Eluru 2010

Transportation Letters
The International Journal of Transportation Research
ISSN: 1942-7867 (Print) 1942-7875 (Online) Journal homepage: http://www.tandfonline.com/loi/ytrl20
An econometric multi-dimensional choice model

of activity-travel behavior
Naveen Eluru, Abdul Pinjari, Ram Pendyala & Chandra Bhat
To cite this article: Naveen Eluru, Abdul Pinjari, Ram Pendyala & Chandra Bhat (2010) An
econometric multi-dimensional choice model of activity-travel behavior, Transportation
Letters, 2:4, 217-230
To link to this article: http://dx.doi.org/10.3328/TL.2010.02.04.217-230
Published online: 07 Sep 2013.
Submit your article to this journal
Article views: 82
View related articles
Citing articles: 1 View citing articles
Full Terms & Conditions of access and use can be found at

http://www.tandfonline.com/action/journalInformation?journalCode=ytrl20
Download by: [85.248.227.163] Date: 14 March 2016, At: 20:28

217
Naveen Eluru, * Abdul R. Pinjari, Ram M. Pendyala, and Chandra R. Bhat
1 2 3 4
An econometric multi-dimensional choice

model of activity-travel behavior
Abstract: Recent evidence suggests that many activity-travel choices are inter-dependent with one another and hence inex-
tricably linked in ways that need to be better understood to help inform the specification of activity-based travel model systems.
Model systems in practice often sequentially link a series of choice dimensions into a deeply nested logit model where acces-
sibility variables (logsum terms) from lower nests cascade up through the structure to the higher levels in the model structure.
While these model systems are convenient from a practical standpoint, they ignore the potential jointness in choice-making
processes and do not effectively and directly capture the effects of spatial land use and built environment characteristics on
activity generation. In this paper, a unified model of activity type choice (generation), time of day choice, mode choice, destina-
tion choice, and time use allocation (duration) is formulated and estimated on a survey sample data set drawn from the 2000
Downloaded by [85.248.227.163] at 20:28 14 March 2016
San Francisco Bay Area Travel Survey (BATS). The model system constitutes a joint multiple discrete continuous extreme
value (MDCEV) – multinomial logit (MNL) model, in which all discrete choices, except for destination choice, and the con-
tinuous duration dimension are modeled using the MDCEV, and destination choice is modeled as a MNL (with sampling of
alternatives) nested and therefore integrated with the MDCEV model component. The parameter estimates of the joint model
offer behaviorally intuitive results that support the integrated treatment of these choice dimensions as a choice “bundle”. The
potential applicability of the model system is demonstrated through a policy simulation example that shows how changes in
travel cost and time variables lead to changes in out-of-home discretionary activity participation.
Keywords: activity type choice, time of day choice, activity duration, mode and destination choice, joint model, simulta-
neous equations model, integrated model, MDCEV-MNL model
1. INTRODUCTION AND MOTIVATION and time (Meloni et al., 2004). The focus on activity-based
approaches is spurred on by enhanced understanding of
Travel demand modeling is characterized by an increas- activity travel behavior dimensions, increasing concerns
ing shift towards activity-based travel demand modeling of global climate change as well as the recent advances in
approaches that explicitly recognize that travel is under- micro-simulation based computation approaches. The move
taken to fulfill activity needs and desires dispersed in space towards microsimulation-based approaches facilitates the
disaggregate representation of behavioral agents and their
*Corresponding Author: interactions, while simultaneously incorporating the ability
to analyze policy impacts and address equity concerns at the
Department of Civil Engineering and Applied Mechanics,
1
McGill University, 817 Sherbrooke Street West, Montreal, Quebec, level of the individual traveler or any sub-market segment of
CANADA H3A 2K6, Tel: 512-436-3803, Fax: 512-475-8744, interest (Miller and Roorda, 2003).
Email: naveen.eluru@mcgill.ca The major objective of the activity based micro-sim-
Department of Civil and Environmental Engineering,
2 ulation approaches is to strive to mimic and replicate
University of South Florida, 4202 E. Fowler Ave., Tampa, FL 33620 activity-travel processes of individuals. These choice pro-
Tel: 813-974- 9671, Fax: 813-974-2957, Email: apinjari@eng.usf.edu cesses include such dimensions as activity type choice, time
Arizona State University, School of Sustainable Engineering and
3
of day choice, trip chaining or linking choice, joint versus
the Built Environment, Room ECG252, Tempe, AZ 85287-5306, solo activity engagement choice, destination choice, mode
Tel: (480) 727-9164; Fax: (480) 965-0557, Email: ram.pendyala@asu.edu
choice, activity sequencing decisions, and activity time allo-
The University of Texas at Austin, Department of Civil,
4
cation (duration) decisions. Many of these choice processes
Architectural and Environmental Engineering, 1 University Station
C1761, Austin, TX 78712-0278, Tel: 512-471-4535, Fax: 512-475-8744,
are discrete in nature (e.g., activity type choice, time of day
Email: bhat@mail.utexas.edu period choice, mode and destination choices), while others
Transportation Letters: The International Journal of Transportation Research (2010) 2: (217-230) J. Ross Publishing, Inc. © 2010
DOI 10.3328/TL.2010.02.04.217-230
218 Transportation Letters: The International Journal of Transportation Research
are continuous (e.g., activity duration). The structure of the and estimating simultaneous equations model systems that
inter-dependencies across these choice variables assumed represent joint choice processes in which individuals and
by the analyst has important implications for activity-based households are making a “package” of activity-travel choices
model specifications. as a “bundle”. In other words, it is conceivable that individual
Given the large number of choice variables consid- agents are making choices regarding the type of activity to
ered in the behavioral process, it is not surprising that pursue, the mode and destination, and the time allocation to
many approaches have opted for a sequential framework in the activity in one swoop, thus motivating the adoption of a
which activity-travel choices are modeled sequentially. These “joint” choice model specification in which unobserved fac-
approaches often resort to the adoption of deeply nested logit tors unknown to the analyst may be simultaneously impact-
models (Ben-Akiva and Lerman, 1985) where one choice ing multiple dimensions of interest (Jara-Diaz et al., 2007).
process is nested within another choice process and so on, The current paper contributes to the literature on modeling
forming a long chain of inter-connected nests to complete activity travel choice processes simultaneously.
the representation of the behavioral process (Bowman, 1995; The growing interest in the ability to model multiple
