The Pakistan Development Review

50:1 (Spring 2011) pp. 6381

Formal Participation in a Milk Supply Chain

and Technical Inefficiency of Smallholder
Dairy Farms in Pakistan
ABID A. BURKI and MUSHTAQ A. KHAN*
This paper provides empirical evidence on the impact on technical inefficiency of
smallholder dairy producers when they formally participate in a milk supply chain. Here the
stochastic production frontier and technical inefficiency effects model are estimated based on
the data gathered from 800 smallholder dairy farms in Pakistan. The results suggest that the
technical inefficiency of the participating farms is significantly reduced. A strong impact of the
supply chain is also detected in reducing technical inefficiency of farms that are located in
remote areas and on those that have larger herd-size. Experienced farmers upto the age of 36
years have the advantage of reducing technical inefficiency. The remaining differences in
relative inefficiency of dairy farms are accounted for by severe long-term depressive disorders.

JEL classification: D24, Q12, Q13, Q18

Keywords: Agri-food Supply Chain, Production Frontiers, Dairy Efficiency, Food
Policy, Pakistan
1. INTRODUCTION
Agri-food supply chain systems have undergone dramatic transformation lately in
many developing countries. Urbanisation, in conjunction with rapid growth in incomes,
has caused the character of urban diets in these countries to shift away from low quality
staple grains towards high quality cereals, then to livestock and dairy products, and
vegetables and fruits [Pingali (2006)]. A combination of these factors have forced many
developing countries to re-orient their production and marketing systems by linking local
producers with the organised commodity networks and super markets to meet the
increasing domestic and global consumer demands. Hence numerous supply chains of
agricultural and food products have been formed by agents engaged in production,
processing, marketing and distribution of these products. The consequences of linking
smallholder producers with the organised supply chain networks catering to domestic or
Abid A. Burki <burki@lums.edu.pk> is Professor, Department of Economics, Lahore University of
Management Sciences, Lahore. Mushtaq A. Khan <mushtaq@lums.edu.pk> is Associated Professor,
Department of Economics, Lahore University of Management Sciences, Lahore.
Authors Note: We would like to thank two anonymous referees for useful comments and suggestions.
We are grateful to Rasheed Ahmad, Syed Babar Ali, Roland Decorvet, Javed Iqbal, Jack Moser, Peter
Wuethrich, and participants of the 5th Biennial Conference of the Hong Kong Economic Association in
Chengdu, China for helpful discussions and comments. We are also thankful for the assistance of Masood
Ashfaq Ahmad on the survey data; Tariq Munir, Sanaullah and Munir Ahmad for conducting the field survey,
and Abubakar Memon for providing excellent research assistance. We gratefully acknowledge partial financial
support from the Lahore University of Management Sciences, and Nestl Pakistan.

