is highest when the lane distribution is even.
3000
2500
2000
1500
1000
← 50
500
← 90
0
0 500 1000 1500 2000
Figure 8.3: Capacity of Finnish two-lane roundabouts with 50 % and 90 % of circulating traffic
in the outer lane
A change in the traffic flows at a subject entry may have an effect on the operation of
other entries, which again may affect the traffic conditions at the subject entry. This is
important especially if the increase of traffic at the subject entry causes oversaturation
at another entry, in which case circulating traffic at the subject entry may decrease.
The method presented here addresses only the capacity of entering flow. No data
is available on the capacity at exits and weaving sections. The German exit capacity
estimate of 1,200 veh/h can be used as a rule of thumb. The effect of crossing pedestrians
at entries and exits can be estimated by the procedure presented in Section 3.8. Traffic
operation at weaving sections can be analyzed using simulation models.
When the capacity of a roundabout has been calculated, other performance measures can
be estimated following the the same procedures as for other unsignalized intersections.
1 Thorsten Miltner at the German–Dutch–Finnish Seminar on Traffic Engineering at 8–9 June 2004 in
Delft. In HBS 2001 (FGSV 2001) adding a second entry lane doubles the capacity.
3000
2500
2000
1500
1000
← 50
500 ← 50
← 90
← 90
0
0 500 1000 1500 2000
Figure 8.4: Capacity of Finnish two-lane roundabouts with 50 % and 90 % of circulating and
entering traffic in the outer lane (solid curves) and with a one-lane entry (dashed curves)
8.4 Level of service

Current level-of-service criteria in Finland (Table 5.14) follow the 1985 HCM
(Transportation Research Board 1985). Reserve capacity is used as a service meas-
ure. In HCM2000 (Transportation Research Board 2000) and in the German HBS 2001
(FGSV 2001) level of service is based on control delay (Table 5.1). Because delay is a
measure experienced directly by road users it should be preferred as a service measure
over the more engineering-oriented reserve capacity.
Table 8.1 displays the current thresholds for reserve capacity with a respective delay
estimate (4.68). The new level-of-service criteria for control delay have been obtained
from HCM2000. Table 8.1 shows that the delay thresholds derived from the reserve
capacity and the suggested new criteria are similar. Setting LOS C threshold to 20 s
would make it closer to current criterion, but that would make the range of traffic
conditions described by LOS C very limited. The German manual (HBS 2001) has
an even higher (30 s) threshold. Thus, the control delays in column “New criteria” of
Table 8.1 are suggested as the new level-of-service criteria in Finland.
8.5 Further research