Bowman and Bradley, 2006; PB Consult, 2005). As it is virtu- choice dimensions simultaneously, where the endogeneity
ally impossible to estimate such long chains of nested logit of many choice variables is explicitly recognized in the activ-
models simultaneously (i.e., in one single step), components ity-travel behavior modeling arena, motivates this paper.
of the nested logit model are usually estimated one step (or Specifically, this paper presents a joint model system of five
maybe two steps) at a time and the logsum from one level is choice dimensions:
carried up to the next higher level, resulting in a sequential
• Activity type choice
estimation and model application approach. Although there
• Activity time of day choice (treated as discrete time
are other behavioral model systems that attempt to move
intervals)
away from such deeply nested logit specifications, such as
• Mode choice
those based on computational process modeling and heuris-
• Destination choice
tic approaches (for example, see Arentze and Timmermans,
• Activity duration (continuous choice dimension)
2005), the fact remains that most current activity-based
model systems break down the behavioral decision process These five dimensions are of particular interest to policy
so that one is modeling only one or two choice processes at makers to devise strategies that influence travel behavior.
any step in the model system. It may be argued that some For instance, to devise a policy to reduce vehicle emissions,
choice processes are best modeled sequentially; for example, it is necessary for any modeling tool to provide informa-
a longer-term choice of work or residential location is likely tion on activity flexibility (activity type and time-of-day),
to precede a shorter-term choice of time of departure for travel mode, , activity location and distance (destination),
a discretionary trip. On the other hand, there are a host of and vehicle soak times (activity duration). These five choice
choices, such as destination choice, mode choice, activity dimensions are of critical interest to any activity-based
type choice, activity duration choice, and activity accompani- model system regardless of the model design that might
ment choice that one would expect to be short-term choices be adopted. Thus, this paper aims to specify and estimate
made contemporaneously. Although a sequential treatment a comprehensive econometric model system that jointly
of such contemporaneous choice mechanisms is convenient models these five choice dimensions in a holistic unifying
from a practical model estimation and application stand- utility-maximization framework. The model system explic-
point, it is unclear whether such model systems truly repli- itly includes consideration of built environment attributes
cate behavioral processes. including level of service variables and spatial land use char-
On the other hand, resorting to a simultaneous equa- acteristics to capture the potential impacts of such variables
tions modeling framework for will allow the analyst to on the activity generation process, a key area that warrants
incorporate the complex inter-dependencies that often char- additional research. Such a model specification provides the
acterize activity-travel choice processes. However, the simul- ability to examine induced and suppressed demand effects in
taneous equations approach results in a substantial increase response to changes in system capacity and level of service.
in computational complexity, particularly when the num- The modeling methodology adopted in this paper builds
ber of choice dimensions being modeled becomes greater on previous work by the authors and constitutes a joint
than two. In fact, one could argue that current practice has multiple discrete continuous extreme value model and mul-
adopted the sequential framework in the activity-based mod- tinomial logit model system (Bhat, 2005, Bhat et al., 2006,
eling realm because of the estimation challenges and compu- Bhat, 2008). The multiple discrete continuous extreme value
tational complexity associated with specifying, identifying, (MDCEV) model component is used to jointly analyze activ-
An econometric multi-dimensional choice model of activity-travel behavior 219
ity type choice, activity time of day choice, mode choice, and and two modes of travel (auto, and non-auto), yielding 60
activity duration. Specifically, the MDCEV model is used to different types of OH discretionary activity episodes (or ptm
represent activity participation (discrete choice) and time combinations). Thus, there are a total of 62 MDCEV choice
use (continuous choice) for different types of activities at alternatives in that one or more of these alternatives may
different time periods of the day by different travel modes. be chosen by an individual through the course of a day.1
The activity location choice is modeled using a multinomial For each of these alternatives, the ψ terms (ψ1, ψ2, and ψptm)
logit (MNL) model nested within the MDCEV framework. are the baseline utility parameters that control the discrete
The model system is estimated for a survey sample drawn choice of the alternative. For all alternatives except the first
from the 2000 San Francisco Bay Area Travel Survey (BATS), alternative, the γ terms (γ2 and γptm) allow for corner solu-
a comprehensive database that includes detailed household tions (i.e., the possibility of not choosing the alternative)
and personal socio-economic, demographic, and activity- as well as satiation effects (i.e., diminishing marginal utility
travel information together with a host of secondary trans- with increasing time investment).2 There is no γ term corre-
portation level-of-service and land use variables. sponding to the first alternative (maintenance activity) as it is
The next section presents the modeling methodology in always chosen by all individuals.
detail. This is followed by a description of the dataset and Finally, let each of the 60 OH discretionary activity epi-
survey sample. The fourth and fifth sections present model sode types (ptm) be defined (by its purpose-timing-mode
estimation and policy simulation results, while the sixth and (ptm) combination) such that an individual participates in no
final section offers concluding remarks. more than one episode of that type in a day. Consequently, if
an individual chooses to undertake an activity episode type
(ptm), it has to be at only one of the several destination alter-
2. MODELING METHODOLOGY natives (l) available to her/him.
Let the index for the activity destination (or location) be
This section presents the modeling methodology for the joint l, and let Nptm be the set of destinations available for an activ-
MDCEV-MNL model structure. First, the utility structure is ity episode type (ptm). Further, for each activity episode type
presented, second, the econometric model specification is (ptm), let γptm be defined as follows (Bhat et al., 2006):
presented, and finally the procedure for sampling of location
choice alternatives is discussed. An intuitive behavioral inter- ψptm= exp ∑ δlptmWlptm (2)
pretation of the model structure is offered as well. l ∈Nptm
where, Wlptm is the utility perceived by the individual for