64

Burki and Khan

international markets are not fully known: Who are the winners and who the losers in an
integration of this kind; how participation in these supply chains affects the relative
inefficiency of smallholder producers; and how does the buyer-side market structure
affects the sustainability of the participating farms? This paper attempts to answer these
questions.
Much of the research into supply chain networks continues to rely on agribusiness
theory [e.g., Dolan and Humphrey (2000); Islam (2008); Sartorius and Kirsten (2007)]. A
vast literature also examines production and distribution planning of supply chains [see,
among others, Ahumada and Villalobos (2009)], while many others address issues related
to public health as in Jevsnik, et al. (2008). A few papers such as Gow and Swinnen
(1998) and Key and Runsten (1999) show that foreign direct investment in developing
nations helps in enforcement of contracts and adoption of new technologies, yet others
[e.g., Dolan and Humphrey (2000) and Weatherspoon and Reardon (2003)] conclude that
FDI negatively affects small local suppliers. Gow and Swinnen (2001) and Dries and
Swinnen (2004) show that FDI related vertical and horizontal integration contributes to
increased access to finance, inputs and productivity growth while Gorton, et al. (2006)
illustrate how asymmetric information between dairy farmers and milk processors leads
to market failure. Some recent studies have voiced concerns about exclusion of smallscale farmers in developing countries from profitable niche markets due to tighter
alignment of supply chains producing for international super markets [e.g., Reardon and
Barrett (2000); Stanton (2000); Unneveher (2000); Sartorius and Kirsten (2007)]. Yet
there is no empirical evidence on the effects of participation of smallholder producers in
supply chain network on their productive efficiency.
This paper provides evidence from the supply chain of milk processing industry in
Pakistan and evaluates how participation of commercial dairy farms in milk supply chain
network, also known as milk district, affects technical inefficiency of the participating
dairy farms, especially in comparison with the record of their rival, traditional milk
collectors or dodhis. Milk supply chain functions on the basis of: (a) self-collection of
farmers milk by the milk plants, e.g., Nestls milk collection model; (b) third-party milk
collection on behalf of processing units, e.g., Haleeb, Nirala, Noon, etc.; and (c) farmer
cooperatives, e.g., HALLA (Idare-e-Kisan).1
Pakistan is the fourth largest producer of milk in the world where three-fourth of
the total milk supply is produced in the Punjab province. The hallmark of the dairy
economy in Pakistan is the dominance of subsistence dairy households that keep buffalos
and cows in small herd-sizes [Burki, et al. (2004)]. Punjab is also home to one of the
largest milk supply chains in Asia. Punjab has the unique feature of having more than 20
private milk processing companies competing to collect farmer milk, including global
giant Nestl, Haleeb Foods, and Halla. Nestl Pakistan has, this year, completed 23 years
of milk collection from rural Punjab while other milk processing units have also made
significant inroads over the last 15 years. While commercial dairy farms are evenly
spread, the milk supply chain mostly consists of central and southern districts of the
Punjab province where population density is relatively low and milk is surplus. However,
1
Nestl Pakistan is the biggest processing industry of the sector, collecting 1040 tons of milk daily from
over 140,000 farmers in about 3500 villages. Other major industry players include Haleeb, Nirala, Halla, Noon,
Millac, Dairy Bell, Dairy Crest, Premier, Army Dairies and Engro Foods.

Milk Supply Chain and Technical Inefficiency

65

this is not the case in northern districts of Punjab, where a vast informal network of
traditional milk collectors, known as dodhis, is still collecting milk from dairy farmers, as
was the case in southern Punjab before the emergence of the milk supply chain. Gains in
technical efficiency of participating dairy farms are expected on account of better
decision-making.
The milk supply chain creates favourable production conditions in the form of
modern milk storage facilities, better and dependable transportation even to remote areas,
regular payment schedules and buyer-side competition leading to higher farm-gate
prices.2 In effect it is expected that the presence of milk supply chain would lead to gains
in technical efficiency of the participating dairy farms.
This paper uses a rich data set of 800 smallholder dairy producers to examine the
extent to which participation in milk supply chain contributes to reducing the technical
inefficiency of these farms. The results suggest that dairy farms in milk supply chain improve
their long term viability by establishing a steady and secure link with the processing industry.
In general, while technical inefficiency of dairy farms located in the milk supply chain is
significantly reduced, the stronger power of the supply chain is detected in further reducing
technical inefficiency of farms situated in remote areas or those with relatively large farm size.
The paper is organised in six sections. Section 2 outlines the survey of dairy
households and sampling methods; Section 3 describes the empirical framework; Section
4 data and variables; Section 5 analyses the estimation results and examines the impact of
milk supply chain on dairy efficiency; Section 6 presents the conclusions of this study.
2. SURVEY OF DAIRY HOUSEHOLDS AND
SAMPLING METHODS
A survey namely, the LUMS3 Survey of Dairy Households in Rural Punjab 2005,
was designed to draw a representative sample of 800 dairy households from rural Punjab,
who owned at least one milching animal (buffalo or cow), sold milk for at least 6 months,
and did not share ownership of farm resources with other households during the calendar
year 2005.4 Punjab is the most populous of the four provinces, which produces nearly 70
percent of total fresh milk supplies in the country. While the dairy farms are evenly
spread in Punjab, the milk supply chain is mostly concentrated in central and southern
Punjab. The dairy survey was conducted between January and April 2006.
The authors used a probability sampling plan where sampled area (rural Punjab)
was divided into sections according to agro-climatic (crop) zones, mouzas/villages and
target groups. To accommodate the different environmental production conditions faced
by the dairy households, Pinckney (1989) was followed and the districts were classified
into five agro-climatic (or crop) zones consisting of (1) wheat-rice, (2) wheat-mix, (3)
wheat-cotton, (4) low intensity barani (rain-fed), and (5) barani regions.
For instance, Nestls milk supply chain model generally functions by setting-up rural milk collection
centres, which provide access to chillers in remote rural areas. Some milk collection networks also provide
dairy extension services.
3
LUMS is short for the Lahore University of Management Sciences.
4
The authors organised and supervised the survey, which was carried out by a three-member team of
trained professional surveyors. A 26-page survey questionnaire was developed and appended by the WHOs
self reporting questionnaire (SRQ-20), meant for measuring prevalence of depressive disorders in the surveyed
dairy farmers.
2