This research was started assuming that only a short theoretical description and a few
field measurements to adjust current estimates of critical gaps and follow-up times and
to calibrate the parameters of Hutsim were required to make a suggestion for new
analysis procedures. It appeared, as it often does, that reality was more complicated.
The research revealed some answers, but raised even more questions. Some of these
questions are described here as items for further research.
Field studies were used to calibrate Hutsim parameters. The estimation of critical gaps
and follow-up times, however, requires more field studies at highly saturated conditions.
Table 8.1: Current and new level-of-service criteria with approximate delays at reserve-capacity
thresholds
Current criteria New criteria
LOS Reserve capacity Delay Delay
(veh/h) (s) (s)
A ≥ 400 ≤ 9 ≤ 10
B ≥ 300 ≤ 12 ≤ 15
C ≥ 200 ≤ 18 ≤ 25
D ≥ 100 ≤ 36 ≤ 35
E ≥ 0 > 36 ≤ 50
F ρ>1 ρ>1 > 50
Besides adjusting the base values the studies should consider
1. The correlation of critical gaps and follow-up times
2. Possible flow dependency of critical gaps and follow-up times
3. The effects of intersection and roundabout geometry.
The maximum likelihood method can be suggested as the statistical estimation method.
Field studies are also required to further calibrate and validate the Rank 3 and Rank 4
capacity analysis procedures.
In order to evaluate the impedance effect of pedestrians, a description of pedestrian-
vehicle interactions and pedestrian flow dynamics at intersections is needed. This
requires new field studies.
At two-lane roundabouts field studies are required to estimate the lane allocation. This
information is needed in the capacity analysis.
The movement capacity equations for Rank 3 and Rank 4 steams were presented as-
suming Poisson arrivals. Extension of these models to Cowan’s M3 distribution would
give additional flexibility and realism.
The simulation experiments revealed some problems in the simulation software. The
most important problem is the lack of anticipation of vehicle acceleration and deceler-
ation in the gap acceptance process. Other, mostly minor, details have been reported
to the software developer.
REFERENCES
REFERENCES
Aagaard, P. 1998. Danish capacity models. In NORDKAP—NORDiskt KAPacitetssamarbete
[NORDKAP—Nordic capacity cooperation] (University of Lund 1998). Sammandrag av
seminarium på Grand Hotel i Lund, Sverige 2–3 juni 1997.
Aagaard, P. E. 1995. Recent research into capacity in Danish roundabouts. Paper No. 950080,
Transportation Research Board 74th Annual Meeting Preprint.
Aakre, A. 1998. Norwegian methods for calculation of capacity and level of service at intersec-
tions. In NORDKAP—NORDiskt KAPacitetssamarbete [NORDKAP—Nordic capacity
cooperation] (University of Lund 1998). Sammandrag av seminarium på Grand Hotel i
Lund, Sverige 2–3 juni 1997.
Adams, W. F. 1936. Road traffic considered as a random series. Journal of the Institution of
Civil Engineers, vol. 4, no. 1, p. 121–130.
Akçelik, R. 1994a. Gap-acceptance modelling by traffic signal analogy. Traffic Engineering
and Control, vol. 35, no. 9, p. 498–506.
Akçelik, R. 1997. Analysis of roundabout performance by modeling approach-flow interac-
tions. In Kyte (1997). P. 15–25.
Akçelik, R. 1998. Roundabouts: Capacity and Performance Analysis. (Research Report
ARR–321). Vermont South, Victoria: ARRB Transport Research.
Akçelik, R. & Chung, E. 1994. Calibration of the bunched exponential distribution of arrival
headways. Road & Transport Research, vol. 3, no. 1, p. 42–59.
Akçelik, R. (ed.) 1994b. Second International Symposium on Highway Capacity. Australian
Road Research Board, Transportation Research Board, Sydney.
Akçelik, R. & Troutbeck, R. 1991. Implementation of the Australian roundabout analysis
method in SIDRA. In Brannolte (1991), p. 17–34.
Allan, R. R. 1968. On the highway crossing problem. In ARRB Proceedings, Vol. 4. Victoria:
Australian Road Research Board. P. 529–548.
Anveden, J. 1988. Swedish research on unsignalized intersections. In Brilon (1988b), p. 80–
102.
Armitage, D. J. & McDonald, M. 1974. Roundabout capacity. Traffic Engineering and
Control, vol. 15, no. 10, p. 812–815.
Ashworth, R. 1969. The capacity of priority-type intersections with a non-uniform distribution
of critical acceptance gaps. Transportation Research, vol. 3, p. 273–278.
Ashworth, R. & Field, J. C. 1973. The capacity of rotary intersections. Journal of the
Institution of Highway Engineers, vol. 20, no. 3, p. 14–21.
Ashworth, R. & Laurence, C. J. D. 1978. A new procedure for capacity calculation at
conventional roundabouts. Proceedings of the Institution of Civil Engineers, Part 2, vol. 65,
p. 1–16.
Austroads 1988. Intersections at Grade. (Guide to Traffic Engineering Practice—Part 5).
Sydney.
Austroads 1993. Roundabouts. (Guide to Traffic Engineering Practice—Part 6). Sydney.
Baass, K. G. 1987. The potential capacity of unsignalized intersections. ITE Journal, vol. 57,
no. 10, p. 43–46.
Beckmann, M., McGuire, C. B. & Winsten, C. B. 1956. Studies in the Economics of
Transportation. New Haven: Yale University Press.
Bergh, T. 1991. Intersections without traffic signals—Swedish experience on capacity and
traffic safety. In Brilon (1991), p. 192–213.
REFERENCES
Blackmore, F. C. 1963. Priority at roundabouts. Traffic Engineering and Control, vol. 4, no. 6,
p. 104–106.
Blackmore, F. C. 1970. Capacity of single-level intersections. (RRL Report LR 356).
Crowthorne: Road Research Laboratory.
Bonneson, J. A. & Fitts, J. W. 1997. Delay to major-street through vehicles at two-way
stop-controlled intersections. In Kyte (1997), p. 45–54.
Bonneson, J. A. & Fitts, J. W. 1999. Delay to major street through vehicles at two-way
stop-controlled intersections. Transportation Research, vol. 33A, no. 3/4, p. 237–253.
Brannolte, U. (ed.) 1991. Highway Capacity and Level of Service. Rotterdam: Balkema.
Branston, D. 1976. Models of single lane time headway distributions. Transportation Science,
vol. 10, no. 2, p. 125–148.
Bratley, P., Fox, B. L. & Schrage, L. E. 1987. A Guide to Simulation. Second ed. New York:
Springer Verlag.
Breiman, L. 1963. The Poisson tendency in traffic distribution. The Annals of Mathematical
Statistics, vol. 32, p. 308–311.
Brennan, M. J. & Fitzgerald, M. P. 1979. A new look at Tanner’s formula for the capacity
of a minor road at an uncontrolled intersection. Traffic Engineering and Control, vol. 20,
no. 4, p. 220.
Briggs, T. 1977. Time headways on crossing the stop-line after queueing at traffic lights. Traffic
Engineering and Control, vol. 18, no. 5, p. 264–265.
Brilon, W. 1988a. Recent developments in calculation methods for unsignalized intersections
in West Germany. In Intersections without Traffic Signals (Brilon 1988b), p. 111–153.
Brilon, W. (ed.) 1988b. Intersections without Traffic Signals. Berlin: Springer Verlag.
Brilon, W. (ed.) 1991. Intersections without Traffic Signals II. Berlin: Springer Verlag.
Brilon, W. (ed.) 2000. Fourth International Symposium on Highway Capacity Proceedings.
(Transportation Research Circular E-C018). Washington, D.C.: Transportation Research
Board.
Brilon, W. & Großmann, M. 1991. The new German guideline for capacity of unsignalized
intersections. In Brannolte (1991), p. 63–73.
Brilon, W., Großmann, M. & Blanke, H. 1994. Verfahren für die Berechnung der Leistungs-
fähigkeit und Qualität des Verkehrsablaufes auf Straßen [Calculation methods for capacity
and quality of traffic flow on roads]. (Forschung Straßenbau und Straßenverkehrstechnik
669). Bonn: Bundesministerium für Verkehr.
Brilon, W., Koenig, R. & Troutbeck, R. J. 1997. Useful estimation procedures for critical
gaps. In Kyte (1997), p. 71–87.
Brilon, W. & Miltner, T. 2002. Verkehrsqualität unterschiedlicher Verkehrsteilnehmerarten
an Knotenpunkten ohne Lichtsignalanlage [Quality of traffic for different modes of travel
at unsignalized intersections]. (Slussbericht). Bochum: Ruhr-Universität Bochum, Lehr-
stuhl für Verkehrswesen. Forschungsauftrag FE-Nr.: 77.434/1999/ des Bundesministeri-
ums für Verkehr, Bau- und Wohnungswesen.
Brilon, W. & Stuwe, B. 1993. Capacity and design of traffic circles in Germany. In Traffic
Flow and Highway Capacity. (Transportation Research Record 1398). Washington, D.C.:
Transportation Research Board. P. 61–67.
Brilon, W. & Wu, N. 1999. Capacity at unsignalized two-stage priority intersection. Trans-
portation Research, vol. 33A, no. 3/4, p. 275–289.
Brilon, W. & Wu, N. 2001. Capacity at unsignalized intersections derived by conflict technique.
In Traffic Flow and Highway Capacity 2001. (Transportation Research Record 1776).
Washington, D.C.: Transportation Research Board. P. 82–90.
REFERENCES
Brilon, W. & Wu, N. 2002. Unsignalized intersections—a third method for analysis. In Taylor
(2002), p. 157–178.
Brilon, W., Wu, N. & Bondzio, L. 1997. Unsignalized intersections in Germany—a state of
the art 1997. In Kyte (1997), p. 61–70.
Brilon, W., Wu, N. & Lemke, K. 1996. Capacity at unsignalized two-stage priority intersec-
tion. In Highway Capacity Analysis for Interrupted and Uninterrupted Flow Facilities.
(Transportation Research Record 1555). Washington, D.C.: Transportation Research
Board. P. 74–82.
Brown, M. 1995. The Design of Roundabouts. London: HMSO.
Buckley, D. J. 1962. Road traffic headway distributions. In Proceedings, Vol. 1, part 1.
Victoria: Australian Road Research Board. P. 153–187.
Buckley, D. J. 1968. A semi-Poisson model of traffic flow. Transportation Science, vol. 2,
no. 2, p. 107–133.
Bunker, J. & Troutbeck, R. 1994a. The Theory of Limited Priority Merging. (Research Report
94-4). Brisbane: Physical Infrastructure Centre, Queensland University of Technology.
Bunker, J. & Troutbeck, R. J. 1994b. Analysis of Limited Priority Theoretical Equations.
(Research Report 94-07). Brisbane: Physical Infrastructure Centre, Queensland Univer-
sity of Technology.
Bureau of Public Roads 1950. Highway Capacity Manual. Washington, D.C.: U.S. Depart-
ment of Commerce.
Catchpole, E. A. & Plank, A. W. 1986. The capacity of a priority intersection. Transportation
Research, vol. 20B, no. 6, p. 441–456.
Catling, I. 1977. A time-dependent approach to junction delays. Traffic Engineering and
Control, vol. 18, no. 11, p. 520–523, 526.
Çinlar, E. 1972. Superposition of stochastic processes. In P. A. W. Lewis (ed.), Stochastic Point
Processes: Statistical Analysis, Theory, and Applications. NewYork: Wiley-Interscience.
P. 549–606.
Clayton, A. J. H. 1941. Road traffic calculations. Journal of the Institution of Civil Engineers,
vol. 16, no. 7, p. 247–284. With discussion and correspondence in no. 8, pp. 588–594.
Clayton, A. J. H. 1945. The capacity of roundabouts at road intersections. Journal of the
Institution of Civil Engineers, vol. 23, no. 3, p. 149–154.
Clayton, A. J. H. 1955. Working capacity of roads. Proceedings of the Institution of Civil
Engineers, Part 2, vol. 4, no. 3, p. 652–696. With discussion.
Cohen, J., Dearnaley, E. J. & Hansel, C. E. M. 1955. The risk taken in crossing a road.
Operations Research Quarterly, vol. 6, p. 120–128.
Cooper, R. B. 1981. Introduction to Queueing Theory. Second ed. London: Edward Arnold
(Publishers) Limited.
Cowan, R. 1987. An extension of Tanner’s result on uncontrolled intersections. Queueing
Systems, vol. 1, p. 249–263.
Cowan, R. J. 1975. Useful headway models. Transportation Research, vol. 9, no. 6, p. 371–375.
Cowan, R. J. 1984. Adams’ formula revised. Traffic Engineering and Control, vol. 25, no. 5,
p. 272–274.
Cox, D. R. & Isham, V. 1980. Point Processes. London: Chapman and Hall Ltd.
Cox, D. R. & Miller, H. D. 1965. The Theory of Stochastic Processes. London: Chapman
and Hall.
Cox, D. R. & Smith, W. L. 1954. On the superposition of renewal processes. Biometrika,
vol. 41, p. 91–99.
REFERENCES
Crowder, M. J., Kimber, A. C., Smith, R. L. & Sweeting, T. J. 1991. Statistical Analysis of
Reliability Data. London: Chapman and Hall Ltd.
Daganzo, C. F. 1977. Traffic delay at unsignalized intersections: Clarification of some issues.
Transportation Science, vol. 11, no. 2, p. 180–189.
Daganzo, C. F. 1997. Fundamentals of Transportation and Traffic Operations. Oxford:
Pergamon.
Darzentas, J. 1981. Gap acceptance: Myth and reality. In V. F. Hurdle, E. Hauer & G. N.
Steuart (ed.), Proceedings of the Eighth International Symposium on Transportation
and Traffic Theory. Toronto: University of Toronto Press. P. 175–192.
Dawson, R. F. & Chimini, L. A. 1968. The hyperlang probability distribution—A generalized
traffic headway model. In Characteristics of Traffic Flow. (Highway Research Record
230). Washington, D.C.: Highway Research Board. P. 1–14.
de Smit, J. H. A. 1971. The transient behaviour of the queue at a fixed cycle traffic light.
Transportation Research, vol. 5, no. 1, p. 1–14.
Drew, D. R. 1967. Gap acceptance characteristics for ramp-freeway surveillance and con-
trol. In Freeway Traffic Characteristics and Control. (Highway Research Record 157).
Washington, D.C.: Highway Research Board. P. 108–143. With discussion.
Drew, D. R. 1968. Traffic Flow Theory and Control. New York: McGraw-Hill.
Duff, J. T. 1968. Traffic management. In E. Davies (ed.), Traffic Engineering Practice. Second
ed. London: E. & F. N. Spon Ltd. Chapter 5, p. 102–128.
Dunne, M. C. & Buckley, D. J. 1972. Delays and capacities at unsignalized intersections. In
ARRB Proceedings, Vol. 6. Victoria: Australian Road Research Board. P. 345–362.
Edwards, A. W. F. 1974. The history of likelihood. International Statistical Review, vol. 42,
p. 9–15.
Edwards, A. W. F. 1992. Likelihood. Expanded ed. Baltimore: The Johns Hopkins University
Press.
Erlang, A. K. 1909. Sandsynlighedsregning og telefonsamtaler [The theory of probabilities
and telephone conversations]. Nyt Tidskrift Matematik B, vol. 20, p. 33–39.
Erlang, A. K. 1917a. Løsning af nogle problemer fra sandsynlighedsregningen af betydning for
de automatiske telefoncentraler [Solution of some problems in the theory of probabilities
of significance in automatic telephone exchanges]. Elektroteknikeren, vol. 13, no. 1,
p. 5–13. Also in English (Erlang 1917b).
Erlang, A. K. 1917b. Solution of some problems in the theory of probabilities of significance
in automatic telephone exchanges. The Post Office Electrical Engineers’ Journal, vol. 10,
p. 189–197.
Evans, D. H., Herman, R. & Weiss, G. H. 1964. The highway merging and queuing problem.
Operations Research, vol. 12, no. 6, p. 832–857.
Feller, W. 1957. An Introduction to Probability Theory and Its Applications. Vol. 1. Second
ed. New York: John Wiley & Sons.
Feller, W. 1966. An Introduction to Probability Theory and Its Applications. Vol. 2. New
York: John Wiley & Sons.
FGS 1972. Berechnung der Leistungsfähigkeit von Knotenpunkten ohne Lichtsignalanlagen
[Capacity calculation for unsignalized intersections]. Köln: Forschungsgesellschaft für
das Straßenwesen.
FGSV 2001. Handbuch für die Bemessung von Straßenverkehrsanlagen [Handbook for assess-
ing road traffic facilities]. Köln: Forschungsgesellschaft für Straßen- und Verkehrswesen.
HBS 2001.
REFERENCES
Fisher, R. B. 1960. A preliminary study of acceptance distributions for merging traffic. M.