2.1 Utility Structure undertaking the OH discretionary activity episode of pur-
Consider the following utility specification for the integrated pose p, during time period t, by traveling on mode m to
analysis of individuals’ activity time-use, timing, mode location l, and δlptm is a dummy variable taking a value of 1
choice, and location choice decisions: if the l th location is chosen for that activity episode such that
x2 62 x
∑ (δlptm) = 1 (i.e., only one location is chosen).
l ∈Nptm
U(x)=ψ1ln x1+ γ2ψ2ln +1 +∑ γptmψptm ln ptm +1 (1)
γ2 ptm=3
γptm With the above definition of ψptm and other terms
described earlier, the individual is assumed to maximize the
In the above equation, the first term ψ1lnx1 corresponds
utility function U(x) in Equation (1) subject to x1 + x2 + ∑
to the utility contribution of the total daily time invested ptm
(x1) in all maintenance activities, and the second term cor- xptm = X; x1 > 0, x2 ≥0, x1 ≥ 0 ∀ptm = 3, 4, … 62. Since the indi-
responds to the utility contribution of the total daily time
invested (x2) in all in-home (IH) discretionary activities. The 1
Without loss of generality, all individuals can be assumed to participate
next set of terms correspond to the utility contribution due to in maintenance activities. On the other hand, an individual can
participate in none, or one, or more of IH discretionary and 5 OH
the time investment (xptm) in out-of-home (OH) discretion-
discretionary activity purposes (p) identified above. If (s)he chooses to
ary activity episode types (indexed by ptm), with each activity participate in OH discretionary activities, (s)he can do so during one
episode type defined by its purpose (p), timing (t), and mode or more of the 6 time periods (t), and access the activities using one
of travel (m). In the current empirical context considered in or more of the 2 travel modes (m). Thus, there is multiple discreteness
this paper, there are five OH discretionary activity purposes in the choices across the activity purpose, activity timing, and travel
(volunteering, socializing, recreation, meals, and shopping), mode dimensions.
2
To distinguish the satiation along OH discretionary activity purpose, activ-
six time periods (3am-7am or early morning, 7am-9am or
ity timing, and travel mode dimensions (and to facilitate estimation),
morning, 9am-12noon or late morning, 12noon-4pm or γptm (ptm = 3, 4, …, 62) is parameterized as γptm = γp × γt × γm, where
afternoon, 4pm-7pm or evening, and 7pm-3am or night), γp, γt, γm , are the estimated dimension-specific satiation parameters.
vidual maximizes U(x) and can choose only one location for ary activity episode ptm, ϕ′wlptm is the observed utility corre-
each activity episode ptm type, the functional form of U(x) sponding to the potential location l for the activity episode,
implies that the individual will consider the location that and ηlptm is the unobserved utility component associated with
provides the maximum utility for each activity episode ptm the location l of activity episode ptm. Similar to ε1 and ε2, the
type in the process of maximizing U(x) (see Bhat et al., 2009). ηptm terms are assumed to be independent and identically dis-
tributed (across different activity episode ptm types) Gumbel
That is, ∑δ Wlptm= max Wlptm, or ψptm =exp max Wlptm ,
lptm terms. Within each activity episode ptm type, however, all of
l ∈Nptm l ∈Nptm l ∈Nptm
the error terms may share common unobserved attributes
Thus, the individual’s utility maximizing problem can be (specific to the activity episode ptm type) generating correla-
written as: tions among the ηptm terms across all potential locations for
x2 the activity episode. Thus, for each activity episode ptm type,
U(x) = ψ1ln x1 + γ2ψ2ln +1 the following distribution of error terms may be used:
γ2
x F(η1ptm,η2ptm,…,ηLptm)=exp–[e–η1ptm/θptm+e–η2ptm/θptm+…+e–ηLptm/θptm] θptm (6)
+∑ γptmexp max W ln ptm +1 (3)
ptm
l ∈Nptm lptm γptm
where the θptm is the dissimilarity parameter indicating the
subject to x1 + x2 + ∑ xptm = X; x1 > 0, x2 ≥0, xptm≥ 0 ∀ptm. level of correlation among the ηLptm terms across all the
ptm
The analyst can solve for the optimal values of x1, x2, potential locations for the activity episode ptm combination.
and xptm by forming the Lagrangian and applying the Kuhn- Given this error distribution, using the properties of Gumbel
Tucker (KT) conditions. Specifically, the following KT con- distribution, Hptm in Equation (4) can be expressed as:
ditions can be formed (see Bhat, 2008): xptm
Hptm= max β′zptm + ϕ′wlptm+ ηlptm – ln +1 (7)
H2 = H1 if x2 > 0 (4) l ∈Nptm γptm
H2 < H1 if x2 = 0
ϕ′wptm xptm
Hptm = H1 if xptm > 0 = β′zptm + θptm ln ∑ exp + ζptm – ln +1
Hptm < H1 if xptm = 0 l ∈Nptm θptm γptm
where, ζptm is a standard independent and identically distrib-
where, uted (across ptm) Gumbel error term. In this equation,
ϕ′wptm
H1 = ln(ψ1)– ln(x1) ln ∑ exp constitutes the logsum term.
l ∈Nptm θptm
x2 Next, following the MDCEV model derivations (see
H2 = ln(ψ1)– ln +1 , and Bhat, 2008), the probability that the individual chooses the
γ2
first Q out of K (= 62) activity purpose-timing-mode alterna-
xptm tives (this may include maintenance as well as the IH discre-
Hptm= max Wlptm – ln +1
l ∈Nptm γptm tionary activities without any timing and mode distinctions)
for time investments x1*, x2*,…, xQ* may be written as:
2.2 Econometric Structure Q Q 1

To complete the model specification, let ψ1 = exp(β′z1 P(x1*, x2*,…, xQ* 0,0,…,0) = ∏ rk ∑ rk
+ ε1) and ψ2 = exp(β′z2 + ε2), where β′z1 and β′z2 are the k= 1 k= 1
K
observed baseline utility components of maintenance and 1 ∑ eVh
IH discretionary activities, respectively, and ε1 and ε2 are h= 1
the corresponding unobserved components assumed to be (Q–1)!, (8)
K Q
independent and identically Gumbel distributed. Further, to ∑ eVh
define ψptm, we expand Wlptm as: h= 1
where 3
Wlptm = β′zptm + ϕ′wlptm + ηlptm (5) 1 1

r1= and rk= ∀k >1
xk* xk* + γk
where, β′zptm is the observed baseline utility corresponding to
the activity purpose, timing, and mode of the OH discretion- Note that the notation for the subscripts of the choice alternatives has been
3
changed to k (=1,2,…62) from 1,2,ptm(=3,4,…62) for convenience.

V1 = β′z1 – ln(x1), is maintained for the location choice as in Equation (10).4

However, sampling the location choice alternatives warrants
x2
V2 = β′z2 – ln +1 , and ϕ′wlk
γ2 a correction to the log-sum term ∑ exp used in
l ∈Nk θk
ϕ′wlk xk
Vk = β′zk – θkln ∑ –ln +1 ; ∀k >2 the MDCEV component of the joint model (see Equation
l ∈Nptm θk γk 7). This is because, in this term, the sum of exponentials of
the utilities (scaled by the dissimilarity parameter) of all the
The conditional probability that location l will be chosen
ϕ′wlk
for an activity episode purpose-timing-mode (ptm) combi- location choice alternatives ∑ exp
is not equal to
nation k, given that xk* >0, is given by: θk
l ∈Nk
the sum of exponentials of the utilities of a sample of those
P(l | xk* > 0; l ∈ Nk) = P[ϕ′wlk + ηlk > ϕ′wl′k+ ηl ′k ∀ l′ ≠ l] (9)
alternatives. This is corrected by incorporating a scaling
Based on the multivariate Gumbel distribution function factor (πk) that is equal to the total number of available loca-
for the ηlk (or ηlptm) terms (l = 1, 2,…, L) from Equation (6), tion choice alternatives divided by the number of sampled
the above probability expression can be computed using the alternatives. Since location choice alternatives are sampled
following standard multinomial logit formula: randomly, and since the random sample varies across indi-
ϕ′wlk viduals and activity purpose-timing-mode (ptm) combina-
exp tions, this scaling factor should help approximate the logsum
θk
P(l | xk* > 0; l ∈ Nk) = (10) term reasonably well. That is:
ϕ′wl′k
∑ exp
θk ϕ′wlk ϕ′wlk
∑ ≈ ln ∑ exp
l ∈Nk
ln πk exp (12)
Next, the unconditional probability that the individual l ∈a random sample of Nk θk l ∈Nk θk
spends x1* amount of time in daily maintenance activi- In this study, 30 location choice alternatives are ran-
ties, x2* amount of time in daily IH-discretionary activities, domly sampled from 1099 potential locations yielding, πk =
x3* amount of time in OH discretionary activity episode 36.63.
purpose-timing-mode (ptm) combination 3 (i.e., k = 3) at
location a, x4* amount of time in OH discretionary activity 2.4 An Intuitive Behavioral Interpretation
episode purpose-timing-mode (ptm) combination 4 (i.e., k =
The probability expression in Equation (11) is a combina-
4) at location b, … and so on, may be written as:
tion of MDCEV and single discrete choice probabilities.
P(x1*, x2*, x3* at a, x4* at b, xQ* at q, 0,0,0,…0) (11) Specifically, for each OH discretionary activity episode
P(x1*, x2*,… xQ*,0,0,…0) × P(a | x3* > 0)…P(q | xQ* > 0) purpose-timing-mode (ptm) combination chosen by an
individual, a single discrete choice model of location choice
is invoked. The parameters ϕ and θk appear in both the
MDCEV probability expression (Equation 8) as well as the
2.3 Sampling of Location Choice Alternatives
standard discrete choice probability expression for the choice
A practical issue with the proposed MDCEV-MNL model (as of activity location (Equation 10) to create jointness between
also with the deeply nested logit approach) is that, since there the multiple discrete-continuous and single discrete choices.
can be a large number of location choice alternatives at the
single discrete choice level (and since multiple single discrete The reader will note here that Equation (10) is derived from a nested
4
choice models may be invoked), the model estimation can extreme value error term distribution as in Equation (6). However,
be highly computation intensive. To reduce the computa- since this distribution assumes the same scale parameter for all loca-
tion time, the analyst can include only a smaller sample of tion choice alternatives associated with the activity episode ptm type,
the location choice alternatives (with the chosen alternative the location choice parameters will be consistent. In essence, as long as
the error distributions do not allow different scale parameters across
in the sample) during estimation. According to McFadden
the location choice alternatives associated with an activity episode ptm
(1978), random sampling of alternatives will not compro- type (i.e., to accommodate spatial correlations, etc.) and no random
mise the consistency of the location choice model parameters coefficients are estimated in the location choice model, one can use a
as long as a simple multinomial logit modeling framework random sample of location choice alternatives to consistently estimate
the model parameters. See Bierlaire et al. (2008) for more details on
sampling related issues with multi-dimensional choice models.
Further, the logsum term (see Equation 7) appearing in the processes can be found in Guo and Bhat (2004) and Pinjari
MDCEV probability expression carries the accessibility of et al. (2009).
destinations (or potential locations) from the single discrete As mentioned in the previous section, the activity
location choice model to the MDCEV model of time invest- choice dimensions modeled in this paper include activity
ment by activity purpose, timing, and travel mode. Thus, type choice, activity time of day choice, travel mode choice,
Equation (11) represents a unified and comprehensive model activity location (destination) choice, and activity time
of activity-travel program generation that incorporates the use allocation (duration).5 The MDCEV model component
influence of accessibility measures on activity time-use, tim- alternatives are formed as combinations of activity type, time
ing, and mode choices. of day, and travel mode while the duration of each activ-
The proposed two-level MDCEV-MNL model is an ity episode constitutes the continuous dependent variable.
attractive alternative to the deeply nested logit modeling Finally, the MNL module accommodates the activity location
approach available in the literature, where accessibility or destination choice. There are: (a) a maintenance activity
measures have to propagate up to the activity generation type, (b) an in-home discretionary activity type, and (c) five
level through multiple levels of a deeply nested logit model. out-of-home discretionary activity types, six time periods,
Further, the MDCEV-MNL model provides a seamless way and two travel modes, yielding a total of 62 possible MDCEV
of incorporating time-use (and the impact of accessibility choice alternatives (2 + 5 × 6 × 2 = 62). It is to be noted that
on time-use) into the framework. Specifically, the modeling the activity timing and travel mode analysis is limited to the
framework explicitly accommodates the concept that indi- five out-of-home discretionary activity types.
vidual’s activity time-use (i.e., time allocation) decisions are In order to control for fundamental differences between
important and influential components of their activity-travel workers and non-workers in their activity engagement pat-
decision-making (Bhat and Koppelman, 1999). On the other terns and choice processes, and in the interest of brevity, the
hand, the deeply nested logit approach does not explicitly analysis in this paper was restricted to the sample of 5,360
incorporate activity time-allocation choices into the analysis non-working individuals (includes employed individuals
framework in a straight forward manner. Another appealing not reporting to work on the survey day) aged 16 years or
feature is that the model recognizes the simultaneity of the above. Descriptive statistics for this sample of individuals
activity time-use, timing, mode choice, and location choice are presented in Table 1. All 5,360 individuals participate
decisions within a unified utility maximization framework. in in-home maintenance for an average duration of nearly
11 hours. Forty percent engage in in-home discretionary
activities for an average duration of about 5.5 hours. Note
3. DATA DESCRIPTION that the average durations are computed over those who
actually participate in the activity type. A little over one-half
The data set used in this paper is derived from the 2000 San of the sample participated in OH discretionary activities, for
Francisco Bay Area Travel Survey (BATS), designed and an average duration of about 2.5 hours. It is found that the
administered by MORPACE International Inc. for the Bay automobile mode is the preferred and dominant mode of
Area Metropolitan Transportation Commission (MTC). The travel accounting for nearly 90 percent of all out-of-home
data includes information on: (1) Individual and household discretionary activity engagement. Non-maintenance shop-
socio-demographics for over 15,000 households in the Bay ping shows a relatively high participation rate, but lower
Area, and (2) All activity episodes (including activity type, time allocation (regardless of mode), while activities such as
start and end times of the activity, geo-referenced location meals, socializing, and recreation show lower participation
of activity participation, and mode of travel to the activity) rates but higher time allocation. Across the top of the table
undertaken by the individuals in all surveyed households for (in the grey shaded row), it is seen that only a very small per-
a two-day period. cent of individuals participate in OH discretionary activities
The travel survey records were augmented extensively in the early morning, and the percentage steadily rises into
with several secondary data items, including land-use char- the afternoon, and then shows a decline towards the night
acteristics, transportation network level-of-service data, and hours. Activities undertaken in the morning and early morn-
Census population and housing data. In addition, geo-ref- ing, however, show the longest average durations relative to
erenced data on businesses, bicycle facilities, highways and those in the afternoon and evening, potentially indicating the
local roads were used to derive spatial variables character- effect of time constraints that might get tighter towards the
izing the activity-travel environment (ATE) in and around latter half of the day. Overall, this table shows the interplay
the household locations of the individuals in the data set.
Details regarding the data preparation and augmentation In the current study, individual time budget required in the MDCEV
5
model is specified as (24 hours – individual’s sleep activity duration).