Tech. Thesis. Sydney: School of Traffic Engineering, University of N.S.W.
Fisk, C. S. 1989. Priority intersection capacity: A generalization of Tanner’s formula. Trans-
portation Research, vol. 23B, no. 4, p. 281–286.
Fisk, C. S. & Tan, H. H. 1989. Delay analysis for priority intersections. Transportation
Research, vol. 23B, no. 6, p. 453–469.
Gani, J. 1969. Recent advances in storage and flooding theory. Advances in Applied Probability,
vol. 1, p. 90–110.
Gartner, N. H. & Messer, C. J. (ed.) 1997. Monograph on Traffic Flow Theory. Federal
Highway Administration, http://www.tfhrc.gov/its/tft/tft.htm, 08/03/2001.
Garwood, F. 1940. An application of the theory of probability to the operation of vehicular-
controlled traffic signals. Journal of the Royal Statistical Society, vol. 7, no. 1, p. 65–77.
Gazis, D. C., Newell, G. F., Warren, P. & Weiss, G. H. 1967. The delay problem for crossing
an n lane highway. In L. C. Edie, R. Herman & R. Rothery (ed.), Vehicular Traffic
Science. (Proceedings of the Third International Symposium on the Theory of Traffic
Flow, New York, June 1965). New York: Elsevier. P. 267–279.
Gazis, D. C. & Potts, R. B. 1965. The over-saturated intersection. In J. Almond (ed.),
Proceedings of The Second International Symposium on The Theory of Road Traffic Flow.
Paris: The Organisation for Economic Co-operation and Development. P. 221–237.
Gipps, P. G. 1982. Some modified gap acceptance formulae. In Traffic Engineering. (Proceed-
ings, Vol. 11, Part 4). Melbourne: Australian Road Research Board. P. 71–75.
Glynn, P. W. 1990. Diffusion approximations. In D. P. Heyman & M. J. Sobel (ed.), Stochastic
Models. Vol. 2 of Handbooks in Operations Research and Management Science. Ams-
terdam: North Holland. Chapter 4, p. 145–198.
Gnedenko, B. V. 1962. The Theory of Probability. New York: Chelsea Publishing Company.
Translated from the Russian by B. D. Seckler.
Gnedenko, B. V. 1967. The Theory of Probability. Fourth ed. New York: Chelsea Publishing
Company. Translated from the Russian by B. D. Seckler.
Goodman, R. 1988. Introduction to Stochastic Models. Menlo Park, Ca.: The Ben-
jamin/Cummings Publishing Company, Inc.
Grabe, W. 1954. Leistungsermittlung von nicht lichtsignalsteuerten Knotenpunkten des
Straßenverkehrs [Capacity calculation of unsignalized intersections in road traffic].
(Forschungsarbeiten aus dem Strassenwesen, Neue Folge 11). Bielefeld: Kirschbaum
Verlag.
Greenshields, B. D., Shapiro, D. & Ericksen, E. L. 1947. Traffic Performance at Urban
Street Intersections. (Technical Report 1). New Haven, Connecticut: Yale Bureau of
Highway Traffic.
Greenshields, B. D. & Weida, F. M. 1952. Statistics with Applications to Highway Traffic
Analyses. Saugatuck, Connecticut: The Eno Foundation for Highway Traffic Control.
Grimmett, G. & Welsh, D. 1986. Probability: An Introduction. Oxford: Clarendon Press.
Grimmett, G. R. & Stirzaker, D. R. 2001. Probability and Random Processes. Oxford:
Oxford University Press.
Gross, D. & Harris, C. M. 1974. Fundamentals of Queueing Theory. New York: John Wiley
& Sons.
Gross, D. & Harris, C. M. 1998. Fundamentals of Queueing Theory. Third ed. New York:
John Wiley & Sons.
REFERENCES
Grossmann, M. 1991. Aktualisiertes Berechnungsverfahren für Knotenpunkte ohne Licht-

signalanlagen [Updated calculation procedure for intersections without traffic signals].
(Strassenbau und Strassenverkehrstechnik 596). Bonn: Bundesminister für Verkehr.
Guichet, B. 1997. Roundabouts in France: Development, safety, design, and capacity. In Kyte
(1997), p. 100–105.
Hagring, O. 1996a. Roundabout Entry Capacity. (Bulletin 135). Lund: Tekniska Högskolan
i Lund, Institutionen för Trafikteknik.
Hagring, O. 1996b. The use of the Cowan M3 distribution for modelling roundabout flow.
Traffic Engineering and Control, vol. 37, no. 5, p. 328–332.
Hagring, O. 1997. Capcal2—A new version of the SNRA capacity, delay, and VOC software.
In Kyte (1997), p. 115–123.
Hagring, O. 1998a. A further generalization of Tanner’s formula. Transportation Research,
vol. 32B, no. 6, p. 423–429.
Hagring, O. 1998b. Vehicle-vehicle Interactions at Roundabouts and their Implications for
the Entry Capacity: A Methodological Study with Applications to Two-lane Roundabouts.
(Bulletin 159). Lund: Department of Traffic Planning and Engineering, Lund Institute of
Technology.
Hagring, O. 2000. Effect of OD flows on roundabout entry capacity. In Brilon (2000), p. 434–
445.
Hagring, O. 2001. Derivation of capacity equation for roundabout entry with mixed circulating
and exiting flows. In Traffic Flow and Highway Capacity 2001. (Transportation Research
Record 1776). Washington, D.C.: Transportation Research Board. P. 91–99.
Hagring, O. & Niittymäki, J. 1997. A modelling of roundabout entry operations. VTIs och
KFBs Forskardagar. Linköping, Sweden, January 8–9, 1997. Unpublished.
Hagring, O., Rouphail, N. M. & Sørensen, H. A. 2003. Comparison of capacity models
for two-lane roundabouts. In 82nd Annual Meeting Compendium of Papers CD-ROM.
Transportation Research Board. Washington, D.C.
Haight, F. A. 1963. Mathematical Theories of Traffic Flow. New York: Academic Press.
Haight, F. A., Bisbee, E. F. & Wojcik, C. 1962. Some mathematical aspects of the problem of
merging. In Theory of Traffic Flow. (Bulletin 356). Washington, D.C.: Highway Research
Board. P. 1–14.
Haight, F. A. & Breuer, M. A. 1960. The Borel-Tanner distribution. Biometrika, vol. 47,
no. 1 and 2, p. 143–150.
Hakkert, A. S., Mahalel, D. & Asante, S. A. 1991. A comparative study of roundabout
capacity procedures. In Brilon (1991), p. 93–106.
Häkli, O. 1971. Liikenteenvälityskyky [Traffic capacity]. In Tekniikan käsikirja 6. Eighth ed.
Jyväskylä: K.J. Gummerus Oy.
Hammond, H. F. & Sorenson, L. J. (ed.) 1941. Traffic Engineering Handbook. New York:
Institute of Traffic Engineers and National Conservation Bureau.
Hanson, A. 1978. Swedish Capacity Manual: Part 2. Capacity of unsignalized intersections. In
Highway Capacity, Measures of Effectiveness, and Flow Theory. (Transportation Research
Record 667). Washington, D.C.: Transportation Research Board. P. 4–11.
Harders, J. 1968. Die Leistungsfähigkeit nicht signalregelter städtischer Verkshrsknoten [Ca-
pacity of unsignalized urban intersections]. (Strassenbau und Strassenverkehrstechnik 76).
Bonn: Bundesminister für Verkehr.
Hawkes, A. G. 1966. Delay at traffic intersections. Journal of the Royal Statistical Society,
vol. 28B, p. 202–212.
REFERENCES
Hawkes, A. G. 1968. Gap-acceptance in road traffic. Journal of Applied Probability, vol. 5,

p. 84–92.
Heidemann, D. 1991. Queue length and waiting-time distributions at priority intersections.
Transportation Research, vol. 25B, no. 4, p. 163–174.
Heidemann, D. 2002. Mathematical analysis of non-stationary queues in traffic flow with
particular consideration of the coordinate transformation technique. In Taylor (2002),
p. 675–695.
Herman, R. & Weiss, G. 1961. Comments on the highway-crossing problem. Operations
Research, vol. 9, no. 6, p. 828–840.
Hewitt, R. H. 1985. A comparison between some methods of measuring critical gap. Traffic
Engineering and Control, vol. 26, no. 1, p. 13–21.
Highway Research Board 1965. Highway Capacity Manual. (Special Report 87). Wash-
ington, D.C.: National Research Council.
Hobbs, F. D. 1974. Traffic Planning and Engineering. Oxford: Pergamon Press.
Høyland, A. & Rausand, M. 1994. Systems Reliability Theory: Models and Statistical
Methods. New York: John Wiley & Sons.
Hurdle, V. F. 1984. Signalized intersection delay model—A primer for the uninitiated. In
Traffic Capacity and Characteristics. (Transportation Research Record 971). Washington,
D.C.: Transportation Research Board. P. 96–105.
Jaiswal, N. K. 1968. Priority Queues. New York: Academic Press.
Jessen, G.-D. 1968. Ein Richtlinienvorschlag für die Behandlung der Leistungsfähigkeit von
Knotenpunkten ohne Signalreglerung [A guideline proposal for estimating capacity of
unsignalized intersections]. Straßenverkehrstechnik, vol. 12, no. 7/8, p. 101–105.
Jormalainen, S. 1991. Kiertoliittymät ja niiden välityskyky [Roundabouts and their capacity].
(Tielaitoksen selvityksiä 23/1991). Helsinki: Tielaitos. TIEL 3200022.
Karlin, S. & Taylor, H. M. 1975. A First Course in Stochastic Processes. Second ed. New
York: Academic Press.
Karlin, S. & Taylor, H. M. 1981. A Second Course in Stochastic Processes. Orlando:
Academic Press.
Kehittämiskeskus 1991. CAPCAL-mikroversion käyttöohje [CAPCAL micro version user’s
manual]. Helsinki: Tiehallitus.
Kehittämiskeskus 1996. Kaksikaistaiset kiertoliittymät [Two-lane roundabouts]. (Tielaitok-
sen selvityksiä 47/1996). Helsinki: Tielaitos. TIEL 3200415.
Kendall, D. G. 1951. Some problems in the theory of queues. Journal of the Royal Statistical
Society, vol. 13, no. 2, p. 151–185. With discussion.
Kendall, D. G. 1953. Stochastic processes occurring in the theory of queues and their analysis
by the method of the imbedded Markov chain. The Annals of Mathematical Statistics,
vol. 24, p. 338–354. With discussion.
Khintchine, A. Y. 1960. Mathematical Methods in the Theory of Queueing. London: Charles
Griffin Ltd. Translated by D. M. Andrews and M. H. Quenouille.
Kimber, R. M. 1980. The traffic capacity of roundabouts. (TRRL Laboratory Report 942).
Crowthorne: Transport and Road Research Laboratory.
Kimber, R. M. 1989. Gap-acceptance and empiricism in capacity prediction. Transportation
Science, vol. 23, no. 2, p. 100–111.
Kimber, R. M. & Coombe, R. D. 1980. The traffic capacity of major/minor priority junc-
tions. (TRRL Supplementary Report 582). Crowthorne: Transport and Road Research
Laboratory.
REFERENCES
Kimber, R. M. & Hollis, E. M. 1978. Peak-period traffic delays at road junctions and other
bottlenecks. Traffic Engineering and Control, vol. 19, no. 10, p. 442–446.
Kimber, R. M. & Hollis, E. M. 1979. Traffic queues and delays at road junctions. (TRRL
Laboratory Report 909). Crowthorne: Transport and Road Research Laboratory.
Kimber, R. M., Marlow, M. & Hollis, E. M. 1977. Flow/delay relationships for major/minor
priority junctions. Traffic Engineering and Control, vol. 18, no. 11, p. 516–519.
Kimber, R. M., Summersgill, I. & Burrow, I. J. 1986. Delay processes at unsignalised junc-
tions: The interrelation between geometric and queuing delay. Transportation Research,
vol. 20B, no. 6, p. 457–476.
King, G. F. & Wilkinson, M. 1976. Relationship of signal design to discharge headway,
approach capacity, and delay. In Capacity and Measurement of Effectiveness. (Trans-
portation Research Record 615). Washington, D.C.: Transportation Research Board.
P. 37–44.
Kleijnen, J. P. C. 1987. Statistical Tools for Simulation Practitioners. New York: Marcel
Dekker, Inc.
Kleinrock, L. 1975. Queueing Systems. Volume I: Theory. New York: John Wiley & Sons.
Kleinrock, L. 1976. Queueing Systems. Volume II: Computer Applications. New York: John
Wiley & Sons.
Koivisto, N. 1999. Valo-ohjaamattomien tasoliittymien liikennevirta ja liikenteen simulointi
[Traffic flow of unsignalized at-grade intersections and traffic simulation]. Master’s thesis.
Espoo: Helsinki University of Technology.
Koshi, M. (ed.) 1990. Transportation and Traffic Theory. (Proceedings of the Eleventh In-
ternational Symposium on Transportation and Traffic Theory, held July 18–20, 1990, in
Yokohama, Japan). New York: Elsevier.
Kosonen, I. 1996. HUTSIM—Simulation Tool for Traffic Signal Control Planning. (Transport-
ation Engineering Publication 89). Otaniemi: Helsinki University of Technology.
Kosonen, I. 1999. HUTSIM—Urban Traffic Simulation and Control Model: Principles and
Applications. (Transportation Engineering Publication 100). Espoo: Helsinki University
of Technology.
Kremser, H. 1962. Ein zusammengesetztes Wartezeitproblem bei Poissonschen verkehrsströ-
men [A complex waiting-time problem under Poisson-distributed traffic flows]. Öster-
reichisches Ingenieur-Archiv, vol. 16, no. 3, p. 231–252.
Kremser, H. 1964. Wartezeiten und Warteschlangen bei Einfädelung eines Poissonprozesses
in einen anderen solchen Prozeß [Delays and queues with one Poisson process merging
into another one]. Österreichisches Ingenieur-Archiv, vol. 18, no. 3–4, p. 269–274.
Kyte, M., Dixon, M. & Basavaraju, P. M. 2003. Why field measurements are different than
model estimates—an analysis framework for capacity and level of service analysis of
unsignalized intersections. In 82nd Annual Meeting Compendium of Papers CD-ROM.
Transportation Research Board. Washington, D.C.
Kyte, M. (ed.) 1997. Proceedings of the Third International Symposium on Intersections
Without Traffic Signals. Transportation Research Board, Federal Highway Association,
National Center for Advanced Transportation Technology, University of Idaho, Transport-
ation Northwest, University of Washington, Portland, Oregon U.S.A.
Kyte, M., Tian, Z., Mir, Z., Hameedmansoor, Z., Kittelson, W., Vandehey, M., Robinson,
B., Brilon, W., Bondzio, L., Wu, N. & Troutbeck, R. 1996. Capacity and Level of
Service at Unsignalized Intersections: Two-Way Stop-Controlled Intersections. (NCHRP
Web Document 5). Washington, D.C.: Transportation Research Board. Project 3-46,
Contractor’s Final Report, Vol. 1.
REFERENCES
Kyte, M., Zegeer, J. & Lall, K. B. 1991. Empirical models for estimating capacity and delay
at stop-controlled intersections in the United States. In Brilon (1991), p. 335–361.
Lapierre, R. & Steierwald, G. (ed.) 1987. Verkehsleittechnik für den Straßenverkehr [Traffic
control engineering for road traffic]. Vol. 1. Berlin: Springer-Verlag.
Lertworawanich, P. & Elefteriadou, L. 2003. Determination of storage lengths of left-
turn lanes at unsignalized intersections using M/G2/1 queuing. In 82nd Annual Meeting
Compendium of Papers CD-ROM. Transportation Research Board. Washington, D.C.
Leutzbach, W. & Köhler, U. 1974. Definitions and relationships for three different time
intervals for delayed vehicles. In D. J. Buckley (ed.), Transportation and Traffic Theory.
(Proceedings of the Sixth International Symposium on Transportation and Traffic Theory.
University of New South Wales, Sydney, Australia, 26–28 August 1974). New York:
Elsevier. P. 87–103.
Liimatainen, A. 1997. Kiertoliittymät Suomessa [Roundabouts in Finland]. In Tie- ja liiken-
netekniikan teemapäivät 26.-27.11.1997. Tielaitos, tie- ja liikennetekniikka. Helsinki.
Unpublished.
Little, J. D. C. 1961. A proof of the queuing formula L = λW . Operations Research, vol. 9,
no. 3, p. 383–387.
Li, W., Wang, W. & Jiang, D. 2003. Unsignalized intersection capacity with mixed vehicle
flows. In 82nd Annual Meeting Compendium of Papers CD-ROM. Transportation Research
Board. Washington, D.C.
Louah, G. 1986. La capacite des carrefours sans feux: Méthodes et modelès [Capacity of
unsignalized intersections: Methods and models]. Bagneux Cedex: Service de d’Etudes
Techniques des Routes et Autoroutes.
Louah, G. 1988. Recent French studies on capacity and waiting times at rural unsignalized
intersections. In Brilon (1988b), p. 248–262.
Luttinen, R. T. 1990. Johdatus aikavälijakaumien teoriaan [Introduction to the theory of head-
way distributions]. (Julkaisu 71). Otaniemi: Teknillinen korkeakoulu, Liikennetekniikka.
Luttinen, R. T. 1991. Testing goodness of fit for 3-parameter gamma distribution. In Proceed-
ings of the Fourth IMSL User Group Europe Conference. IMSL and CRPE-CNET/CNRS.
Paris. P. B10/1–13.
Luttinen, R. T. 1994. Identification and estimation of headway distributions. In Akçelik
(1994b), p. 427–446.
Luttinen, R. T. 1996. Statistical Analysis of Vehicle Time Headways. (Transportation Engin-
eering Publication 87). Otaniemi: Helsinki University of Technology.
Luttinen, R. T. 1998. Laajennetun T-liittymän välityskyky [Capacity at a two-stage T-
intersection]. (Tielaitoksen sisäisiä julkaisuja 2/1998). Helsinki: Tielaitos.
Luttinen, R. T. 1999. Properties of Cowan’s M3 headway distribution. In Highway Capacity,
Quality of Service, and Traffic Flow and Characteristics. (Transportation Research Record
1678). Washington, D.C.: Transportation Research Board. P. 189–196.
Luttinen, R. T. 2001. Capacity and Level of Service on Finnish Two-Lane Highways. (Finnra
Reports 18/2001). Helsinki: Finnish Road Administration.
Luttinen, R. T. 2003. Capacity at Unsignalized Intersections. (TL Research Report 3). Lahti:
TL Consulting Engineers, Ltd.
Luttinen, R. T. 2004a. Capacity of unsignalized intersections under random arrivals. Paper
presented at German-Dutch-Finnish Seminar on Traffic Engineering, Delft, June 8- 9,
2004.
Luttinen, R. T. 2004b. Movement capacity at two-way stop-controlled intersections. In
83rd Annual Meeting Compendium of Papers CD-ROM. Transportation Research Board.
Washington, D.C.
REFERENCES
Luttinen, R. T. & Nevala, R. 2002. Capacity and Level of Service of Finnish Signalized
Intersections. (Finnra Reports 25/2002). Helsinki: Finnish Road Administration.
Lyly, S. (ed.) 1988. Liikenne ja väylät [Transportation Planning and Facilities]. (In two
volumes: RIL 165-I and RIL 165-II). Helsinki: Suomen Rakennusinsinöörien Liitto.
Madanat, S. M., Cassidy, M. J. & Wang, M.-H. 1994. Probabilistic delay model at stop-
controlled intersection. Journal of Transportation Engineering, vol. 120, no. 1, p. 21–36.
Major, N. G. & Buckley, D. J. 1962. Entry to a traffic stream. In ARRB Proceedings, Vol. 1.
Victoria: Australian Road Research Board.
Matloff, N. S. 1988. Probability Modeling and Computer Simulation. Boston: PWS-KENT
Publishing Company.
May, A. D. & Keller, H. E. M. 1967. A deterministic queueing model. Transportation
Research, vol. 1, no. 2, p. 117–128.
Mayne, A. J. 1954. Some further results in the theory of pedestrians and road traffic. Biometrika,
vol. 41, p. 375–389.
McDonald, M. & Armitage, D. J. 1978. The capacity of roundabouts. Traffic Engineering
and Control, vol. 19, no. 10, p. 447–450.
McDonald, M., Hounsell, N. B. & Kimber, R. M. 1984. Deometric delay at non-signalised
intersections. (TRRL Supplementary Report 810). Crowthorne: Transport and Road
Research Laboratory.
McLean, J. R. 1989. Two-Lane Highway Traffic Operations: Theory and Practice. New York:
Gordon and Breach Science Publishers.
McNeil, D. R. & Weiss, G. H. 1974. Delay problems for isolated intersections. In D. C. Gazis
(ed.), Traffic Science. New York: John Wiley & Sons. Chapter 2, p. 109–174.
Miller, A. J. 1961. A queueing model for road traffic flow. Journal of the Royal Statistical
Society, vol. 23B, no. 1, p. 64–75.
Miller, A. J. & Pretty, R. L. 1968. Overtaking on two-lane rural roads. In Proceedings of the
Fourth Australian Road Research Board Conference. Victoria: Australian Road Research
Board. P. 582–594.
Morse, P. M. 1958. Queues, Inventories and Maintenance. New York: John Wiley & Sons.
Nelson, W. 1982. Applied Life Data Analysis. New York: John Wiley & Sons.
Neuburger, H. 1971. The economics of heavily congested roads. Transportation Research,
vol. 5, no. 4, p. 283–293.
Neuts, M. F. 1989. Structured Stochastic Matrices of M/G/1 Type and Their Applications. New
York: Marcel Dekker, Inc.
Newell, G. F. 1965. Approximation methods for queues with application to the fixed-cycle
traffic light. SIAM Review, vol. 7, no. 2, p. 223–240.
Newell, G. F. 1971. Applications of Queueing Theory. London: Chapman and Hall.
Newell, G. F. 1982. Applications of Queueing Theory. Second ed. London: Chapman and
Hall.
Niittymäki, J. 1993. Calibration of HUTSIM traffic signal simulator. In Simulering och
Trafiksignalstyrning—Seminariet [Simulation and traffic signal control—Seminar]. TFK.
Stockholm.
Niittymäki, J. 1998. Isolated Traffic Signals — Vehicle Dynamics and Fuzzy Control. (Trans-
portation Engineering Publication 94). Otaniemi: Helsinki University of Technology.
Niittymäki, J. & Pursula, M. 1994. Valo-ohjattujen liittymien simulointi ja ajodynamiikka
[Simulation and vehicle dynamics of signalized intersections]. (Julkaisu 81). Otaniemi:
Teknillinen korkeakoulu, Liikennetekniikka.
REFERENCES
Niittymäki, J. & Pursula, M. 1997. Saturation flows at signal-group-controlled traffic sig-