Table 1. Descriptive Statistics of Activity Participation and Time-Use by Activity Purpose, Activity Timing and Travel Mode
ACTIVITY TIMING
ACTIVITY PURPOSE and Number (%) of non- Early Morning Late Morning Afternoon Evening Night
TRAVEL MODE workers participating, Morning (7am-9am) (9am-12pm) (12pm-4pm) (4pm-7pm) (7pm-
and mean duration of (3am-7am) 3am)
participation among
those participating
Maintenance 5360 (100%) 651 min -- -- -- -- -- --
IH Discretionary 2133 (39.8%) 341 min -- -- -- -- -- --
OH Discretionary 2752 (51.3%) 163 min -- -- -- -- -- --
OH Discretionary Auto mode 2473 (89.9%) 158 min
Volunteering 396 (14.4%) 149 min 4 (1.0%) 81 (20.5%) 137 (34.6%) 89 (22.5%) 72 (18.2%) 63 (15.9%)
Socializing 508 (18.5%) 128 min 6 (1.2%) 20 ( 3.9%) 125 (24.6%) 159 (31.3%) 97 (19.1%) 77 (15.2%)
Meals 809 (29.4%) 115 min 13 (1.6%) 90 (11.1%) 206 (25.5%) 270 (33.4%) 223 (27.6%) 84 (10.4%)
Non-Maintenance Shopping 1092 (39.7%) 60 min 4 (0.4%) 46 ( 4.2%) 372 (34.1%) 571 (52.3%) 175 (16.0%) 53 ( 4.9%)
Recreation 738 (26.8%) 145 min 33 (4.5 %) 116 (15.7%) 256 (34.7%) 200 (27.1%) 115 (15.6%) 88 (11.9%)
OH Discretionary Non Auto 432 (15.7%) 134 min
mode
Volunteering 37 (1.3%) 170 min 2 (5.4%) 9 (24.3%) 10 (27.0%) 8 (21.6%) 3 (8.1%) 6 (16.2%)
Socializing 72 (2.6%) 140 min 0 (0.0%) 3 (4.2%) 19 (4.2%) 27 (37.5%) 21 (29.2%) 4 (5.6%)
Meals 135 (4.9%) 119 min 1 (0.7%) 9 (6.7%) 35 (25.9%) 54 (40.0%) 25 (18.5%) 18 (13.3%)
Non-Maintenance Shopping 132 (4.8%) 59 min 0 (0.0%) 4 (3.0%) 50 (37.9%) 62 (47.0%) 12 (9.1%) 6 (4.5%)
Recreation 131 (4.8%) 136 min 1 (0.8%) 14 (10.7%) 52 (39.7%) 33 (25.2%) 32 (24.4%) 6 (4.6%)
a
The reader will note here that the average time investments reported in this table are for only those who participated in the corresponding activity
purpose or for those who participated in OH discretionary activities during the corresponding time period. Also, the activity participation percent-
ages across all activity purposes (or across all time periods, or modes) may sum to more than 100% because of multiple discreteness (i.e., partici-
pation in multiple activity purposes and/or during multiple time periods and/or travel by multiple modes over a day). For example, a non-worker can
undertake both OH recreation and OH meal activities on a day.
b
Percentages in this row are out of the 2752 non-workers who participated in at least one OH discretionary activity during the day.
c
Percentages in this column, from this row onward, are out of the 2473 non-workers who traveled by auto mode for at least one OH discretionary
activity during the day.
d
Percentages from this row and column onward (within this block of rows) are based on total number of non-workers participating in row activity
purpose [(4/396)×100=1.0%].
among the dimensions of activity-travel participation that However, extending the estimation process to incorporate
merit a unified approach towards modeling these behavioral 60 MNL models of destination choice is straightforward by
characteristics. specifying dimension-specific model coefficients; the model
specification here is one in which all destination choice
model coefficients are restricted to be identical across all
4. EMPIRICAL ANALYSIS activity purpose categories, timing categories, and mode cat-
egories. Several variables were included in the model speci-
4.1 Model Specification and Estimation fication including household and personal socio-economic
Results and demographic variables, contextual variables such as day
of week and season of the year, and a host of spatial variables
Model estimation was performed using Gauss code written
characterizing the activity-travel environment (ATE) around
specifically to estimate the joint MDCEV-MNL model sys-
the household locations, not to mention several transporta-
tem. Although it would have been ideal to estimate a separate
tion network level of service variables. The spatial ATE vari-
destination choice model for each of the 60 OH discretionary
ables included density measures, activity opportunity and
activity purpose-timing-mode (ptm) combination categories,
accessibility measures, and population and housing data for
for this initial effort, a single MNL location choice model
the neighborhood (traffic analysis zone). The ATE measures
was estimated for all discretionary activity ptm categories.
were considered at the level of the traffic analysis zone and at
Table 2 The MDCEV Model Results: Baseline Parameter Estimates

Household (HH) Socio-demographics
Kids of Kids of Number # of adults

Single age < 5 age of kids of in HH who HH annual HH annual # of
member yrs 5-15 yrs age worked on income income vehicles in
HH size HH present present <15 yrs the day < 45k >100k HH
‘Activity Purpose’
Dimension
IH and OH Maintenance 0.071 - - - - - - - -
(3.74)
IH Discretionary - - - - - - 0.168 - -0.061
(2.92) (-1.89)
OH Volunteering - - - - - - - - -
OH Socializing - 0.420 - - - - - 0.169 -
(3.73) (3.61)
OH Recreation - - - - - - - 0.169 -
(3.61)
OH Meals - - - - - - - 0.169 -
(3.61)
OH Non-Maintenance - - - - - - - 0.169 -
Shopping (3.61)
‘Activity Timing’
Dimension
Early Morning - - - - - - - - -
Morning - - 0.125 0.297 - - - - -
(1.77) (2.10)
Late Morning - - 0.125 - - -0.170 - - -
(1.77) (-4.43)
Afternoon - - 0.125 - - -0.170 - - -
(1.77) (-4.43)
Evening - - 0.125 0.428 - - - - -
(1.77) (3.88)
Night - - - - - - - - -
‘Travel Mode’
Dimension
Auto mode
Non-auto mode -1.190
(-31.90)
Interactions
OH Recreation - 0.363 - - - - - - -
– Evening (3.53)
OH Recreation - - - - - - - 0.463 -
– Non-auto (2.11)
OH Meals - - - - 0.154 - - - -
– Non-auto (1.17)
OH Meals - - - - -0.535
– Non-auto – Evening (-1.36)
finer spatial resolutions, including within 0.25 mile, 1 mile, MDCEV-MNL model that does not incorporate the log-
and 5 mile radii buffers of the household location (see Guo sum parameters in the MDCEV component. The goodness
and Bhat, 2004 and Pinjari et al, 2009 for complete details). of fit of the two models were compared using the Bayesian
In the current research effort, a comparison was made Information Criterion (BIC), which is given by the expres-
between the joint MDCEV-MNL model that integrates sion –2 × ln(L) + number of parameters × ln(Q), where ln(Q)
destination choice with activity choices and an independent is the log-likelihood value at convergence and Q is the num-
Table 2 (Continued) The MDCEV Model Results: Baseline Parameter Estimates