nals. In Highway Capacity Issues and Analysis. (Transportation Research Record 1572).
Niittymäki, J. & Uusiheimala, M. 1996. Kritiska tidsavstånd i korsningar utan signalreglering
[Critical gaps at unsignalized intersections]. In VTI & KFB Forskardagar. (KFB & VTI
forskning/research 18, part 5). KFB & VTI. Linköping. P. 167–177.
Nordqvist, S. 1958. Gators och vägars kapacitet [Capacity of streets and highways]. (Med-
delande nr 39). Stocholm: IVA:s Transportforskningskommission.
Nordqvist, S., Bång, K.-L. & Hansson, A. 1973. Kapacitetsutredning: Literaturstudier och
analys [An inquiry on capacity: Literature studies and analyses]. Np.: Statens Vägverk.
TV 118.
OECD Road Research Group 1974. Capacity of At-Grade Junctions. (Road Research).
Paris: Organisation for Economic Co-operation and Development.
O’Flaherty, C. A. 1997. Intersection design and capacity. In C. A. O’Flaherty (ed.),
Transport Planning and Traffic Engineering. London: Arnold. Chapter 20, p. 356–399.
Oliver, R. M. 1961. A traffic counting distribution. Operations Research, vol. 9, no. 6,
p. 802–810.
Palm, C. 1943. Intensitätsschwankungen im Fernsprechverkehr [Intensity fluctuations in tele-
phone traffic]. (Ericsson Technics 44). Stockholm: Telefonaktiebolaget L M Ericsson.
Parzen, E. 1962. Stochastic Processes. New York: Holden-Day.
Pasanen, E. 1991. Ajonopeudet ja jalankulkijan turvallisuus [Driving speeds and pedestrian
safety]. (Julkaisu 72). Otaniemi: Teknillinen korkeakoulu, Liikennetekniikka.
Pécheux, K. K., Pietrucha, M. T. & Jovanis, P. P. 2000. User perception of level of service
at signalized intersections: Methodological issues. In Brilon (2000), p. 322–335.
Philbrick, M. J. 1977. In search of a new capacity formula for conventional roundabouts.
(TRRL Laboratory Report 773). Crowthorne: Transport and Road Research Laboratory.
Plank, A. W. 1982. The capacity of a priority intersection—two approaches. Traffic Engin-
eering and Control, vol. 23, no. 2, p. 88–92.
Plank, A. W. & Catchpole, E. A. 1984. A general capacity formula for an uncontrolled
intersection. Traffic Engineering and Control, vol. 25, no. 6, p. 327–329.
Pollaczek, F. 1930. Über eine Aufgabe der Wahrscheinlichkeitstheorie I–II [Concerning a
question in probability theory]. Mathematische Zeitschrift, vol. 32, p. 64–100, 729–750.
Pöschl, F. J. 1983. Die nicht signalgesteuerte nebenstraßenzufahrt als verallgemeinertes
M/G/1-warteschlangensystem [The unsignalized minor street entry as a generalized
M/G/1 queuing system]. Zeitschrift für Operations Research, vol. 27, p. B91–B111.
Prabhu, N. U. 1965. Queues and Inventories: A Study of Their Basic Stochastic Processes.
New York: John Wiley & Sons.
Prevedouros, P. D. 1988. A model of unsignalized intersection capacity based on Erlang-3
gap distribution. In Brilon (1988b). P. 165–179.
Pursula, M., Hagring, O. & Niittymäki, J. 1997. Simulation and capacity calculation of
intersections without traffic signals—a discussion and a roundabout case study. In Kyte
(1997), p. 208–216.
Pursula, M. & Peltola, V. 1997. Capacity analysis and design practice for unsignalized
at-grade intersections in Finland. In Kyte (1997), p. 197–207.
Pursula, M. & Peltola, V. 1998. Finska beräkningsmetoder för korsningarnas kapacitet och
fördröjningar [Finnish calculation methods for capacity and delays of intersections]. In
NORDKAP—NORDiskt KAPacitetssamarbete [NORDKAP—Nordic capacity coopera-
tion] (University of Lund 1998). Sammandrag av seminarium på Grand Hotel i Lund,
Sverige 2–3 juni 1997.
REFERENCES
Pursula, M. & Ristikartano, J. 1988. Liikennevirta ja välityskyky [Traffic flow and capacity].
In Lyly (1988). Chapter B4, p. 131–177.
Raff, M. S. 1950. A Volume Warrant For Urban Stop Signs. Saugatuck, Connecticut: The Eno
Foundation for Highway Traffic Control.
Road Research Laboratory 1965. Research on Road Traffic. London: Her Majesty’s
Stationery Office.
Robertson, D. I. 1969. TRANSYT: A Traffic Network Study Tool. (RRL Report LR 253).
Crowthorne, Berkshire: Road Research Laboratory.
Ross, S. 2002. A First Course in Probability. Sixth ed. Upper Saddle River: Prentice-Hall.
Ross, S. M. 1996. Stochastic Processes. Second ed. New York: John Wiley & Sons.
Rouphail, N. M. & Akçelik, R. 1992. Oversaturation delay estimates with consideration of
peaking. In Highway Capacity and Traffic Flow. (Transportation Research Record 1365).
Rouphail, N. M., Tarko, A. & Li, J. 1997. Traffic flow at signalized intersections. In Gartner
& Messer (1997). Chapter 9.
Schnabel, W. & Lohse, D. 1997. Grundlagen der Straßenverkehrstechnik und der Verkehr-
splanung [Fundamentals of traffic engineering and transportation planning]. Vol. 1.
Second ed. Berlin: Verlag für Bauwesen.
Semmens, M. C. 1988. The capacity of non-signalised road junctions with high-speed con-
trolling flows. Traffic Engineering and Control, vol. 29, no. 5, p. 280–283.
Shrope, E. B. 1952. Testing a traffic circle for possible capacity. Proceedings of the Highway
Research Board, vol. 31, p. 415–424.
Siegloch, W. 1973. Die Leistungsermittlung an Knotenpunkten ohne Lichtsignalsteuerung
[Capacity calculations at unsignalized intersections]. (Strassenbau und Strassenverkehr-
stechnik 154). Bonn: Bundesminister für Verkehr.
Simon, M. J. 1991. Roundabouts in Switzerland. In Brilon (1991), p. 41–52.
SNRA 1995a. CAPCAL 2: Model description of Intersection with signalcontrol. Borlänge:
Swedish National Road Administration. Publication 1995:008E.
SNRA 1995b. CAPCAL 2: Model description of Intersection without signalcontrol. Borlänge:
Swedish National Road Administration. Publication 1995:007E.
SNRA 1995c. CAPCAL 2: Model description of Roundabouts. Borlänge: Swedish National
Road Administration. Publication 1995:009E.
Song, L., Dunne, M. C. & Black, J. A. 1993. Models of delay and accident risk to pedes-
trians. In C. F. Daganzo (ed.), Transportation and Traffic Theory. (Proceedings of the
12th International Symposium on Transportation and Traffic Theory, Berkeley, California,
USA, 21–23 July, 1993). Amsterdam: Elsevier. P. 295–312.
Spanier, J. & Oldham, K. B. 1987. An Atlas of Functions. New York: Hemisphere Publishing
Corporation.
Statens vägverk 1977. Beräkning av kapacitet, kölängd, fördröjning i vägtrafikanläggningar
[Calculation of capacity, queue length, delay in road traffic facilities]. Stockholm. TV
131.
Statens Vegvesen 1985. Kapasitet i kryss [Capacity at an intersection]. Oslo. Håndbok–127.
Stuwe, B. 1991. Capacity and safety of roundabouts—German results. In Brilon (1991),
p. 1–12.
Sullivan, D. P. & Troutbeck, R. J. 1994. The use of Cowan’s M3 headway distribution for
modelling urban traffic flow. Traffic Engineering and Control, vol. 35, no. 7/8, p. 445–450.
REFERENCES
Syyrakki, V., Lyly, S. & Granberg, T. 1975. Liikennevirta ja liikenteenvälityskyky [Traffic

flow and capacity]. In S. Lyly (ed.), Liikenne ja väylät [Transportation Planning and
Facilities]. Helsinki: Suomen Rakennusinsinöörien Liitto. Chapter 3, p. 65–94.
Tanner, J. C. 1951. The delay to pedestrians crossing a road. Biometrika, vol. 38, p. 383–392.
Tanner, J. C. 1953. A problem of interference between two queues. Biometrika, vol. 40,
p. 58–69.
Tanner, J. C. 1961a. Delays on a two-lane road. Journal of the Royal Statistical Society,
vol. 23B, no. 1, p. 38–63.
Tanner, J. C. 1961b. A derivation of the Borel distribution. Biometrika, vol. 48, p. 222–224.
Tanner, J. C. 1962. A theoretical analysis of delays at an uncontrolled intersection. Biometrika,
vol. 49, no. 1 and 2, p. 163–170.
Tanner, J. C. 1967. The capacity of an uncontrolled intersection. Biometrika, vol. 54, no. 3
and 4, p. 657–658.
Tanyel, S. & Yayla, N. 2003. A discussion on the parameters of Cowan M3 distribution for
Turkey. Transportation Research, vol. 37A, no. 2, p. 129–143.
Taylor, M. A. P. (ed.) 2002. Transportation and Traffic Theory in the 21st Century. (Proceedings
of the 15th International Symposium on Transportation and Traffic Theory, Adelaide,
Australia, 16–18 July 2002). Amsterdam: Pergamon.
Teply, S. & Jones, A. M. 1991. Saturation flow: Do we speak the same language? In Freeway
Operations, Highway Capacity, and Traffic Flow. (Transportation Research Record 1320).
Thedéen, T. 1964. A note on the Poisson tendency in traffic distribution. The Annals of
Mathematical Statistics, vol. 35, no. 4, p. 1823–1824.
Thompson, W. A. 1988. Point Process Models with Applications to Safety and Reliability.
London: Chapman and Hall Ltd.
Tian, Z. Z., Troutbeck, R., Kyte, M., Brilon, W., Vandehey, M., Kittelson, W. & Robin-
son, B. 2000. A further investigation on critical gap and follow-up time. In Brilon (2000),
p. 397–408.
Tiehallitus 1992. Kiertoliittymät: Suunnitteluohje [Roundabouts: Design guidelines]. Hel-
sinki: Tielaitos. TIEL 2130010.
Tie- ja liikennetekniikka 2001. Tasoliittymät: Suunnitteluvaiheen ohjaus [At-grade inter-
sections: Design guide]. Helsinki: Tiehallinto. TIEH 2100001-01.
Tiensuunnittelutoimisto 1978. Valo-ohjaamattoman tasoliittymän välityskyky: Laskenta-
ohje [Capacity of an unsignalized intersection: Calculation guidelines]. Helsinki: Tie- ja
vesirakennushallitus. TVH 722306.
Tiensuunnittelutoimisto 1987. CAPCAL-tietokoneohjelman käyttöohje 2 [CAPCAL user’s
manual 2]. Helsinki: Tie- ja vesirakennushallitus. TVH 723860.
Tijms, H. C. 2003. A First Course in Stochastic Models. Chichester: John Wiley & Sons.
Tracz, M., Chodur, J. & Gondek, S. 1990. The use of reserve capacity and delay as the
complementary measures of junction effectiveness. In Koshi (1990), p. 497–516.
Transportation Research Board 1985. Highway Capacity Manual. (Special Report 209).
Washington, D.C.: National Research Council.
Transportation Research Board 1994. Highway Capacity Manual. (Special Report 209).
Third ed. Washington, D.C.: National Research Council. Updated October 1994.
Transportation Research Board 1997. Review of International Practices Used To Evaluate
Unsignalized Intersections. (Transportation Research Circular 468). Washington, D.C.:
National Research Council.
REFERENCES
Transportation Research Board 2000. Highway Capacity Manual. Washington, D.C.:

National Research Council. HCM2000, metric units.
TRB Committee A3A10, Highway Capacity and Quality of Service 2003.
Approved Corrections and Changes for the Highway Capacity Manual 2000.
http://www.a3a10.gati.org/public/errata-2003-01-29.pdf, 2003-01-29.
Troutbeck, R. 1992. Estimating the Critical Acceptance Gap from Traffic Movements. (Re-
search Report 92-5). Brisbane: Physical Infrastructure Centre, Queensland University of
Technology.
Troutbeck, R. 1995. The Capacity of a Limited Priority Merge. (Research Report 95-8).
Brisbane: Physical Infrastructure Centre, Queensland University of Technology.
Troutbeck, R. J. 1986. Average delay at an unsignalized intersection with two major streams
each having a dichotomized headway distribution. Transportation Science, vol. 20, no. 4,
p. 272–286.
Troutbeck, R. J. 1988. Current and future Australian practices for the design of unsignalized
intersections. In Brilon (1988b), p. 1–19.
Troutbeck, R. J. 1990. Roundabout capacity and the associated delay. In Koshi (1990),
p. 39–57.
Troutbeck, R. J. 1991. Recent Australian unsignalized intersection research and practices. In
Brilon (1991), p. 238–257.
Troutbeck, R. J. 1993. Capacity and design of traffic circles in Australia. In Traffic Flow
and Highway Capacity. (Transportation Research Record 1398). Washington, D.C.:
Transportation Research Board. P. 68–74.
Troutbeck, R. J. 1997a. The Analysis of the Performance of Roundabouts. (Research Report
97-8). Brisbane: Physical Infrastructure Centre, Queensland University of Technology.
Troutbeck, R. J. 1997b. A review on the process to estimate the Cowan M3 headway distri-
bution parameters. Traffic Engineering and Control, vol. 38, no. 11, p. 600–603.
Troutbeck, R. J. 1999. Capacity of limited-priority merge. In Highway Capacity, Quality of
Service, and Traffic Flow and Characteristics. (Transportation Research Record 1678).
Troutbeck, R. J. 2000. Modeling of signalized and unsignalized junctions. In D. A. Hensher
& K. J. Button (ed.), Handbook of Transport Modelling. (Handbooks in Transport 1).
Amsterdam: Pergamon. P. 375–391.
Troutbeck, R. J. 2002. The performance of uncontrolled merges using a limited priority
process. In Taylor (2002), p. 463–482.
Troutbeck, R. J. & Brilon, W. 1997. Unsignalized intersection theory. In Gartner & Messer
(1997), p. 8:1–8:47.
Troutbeck, R. J. & Kako, S. 1997. Limited priority merge at unsignalized intersections. In
Kyte (1997), p. 294–302.
Troutbeck, R. J. & Kako, S. 1999. Limited priority merge at unsignalized intersections.
Transportation Research, vol. 33A, no. 3/4, p. 291–304.
Troutbeck, R. J. & Walsh, D. J. 1994. The differences between queueing theory and gap
acceptance theory. In Akçelik (1994b), p. 617–635.
University of Lund 1998. NORDKAP—NORDiskt KAPacitetssamarbete [NORDKAP—
Nordic capacity cooperation]. (Bulletin 156). Lund. Sammandrag av seminarium på
Grand Hotel i Lund, Sverige 2–3 juni 1997.
Uusiheimala, M. 1995. Kriittisen aikavälin tarkastelu [A study of critical gaps]. (Special
assignment). Otaniemi: Teknillinen korkeakoulu, liikennelaboratorio.
Vejdirektoratet 1999a. Kapacitet og serviceniveau: Baggrund og dokumentation [Capacity

and level of service: Background and documentation]. Draft.
Vejdirektoratet 1999b. Kapacitet og serviceniveau [Capacity and level of service]. Draft.
Voigt, W. 1973. Modificationen des zeitlückenverfahrens [Modifications of Time Headway
Process]. Die Strasse, vol. 13, no. 6, p. 224–227.
Wang, Z. 2000. Critical gap and traffic volume—re-examining the relationship. Traffic Engin-
eering and Control, vol. 41, no. 7, p. 275–278.
Wardrop, J. G. 1957. The traffic capacity of weaving sections of roundabouts. In M. Davies,
R. T. Eddison & T. Page (ed.), Proceedings of the First International Conference on
Operational Research. Baltimore: Operations Research Society of America. P. 266–281.
With discussion.
Watson, H. 1933. Street Traffic Flow. London: Chapman and Hall Ltd.
Webster, F. V. 1958. Traffic Signal Settings. (Road Research Technical Paper No. 39). London:
Her Majesty’s Stationery Office.
Webster, F. V. 1960. Greenford roundabout experiment. Traffic Engineering and Control,
vol. 2, no. 5, p. 266–271.
Wegmann, H. 1991. A general capacity formula for unsignalized intersections. In Brilon
(1991), p. 177–191.
Weiss, G. 1963. A note on highway transparency. Operations Research, vol. 11, p. 460–461.
Letter to the Editor.
Weiss, G. & Herman, R. 1962. Statistical properties of low-density traffic. Quarterly of Applied
Mathematics, vol. 20, p. 121–130.
Weiss, G. H. & Maradudin, A. A. 1962. Some problems in traffic delay. Operations Research,
vol. 10, p. 74–104.
Wolff, R. W. 1989. Stochastic Modeling and the Theory of Queues. Englewood Cliffs:
Prentice-Hall International, Inc.
Wu, N. 1994. An approximation for the distribution of queue lengths at unsignalized intersec-
tions. In Akçelik (1994b), p. 717–736.
Yagar, S. 1977. Minimizing delays for transient demands with application to signalized road
junctions. Transportation Research, vol. 11, no. 1, p. 53–62.
Yeo, G. F. 1962. Single server queues with modified service mechanisms. Journal of the
Australian Mathematical Society, vol. 2, p. 499–507.
Yeo, G. F. & Weesakul, B. 1964. Delays to road traffic at an intersection. Journal of Applied
Probability, vol. 1, p. 291–310.
APPENDICES
APPENDICES
CRITICAL GAP ESTIMATES
A CRITICAL GAP ESTIMATES
Table A.1: Critical gap estimates for major-road left-turn movements at unsignalized intersec-
tions with major-road speed limit 50 km/h
Maximum likelihood Raff’s
Noncensored Censored method
Location N tˆc N tˆc N tˆc
Postitie 16 5.0 11 3.7 16 4.9
Seutula 274 5.6 165 5.2 274 5.6
Seutula 169 5.8 90 5.3 169 5.7
Kuitinmäki 90 5.9 29 4.4 90 5.6
Vihti 89 6.0 27 4.6 89 5.5
Vanha Turuntie-Jorvi 31 7.2 11 6.5 31 7.6
Sture1 22 10.1 2 4.6 22 10.3
Average 6.6 4.9 6.4
Weighted average 6.0 5.1 5.9
Table A.2: Critical gap estimates for minor-road right-turn movements at yield-controlled
intersections with major-road speed limit 50 km/h
Seutula 161 5.6 90 5.0 161 5.2
Seutula 167 6.3 79 5.7 167 6.3
Average 7.3 5.8 6.9
Table A.3: Critical gap estimates for minor-road left-turn movements at yield-controlled inter-
sections with major-road speed limit 50 km/h
Rajatorppa 113 5.0 87 4.8 113 4.9
Kuitinmäki 156 6.3 86 5.5 156 6.2
Otaniemi 47 6.7 25 5.2 47 6.5
Malminkartano 57 6.9 23 6.3 57 6.6
Pukinmäki 41 8.4 9 6.1 41 7.8
Average 6.9 5.8 6.6
Table A.4: Critical gap estimates for major-road left-turn movements at unsignalized intersec-
Leppävaara 38 5.3 16 4.5 38 4.7
Leppävaara 115 5.6 54 5.1 115 5.5
Karak2 33 6.2 12 5.1 33 5.7
Leppävaara 97 6.5 34 4.8 97 5.7
Seurasaari 25 6.7 6 4.1 25 5.5
Saukkola 24 7.4 5 4.4 24 6.7
Saukkola 24 8.0 6 7.7 24 7.8
Porintie 52 7.2 21 5.6 52 6.6
Average 6.6 5.2 6.0
Table A.5: Critical gap estimates for minor-road right-turn movements at unsignalized inter-
sections with major-road speed limit 60 km/h
Location Control N tˆc N tˆc N tˆc
Leppävaara Yield 89 5.8 41 4.8 89 5.4
Leppävaara Yield 140 6.2 69 5.2 140 5.7
Saukkola Yield 9 6.8 7 6.8 9 6.8
Porintie Stop 180 8.9 53 6.8 180 8.6
Average 6.9 5.9 6.0
Table A.6: Critical gap estimate for minor-road left-turn movements at a yield-controlled
intersection with major-road speed limit 60 km/h
Saukkola 32 8.3 11 6.3 32 8.0
Table A.7: Critical gap estimates for minor-road right-turn movements at yield-controlled
intersections with major-road speed limit 80 km/h
Numpj1 80 7.9 59 7.1 80 7.5
Myllylampi 20 7.8 16 7.3 20 7.4
Hanko-Hyvinkää-Nummela 44 10.7 25 10.0 44 10.4
Average 8.8 8.1 8.4
Table A.8: Critical gap estimates for minor-road left-turn movements at unsignalized intersec-
Jorvaksentie-Sundsberg 69 6.6 62 6.5 69 6.7
Jorvaksentie-Sundsberg 96 6.6 84 6.5 96 6.7
Vihdintie-Lahnus 41 8.9 31 8.9 41 9.5
Numet2 25 10.0 13 9.3 25 10.4
Average 8.3 7.9 8.4
Table A.9: Critical gap estimates for roundabouts

Central island Maximum likelihood Raff’s
diameter Noncensored Censored method
Location (m) N tˆc N tˆc N tˆc
Nummela 1 8.0 50 4.9 18 4.8 50 5.1
Nummela 2 8.0 69 5.1 14 4.1 69 4.4
Tapanila 11.0 20 6.1 5 2.6 20 4.7
Pukinmäki 12.5 42 2.9 27 2.4 42 2.5
Kitee 17.0 53 4.2 39 3.9 53 4.2
Veikkola 25.0 41 4.3 38 4.2 41 4.3
Katinen 26.0 13 5.2 3 3.7 13 4.0
Katuma 38.0 61 4.6 21 3.9 61 4.9
Hyrylä 55.0 302 4.5 204 4.1 302 4.3
Lappeenranta 140 5.5 53 4.6 140 5.0
Average 4.6 3.7 4.3
HUTSIM INITIALIZATION FILE
B HUTSIM INITIALIZATION FILE

A typical Hutsim initialization file, which shows the most important parameters, is
displayed below.
1151 HUTSIM 5.373I/TPMA Initialization File