Individual Socio-demographics Contextual ATE attributes
Total
Age Age Retail employ- Density
< 30 > 65 Licensed Physically employ- Population ment of high-
Female yrs yrs to drive disabled Employed Friday Fall Rain ment density density ways
‘Activity
Purpose’
Dimension
IH and OH 0.315 - - - - -0.173 - - - - - - -
Maintenance (7.29) (-3.59)
IH Discretionary - - - - - -0.1263 - -0.105 - - - - -
(-1.98) (-1.98)
OH Volunteering 0.350 - 0.617 0.743 -0.249 - - - - - - - -
(3.46) (6.51) (9.46) (-3.09)
OH Socializing - 0.467 - 0.743 -0.249 - 0.242 - - - - - -
(2.70) (9.46) (-3.09) (4.31)
OH Recreation - - - 0.743 -0.249 - 0.357 - - - - - -
(9.46) (-3.09) (4.14)
OH Meals - - - 0.743 -0.249 - 0.242 - - - - - -

(9.46) (-3.09) (4.31)
OH Non- - - - 0.743 -0.249 - 0.242 - - - - - -
Maintenance (9.46) (-3.09) (4.31)
Shopping
‘Activity
Timing’
Dimension
Early Morning - - - - - - - - - - - - -
Morning - - - - - - - - - - - - -
Late Morning 0.286 - - - - - - - - - - - -
(5.10)
Afternoon 0.286 - - - - - - - - - - - -
(5.10)
Evening - 0.308 - - - - - - - - - - -
(1.85)
Night - 0.739 -0.482 - - - 0.404 - - - - - -
(4.78) (-3.70) (3.23)
‘Travel Mode’
Dimension
Auto mode - - - - - - - - - - - - -
Non-auto mode -0.220 - - - - - - - - - - - -
(-2.92)
Interactions
OH Non- - - - - - - - - - -0.001 - - -
Maintenance (-2.59)
Shopping –
Afternoon
OH activity Non- - - - - - 0.369 - - - - - - -
auto - Afternoon (1.91)
(except shop-
ping)
OH Meals - - - - - - - - - -0.263 - - - -
Non-auto (-0.85)
OH Meals, - - - - - - - - - - 0.016 (5.43) - -
shopping -Non-
auto
OH Recreation - - - - - - - - - - - - 0.006 -
Non-auto (0.75)
OH Social, meals - - - - - - - - - - - - -0.069
- Non-auto (-0.88)
ber of observations. The model that results in the lower BIC who are licensed to drive have a greater propensity for out-of-
value is the preferred model. The BIC value for the MDCEV- home activities, while the reverse is true for those physically
MNL model (with 103 model parameters) is 150514.2, which disabled. Employed individuals engage less in maintenance
is considerably lower than that for the independent MDCEV- activities and in-home discretionary activities, even on days
MNL model (152334.2 with 102 model parameters). Thus, that they do not work (this analysis was limited to non-work-
the BIC suggests that the simultaneous MDCEV-MNL model ing days for all 5,360 individuals, whether they are employed
of integrated activity choices and destination choice provides or not). Fridays are associated with greater out-of-home dis-
a superior statistical fit than the independent MDCEV-MNL cretionary activity participation, and night time activities. On
model that ignores endogeneity arising from the presence of rainy days, it is less likely that individuals will eat out using
common unobserved attributes that simultaneously influ- non-auto modes. Population density contributes positively
ence multiple choice dimensions. to out-of-home meals, shopping by non-auto mode, possibly
The discussion in this paper is limited to the results of because such areas are better served by transit and have better
the joint MDCEV-MNL model. The MDCEV component is walk and bicycle access to destinations. Overall, the findings
specified (and the results are presented) in such a way that the are consistent with expectations and consistent with those
effect of each variable is first identified separately along the found earlier by Pinjari and Bhat (2010).
activity purpose, activity timing and travel mode dimensions. The estimation results for the destination choice model
Subsequently, any interaction effects of the variable over and are presented in Table 3. The destination choice model
above the uni-dimensional effects are identified. A blank component was estimated with 30 randomly sampled choice
entry corresponding to the effect of a variable indicates no alternatives for each location choice decision. The effects of
significant effect of the variable on the integrated choice pro- transportation network level of service, built environment,
cess. Further, the effects of variables on the baseline utilities and demographic interaction terms were represented in
have been constrained to be equal if coefficient equality could the final model specification. Auto travel times and costs
not be rejected based on statistical tests. Finally, t-statistics decrease the utility associated with choosing a destination for
are presented in parentheses. The final specification of the any activity type. The presence of bicycle lanes, total employ-
MDCEV component of the model is presented in Tables 2. ment, the size of the zone, and zonal household income posi-
In the interest of brevity, and considering the large number tively impact destination choice for discretionary activities
of alternatives (62), tables showing estimates of baseline pref- while retail and service employment, increasing fraction of
erence constants and satiation parameters are not furnished land devoted to residential uses in the zone, and accessibil-
here. ity to passive and natural recreation contribute negatively to
Overall, the model results show indications as expected. destination choice for the activity categories considered in
Larger household sizes are associated with greater levels of this paper. The long list of interaction terms demonstrates
participation in maintenance activities (in and out of home), how household and personal socio-economic and demo-
while single persons are more prone to out-of-home socializ- graphic characteristics play a key role in influencing destina-
ing and recreation in the evening. The presence of very young tion choice for discretionary activities undertaken outside
kids motivates activity engagement in the prime period of the home. In the interest of brevity, a detailed explanation is
day as opposed to early mornings and late nights, although not provided here, but suffice to say that all of the interac-
those with school age children are more restricted to pre- and tion terms included in the model specification are highly
post-school hours. The number of working adults contrib- significant and indicate that household socio-economic and
utes negatively to activity engagement in the middle of the demographic characteristics serve to moderate or enhance
day, presumably due to work constraints. Lower income the likelihood of choosing a certain type of destination for
individuals are more prone to in-home discretionary activi- activity engagement. For example, females are more prone to
ties, while higher income individuals are prone to undertake choosing destinations with high density of eat-out centers, as
out-of-home activities, consistent with expectations. Higher are older people and higher income individuals. Those with
levels of car ownership contribute negatively to in-home kids and in larger households are less prone to choose zones
activity participation and non-auto mode use. with high household density as destinations, presumably
Females are more likely to engage in volunteering and because they prefer more open space and suburban locations
maintenance activities, particularly in the midday period, to accommodate family activities.
confirming the role of gender differences in activity engage- The logsum parameters (θptm) estimated for each activ-
ment. Younger individuals are likely to socialize in the eve- ity purpose, timing, and travel mode combination were not
ning and night, while older individuals (65+ years) are more statistically different from unity. In the final model estima-
likely to volunteer and not undertake night activities. Those tion, all logsum parameters were restricted equal to one. This
Table 3 MNL Component (Location Choice) Model of the MDCEV alternatives through the log-sum variables.
Estimation Results This enables the upper-level choices (i.e., the activity type,
Variable Coefficient t-stat duration, timing, and mode choices) to be sensitive to level-
LOS Measures of-service and land-use attributes, and helps capture the
Auto peak travel time (in minutes) -0.012 -11.82 impacts of policies on different dimensions of activity-travel
Auto peak travel cost (in dollars) -0.056 -2.59 behavior.
ATE Attributes
Density of bicycle lanes 0.129 7.75 4.2 Policy Simulation
Retail employment -0.005 -5.70 The major objective of this paper was to develop a unified
Service employment -0.005 -4.47 model of activity-travel and location choices and time use
Logarithm of Total employment 0.405 29.06 that would allow one to examine the influence of level of
Fraction of residential land-use -2.272 -41.69 service measures and activity-travel environment (ATE)
Logarithm of zonal area 0.056 5.44 attributes on these choice dimensions in an integrated man-
Mean zonal household income 0.007 9.19 ner. To demonstrate the capabilities of the model system
Accessibility to passive and natural -0.364 -2.92 presented in this paper, the model was used to examine the
recreation impacts of the following scenarios on activity and time use
Interaction with socio-demographics behavior:

Density of bicycle lanes * age/100 -0.110 -4.84 • Doubling travel cost across all time periods
Density of bicycle lanes * 0.042 5.46 • Doubling travel cost during peak periods
Continuous income x 10-5
• Doubling travel cost for auto mode
Density of bicycle lanes * household 0.025 4.94
vehicles • Doubling travel time across all time periods
Density of eat-out centers * female 0.003 3.26
• Doubling travel time during peak periods
Density of eat-out centers * 0.010 13.31
• Doubling travel time by auto mode
Continuous income x 10-5
Density of eat-out centers * age/100 0.027 32.64
Logsum variables computed using the activity destina-
Density of eat-out centers * house- 0.014 28.96
tion choice MNL model were used as explanatory variables
hold size in the MDCEV model to predict individual’s participation in
Density of eat-out centers * Own 0.002 2.17 and time allocation to activities by activity purpose, timing,
household and mode. For each policy scenario, logsum variables were
Logarithm of household population 0.102 6.42 computed for all 60 OH discretionary activity purpose, tim-
* age/100 ing, and mode combinations (for use in the base case predic-
Logarithm of household population * 0.011 3.32 tion), and then updated for the specific timing or travel mode
household vehicles
categories for which the policy applied (for the policy case
Household density * No. of kids < -0.006 -1.32
prediction). The prediction using MDCEV was carried out
15yrs
for all individuals in the sample using 1000 replications of
Household density * household size -0.001 -0.36
the error term draws for each individual. Additional details
Household density * household 0.009 3.41
vehicles
about the forecasting procedure using the MDCEV model
Accessibility to employment * -0.003 -6.25
are provided in Pinjari and Bhat (2009).
household size The forecasts under alternative scenarios are presented
Accessibility to employment * Own 0.008 15.38 in Table 4. Specifically, the influence of each policy is
household reported as an aggregate percent change in the amount of
time invested in maintenance activities, in-home discretion-
implies the absence of common unobserved factors across all ary activities, and out-of-home discretionary activities by
location choice alternatives specific to an activity type, tim- purpose, time of day, and mode (relative to the base case).
ing, and mode combination. In general, the results provide indications along expected
Although the log-sum parameter estimates do not indi- lines. Increases in travel cost lead to reduced out-of-home
cate significant correlation among unobserved factors across activity engagement and slight increases in in-home activity
the destination choice alternatives, the MDCEV-MNL model engagement. Increases in travel cost during the peak period
is formulated such that the expected maximum utility values impact volunteer, eat-meal, and recreation activities more
from the MNL alternatives appear in the utility functions than others, and reduce peak period activity engagement
Table 4. Policy Simulation Results

Alternative Activity Purpose
Scenario details Maintenance IH Discretionary OH Volunteer OH Social OH Meals OH Shopping OH Recreation
Travel cost measure increased
0.01 0.02 -0.99 -1.00 -0.84 -0.91 -0.93
by 100% for all time periods
0.00 0.00 -0.58 -0.05 -0.46 0.07 -0.29
by 100% for peak periods
0.01 0.01 -1.16 -1.21 -0.27 -0.31 -0.83
by 100% for auto mode
Travel time measure increased
0.04 0.06 -3.36 -3.40 -2.86 -3.09 -3.18
0.01 0.02 -1.88 -0.15 -1.53 0.22 -0.95
0.03 0.04 -3.85 -3.99 -0.95 -1.05 -2.73
Alternatives Activity Training
Scenario details Early Morning Morning Late Morning Afternoon Evening Night

-0.92 -0.90 -0.92 -0.96 -0.92 -0.87
1.34 -3.89 1.30 1.26 -3.93 1.34
-0.77 -0.75 -0.69 -0.64 -0.68 -0.76
-3.11 -3.07 -3.13 -3.26 -3.13 -2.95
4.41 -12.70 4.22 4.12 -12.83 4.37
-2.54 -2.51 -2.30 -2.12 -2.27 -2.54
Alternatives Travel Mode
Scenario details Auto Non-auto
-1.00 -0.75
-0.30 -0.19
-2.10 2.48
-3.39 -2.57
-0.99 -0.64
-7.03 8.34
while increasing off-peak activity engagement. Increases in a time-specific peak-period travel cost or time increase. It
auto travel costs and times reduce the use of auto mode for appears that individuals are more likely to respond to price
activity engagement and contribute to enhanced mode shares and time signals that cover an entire day as opposed to those
for non-auto modes. In general, travel time increases appear that are narrower in the time band of influence. Overall,
to have larger impacts than travel costs, suggesting that indi- the policy simulation results clearly show that the model is
viduals are more time-sensitive when making activity-travel effective in capturing the responses of individuals to system
choices. In terms of the modal impact, it appears that all changes in a unifying framework.
day travel cost or time increases have a greater impact than
5. CONCLUSIONS mode-specific) coefficients for transportation network level