101 *.*; Input File Mask
102 Test.trf; Input File
103 Test.out; Output File
104 Test.rep; Report File
105 ; Default Directory
111 4 Input Mode Traf-file
112 4 Output Mode Delay file
113 1 Report Mode BasicReport
121 0 TimeHour (0-60)
122 60 TimeMin (0-60)
123 10 NoRepINT (minutes)
124 0 NoGenINT (minutes)
125 60 RepINT (minutes)
131 0 VCOL (vehicle color coding)
132 0 DWIN (downer window mode)
133 0 PLOT (plotter mode)
134 0 PIPE (pipes visible)
135 1 ANIM (animation mode)
136 1 TIME (time update mode)
201 1 Real Time Connection
202 10 Time Step Unit (1/100 s)
203 0 Extra Load (ms)
204 2.00 Low Frequency Warning (Hz)
205 1 Screen Update Interval (sec)
211 3 Graph Time Scale (pixels/second)
212 3 Graph Speed Scale (pixels/unit)
213 300 Graph Display Area (pixels)
221 1 Sound Effects
251 41952 Seed Number (0-63999)
252 1.00 Traffic Scale Factor
253 0.50 Minimum Arrival Time Headway (s)
261 2 Max. Group Number
262 0 Max. Report Number
263 1 Stop Criteria (Level number)
264 0.50 Option-Zone Weight (n=1)
265 1.00 Option-Zone Weight (n>1)
266 ,; Column Delimiter
267 6 Column Spacing
271 0.026 Delay Cost (1/sec)
272 0.180 Stop Cost (1/stop)
273 2.000 Option-Zone Cost (1/veh)
274 10.000 Queue Criteria (dist/m)
301 2.50 Unit Speed Value (km/h)
302 1.20 Following Time Headway (s)
303 0.20 Stable Area Width (s)
304 1.20 Stop Distance (m)
305 0.05 Acceleration Adjustment
308 0.005000 Speed Limit Coefficient
321 300 Max Obstacle Sight (m)
322 300 Max. Vehicle Sight (m)
323 1.00 Average Reaction Time (s)
325 1 Reaction Mode (0=No Reaction Delay)
326 2 Creeping Mode (0=No Creeping)
327 0.05 Creeping Unstability Factor
351 0.90 Left Changing Treshold
352 0.05 Right Changing Treshold
353 0.80 Forced Lnc Gap Back Left
354 0.80 Forced Lnc Gap Front Left
373 0.50 Forced Lnc Gap Back Right
374 0.50 Forced Lnc Gap Front Right
361 0.20 Maximum Gap Decrease \% (Forced Lane Change)
362 20 Forced Lane Change Gap Decreasing Time
363 0.70 Polite Drivers %
355 10.00 Minimum Time in Lane (s)
356 1 Lane Switch Allowed
357 0 Left Handed Traffic
358 1.10 Voluntary Lnc Gap Back Left
359 1.10 Voluntary Lnc Gap Front Left
378 1.10 Voluntary Lnc Gap Back Right
379 1.10 Voluntary Lnc Gap Front Right
360 7.50 Speed Loss Threshold (km/h)
701 0.70 AICC Following Gap
702 1.00 AICC Stopping Distance
721 0 Route Guidance Mode
722 0 Route Guided Vehs (%)
723 0.90 Route Choice Threshold
724 0.20 Travel Time Averaging
400 10 Number of Vehicle Types
100 Normal Deceleration Rates
401 1.9 1.6 1.2 1.7 1.7 1.5 1.3 5.0 1.9 1.9
100 High Deceleration Rates
402 3.2 3.0 2.0 2.8 2.8 2.6 2.4 6.0 3.2 3.2
100 Normal Acceleration Rates
403 1.6 0.7 1.0 0.5 0.6 0.4 0.2 5.0 1.6 1.6
100 High Acceleration Rates
404 2.4 1.0 1.4 0.7 0.8 0.6 0.3 6.0 2.4 2.4
100 Vehicle Lengths
405 4.0 7.0 12.0 18.0 24.0 24.0 24.0 40.0 4.0 4.0
HUTSIM INITIALIZATION FILE
100 Vehicle Type Distributions

501 88 4 2 2 2 1 1 0 0 0
502 100 0 0 0 0 0 0 0 0 0
503 85 4 11 0 0 0 0 0 0 0
504 57 21 4 18 0 0 0 0 0 0
505 87 4 9 0 0 0 0 0 0 0
506 100 0 0 0 0 0 0 0 0 0
507 0 0 100 0 0 0 0 0 0 0
508 0 0 0 0 100 0 0 0 0 0
509 50 0 50 0 0 0 0 0 0 0
100 Car-following Gap Distributions 0.1-5.0s
800 0 Default Distr No (0=Not Used)
801 0 0 0 0 0 0 0 0 0 0 0 100 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
802 0 0 0 0 0 0 0 0 0 10 20 40 20 10 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
803 0 0 0 0 0 0 10 10 10 10 10 20 10 10 10 10 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
804 0 0 0 0 0 0 10 10 10 10 10 10 10 10 10 10 10 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
805 0 0 0 0 0 0 10 10 10 10 10 20 10 10 10 10 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
806 0 0 0 0 0 0 10 10 10 10 10 20 10 10 10 10 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
807 0 0 0 0 0 0 10 10 10 10 10 20 10 10 10 10 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
808 0 0 0 0 0 0 10 10 10 10 10 20 10 10 10 10 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
809 0 0 0 0 0 0 10 10 10 10 10 20 10 10 10 10 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
100 Desired Speed Level Distributions
900 0 Default Distr No (0=Not Used)
901 0 0 0 100 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
902 0 0 0 0 0 0 0 100 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
903 0 0 0 0 0 0 0 0 0 0 0 100 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
904 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
905 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
906 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
907 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
100 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
908 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 100 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
909 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 100 0 0 0 0 0 0 0 0 0 0 0 0 0 0
910 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 100 0 0 0 0 0 0 0 0 0 0
911 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 0 0 0 0 0 0
912 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 100 0 0
913 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
914 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
915 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
916 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
917 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
918 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
919 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
1000 End of Initialization File
ISBN 951-803-181-9
ISSN 1459-1553
TIEH 3200849E-v

Capacity and Level of Service at Finnish PDF

Uploaded by

Copyright:

Available Formats

Capacity and Level of Service at Finnish PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Capacity and Level of Service at Finnish PDF

Uploaded by

Copyright:

Available Formats

S12 Solutions to improve main roads

Capacity and Level of Service at Finnish

Tiehallinnon selvityksiä 1/2004

Capacity and Level of Service at Finnish

Finnish Road Administration

PDF version (www.tiehallinto.ﬁ/julkaisut )

The publication is sold by

Finnish Road Administration

Valo-ohjauksettomien liittymien välityskyvyn ja viipeiden arviointi Suomessa perustuu

Helsinki, 3rd July 2004

Finnish Road Administration

SYMBOLS AND ABBREVIATIONS 18

1.1 Methodological approaches 22

2 GAP ACCEPTANCE THEORY 25

2.1 Relative priority of trafﬁc streams 25

3 CAPACITY AT UNSIGNALIZED INTERSECTIONS 52

3.1 Deﬁnition of capacity 52

3.5.2 Queuing model for Rank 3 stream capacity 78

4 DELAYS AT UNSIGNALIZED INTERSECTIONS 105

4.1 Deﬁnitions of delay 105

5 PERFORMANCE ANALYSIS METHODOLOGIES 138

5.1 Performance measures 138

5.4.5 Capacity 150

6 SIMULATION STUDIES 156

6.1 Hutsim as a simulation tool 156

7 CRITICAL GAPS AND FOLLOW-UP TIMES 178

7.1 Estimation methods 178

8 CONCLUSIONS AND RECOMMENDATIONS 183

8.1 Limitations of simulation experiments 183

A CRITICAL GAP ESTIMATES 206

B HUTSIM INITIALIZATION FILE 209

Approach: A set of lanes at an intersection that accommodates all left-turn, through

Calibration: The process of adjusting model parameters in order to minimize the

Conflicting movements: Traffic streams in conflict at an intersection.

Counting process: A continuous-time process that represents the number of occur-

Deceleration delay: Time lost due to deceleration.

Degree of saturation: Same as demand-to-capacity ratio.

Delay: The additional travel time experienced by a driver, passenger or pedestrian.

Deterministic model: A mathematical model that is not subject to randomness.

Gap: A time interval between two consecutive major-stream vehicles evaluated by a

Gap acceptance: The process by which a minor-stream vehicle accepts an available

Ideal conditions: Base conditions.

Impedance: The reduction in the capacity of a minor stream, caused by congestion in

Quality of service: The capability of a transportation facility or system under given

Queue: A line of vehicles, bicycles or persons waiting to be served by a system.

Rank: A level of priority in the hierarchy of trafﬁc streams at an unsignalized inter-

Recreational vehicle: A heavy vehicle, generally operated by a private motorist, en-

Service measure: A speciﬁc performance measure used to assign a level of service.

Stationary conditions: Conditions not changing during the analysis period.

Stochastic process: A series of random variables indexed over time or space. In a

Stream: A distinguishable element of the total trafﬁc ﬂow traversing an intersection—

Through vehicles: All vehicles passing directly through an intersection.

Traffic condition: A characteristic of traffic flow, including distribution of vehicle

Volume: The number of persons or vehicles passing a point on a lane, roadway, or

Waiting time in queue: Time spent in a queue waiting for service.

Waiting time in system: Time spent in queue and in service.

Yield line: A line of triangles at an unsignalized intersection to intensify the effect of

SYMBOLS AND ABBREVIATIONS

lw Length of weaving section