of service measures and activity-travel environment attri-
This study aims to present a comprehensive unified model butes in the location choice models. In addition, one needs
system of activity-travel choices that is consistent with to note that the policy forecasts provided by the MDCEV
microeconomic utility maximization theory of behavior. model are potentially restrictive in that the total time alloca-
The activity-travel choice dimensions analyzed in this paper tion (budget constraint) is assumed constant across all policy
include activity type choice, time of day choice, mode choice, scenarios (there is simply a reallocation of time across activ-
destination choice, and activity time allocation or duration. ity categories, but maintaining total time expenditure for all
All discrete choices, except for activity destination choice, activity categories considered constant). Overcoming this
and the continuous choice dimension of activity duration limitation is another direction for future research.
are modeled simultaneously using the multiple discrete
continuous extreme value (MDCEV) model form while the
destination choice is modeled using a classic multinomial ACKNOWLEDGEMENTS
logit model (MNL) component. The model components
are tied together within a utility maximization-consistent The authors acknowledge the helpful comments of anony-
framework using logsum variables that reflect the acces- mous reviewers on an earlier version of the paper. The
sibility of destinations for each activity type, timing, and authors are grateful to Lisa Macias for her help in typesetting
mode combination. Model estimation results and the policy and formatting this document.
simulation analysis showed that the joint model system has
merit, offers behaviorally intuitive interpretation, and offers
a goodness of fit statistically superior to that offered by an REFERENCES
independent model system that treats various choice dimen-
Arentze, T.A., and H.J.P. Timmermans. (2005) Albatross
sions separately and sequentially. The model specifications
Version 2: A Learning-Based Transportation Oriented
included built environment and transportation network
Simulation System. European Institute of Retailing and
level of service attributes demonstrating the impact of these
Services Studies, Eindhoven, The Netherlands.
variables on activity-travel dimensions. The model system is
Ben-Akiva, M., and Lerman, S. R. (1985) Discrete Choice
presented for a non-worker sample drawn from the 2000 San
Analysis: Theory and Application to Travel Demand,
Francisco Bay Area Travel Survey (BATS). The efficacy of the
Cambridge, Massachusetts: MIT Press.
model system presented in the current paper is demonstrated
through a host of policy scenarios. These policy scenario Bhat, C.R. (2005) A Multiple Discrete-Continuous
evaluations clearly indicate how changes in activity location Extreme Value Model: Formulation and Application
specific attributes and level-of-service attributes influence to Discretionary Time-Use Decisions. Transportation
activity participation decisions. One of the key empirical Research Part B, 39(8), 679-707.
findings of this analysis is that the built environment and Bhat, C.R. (2008) The multiple discrete-continuous extreme
transportation network level of service attributes of the des- value (MDCEV) model: role of utility function param-
tinations significantly impact activity time use allocation, an eters, identification considerations, and model exten-
aspect that is often overlooked in the literature. sions. Transportation Research Part B, 42(3), 274-303.
The model form adopted in this paper has key impli- Bhat, C.R., and F.S. Koppelman (1999) A retrospective and
cations for activity-travel demand model development. It prospective survey of time-use research. Transportation
appears that the findings reported here support the notion 26(2), 119-139.
that individuals make several contemporaneous activity- Bhat, C.R., S. Sen, and N. Eluru (2009) The Impact of
travel choices jointly as a “bundle”, calling for the simulta- Demographics, Built Environment Attributes, Vehicle
neous modeling of various choice dimensions in a unifying Characteristics, and Gasoline Prices on Household
framework. Activity-travel model systems that purport to Vehicle Holdings and Use. Transportation Research Part
simulate the behavior of agents along the time axis may ben- B, 43(1), 1-18.
efit from the adoption of model forms that are able to simul- Bhat, C.R., S. Srinivasan, and S. Sen (2006) A Joint Model
taneously predict multiple choice dimensions as a “bundle”. for the Perfect and Imperfect Substitute Goods
Ignoring to do so may yield erroneous policy scenario pre- Case: Application to Activity Time-Use Decisions.
dictions. Transportation Research Part B, Vol. 40, No. 10, 827-850.
The current study may be enhanced further by estimat- Bierlaire, M., D. Bolduc, and D. McFadden (2008) The
ing dimension-specific (i.e., activity purpose-, timing-, and Estimation of Generalized Extreme Value Models from
Choice-based Samples. Transportation Research Part B,

42(4), 381-394.
Bowman, J. (1995) Activity Based Disaggregate Travel
Demand Model System with Daily Activity Schedules.
MS Thesis in Transportation, Massachusetts Institute of
Technology.
Bowman, J.L., and M. A. Bradley (2006) Upward Integration
of Hierarchical Activity-based Models. Working paper.
http://www.jbowman.net/papers/20060117%20Bowman%20
and%20Bradley.pdf
Guo, J.Y., and C.R. Bhat (2004) Modifiable areal units: prob-
lem or perception in modeling of residential location
choice? Transportation Research Record 1898, 138-147.
Jara-Diaz, S.R., M. Munizaga, P. Greeven, and R. Guerra
(2007) The unified expanded goods-activities-travel
model: theory and results. Paper presented at the 11th
World Conference on Transport Research, Berkeley, June
24-28.
McFadden, D. (1978) Modelling the choice of residential
location. In A. Karlquist (ed.) Spatial Interaction Theory
and Residential Location, North-Holland, Amsterdam,
pp. 75-96.
Meloni, I., L. Guala, and A. Loddo (2004) Time allocation
to discretionary in home, out of home activities and to
trips. Transportation, 31(1), 69-96.
Miller, E.J. and M.J. Roorda (2003) A prototype model
of activity scheduling. Transportation Research Record
1831, 114-121.
PB Consult (2005) The MORPC Travel Demand Model:
Validation and Final Report. The Mid-Ohio Regional
Planning Commission, Columbus, Ohio.
Pinjari, A.R., and C.R. Bhat (2009) An Efficient Forecasting
Procedure for Kuhn-Tucker Consumer Demand Model
Systems: Application to Residential Energy Consumption
Analysis. Technical paper, Department of Civil &
Environmental Engineering, University of South Florida.
Pinjari, A.R., and C.R. Bhat (2010) A Multiple Discrete-
Continuous Nested Extreme Value (MDCNEV)
Model: Formulation and Application to Non-Worker
Activity Time-Use and Timing Behavior on Weekdays.
Transportation Research Part B, 44(4), 562-583.
Pinjari, A.R., C.R. Bhat, and D.A. Hensher (2009) Residential
Self-Selection Effects in an Activity Time-use Behavior
Model. Transportation Research Part B, 43(7), pp. 729-
748.

Eluru 2010

Uploaded by

Copyright:

Available Formats

Eluru 2010

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

Eluru 2010

Uploaded by

Copyright:

Available Formats

Transportation Letters

The International Journal of Transportation Research

ISSN: 1942-7867 (Print) 1942-7875 (Online) Journal homepage: http://www.tandfonline.com/loi/ytrl20

An econometric multi-dimensional choice model

Naveen Eluru, Abdul Pinjari, Ram Pendyala & Chandra Bhat

To link to this article: http://dx.doi.org/10.3328/TL.2010.02.04.217-230

Published online: 07 Sep 2013.

Submit your article to this journal

View related articles

Citing articles: 1 View citing articles

Full Terms & Conditions of access and use can be found at

Download by: [85.248.227.163] Date: 14 March 2016, At: 20:28

An econometric multi-dimensional choice

where, Wlptm is the utility perceived by the individual for

2.2 Econometric Structure Q Q 1

Wlptm = β′zptm + ϕ′wlptm + ηlptm (5) 1 1

changed to k (=1,2,…62) from 1,2,ptm(=3,4,…62) for convenience.

V1 = β′z1 – ln(x1), is maintained for the location choice as in Equation (10).4

model is specified as (24 hours – individual’s sleep activity duration).

Table 2 The MDCEV Model Results: Baseline Parameter Estimates

Kids of Kids of Number # of adults

Table 2 (Continued) The MDCEV Model Results: Baseline Parameter Estimates

OH Meals - - - 0.743 -0.249 - 0.242 - - - - - -

Interaction with socio-demographics behavior:

Table 4. Policy Simulation Results

Travel cost measure increased

5. CONCLUSIONS mode-specific) coefficients for transportation network level

Choice-based Samples. Transportation Research Part B,

You might also like