66

Burki and Khan

In stage 1, ten districts were randomly picked (two from each agro-climatic zone)
from 34 districts of Punjab.5 In stage 2, four mouzas6/villages were randomly drawn from
each selected district based on the list obtained from Pakistan Mouza Statistics 1998
[Pakistan (1999)]. Out of 40 mouzas/villages sampled, 26 had at least one player from
milk processing industry collecting milk. In stage 3, lists of commercial dairy households
in selected mouzas/villages were prepared in consultation with notables of the areas and
local milk collection units of the processing industry. Based on the lists, 20 dairy
households were randomly selected from each mouza/village, with equal probability.
Five replacement dairy households were also selected from each mouza/village to replace
those who could not be interviewed. Of the 800 dairy households sampled, 160 were
drawn from each agro-climatic zone. Around 77 percent of the farms owned up to 4
milching animals, 21 percent owned 510 animals and only 2 percent owned 1130
animals. Thus small and subsistence dairy farms, which are the hallmark of Pakistans
dairy economy, were well represented in the survey design.
3. ESTIMATION PROCEDURES
The empirical framework employed in this paper involves the stochastic frontier
approach, first introduced by Aigner, et al. (1977) and Meeusen and Van den Broeck
(1977), which postulates the existence of technical inefficiency in the production process.
This approach uses the concept of a frontier that depicts maximum output obtainable
from given inputs, where technical inefficiency of a farm is estimated by deviations from
the frontier. To illustrate, let the milk production technology be represented by
yi = f (xi ; )

evi ui

where yi is the output of the ith dairy farm, xi (i = 1,,n) is a 1 k vector of values of
known functions of inputs for the ith dairy farm, is a k 1 vector of unknown
parameters to be estimated, and f (xi ; ) is the frontier production function (usually
assumed as Cobb-Douglas). As usual in frontier literature, the stochastic composite error
term in Equation (1) is decomposed into vi and ui where vi is typically the symmetric error
term taken as normal, independently and identically distributed (iid) as N (0,

v2 ), which

captures the random effects of measurement errors in output, external shocks and events
outside a farms control, while ui > 0 is the asymmetric technical inefficiency measure
(usually assumed as half-normal, exponential, gamma or truncated normal distribution)
representing farm-specific inefficiency effects reflecting the extent of the stochastic
shortfall of the ith dairy farm output from the frontier. Following Battese and Coelli
(1993, 1995), technical inefficiency is related to a vector of farm specific attributes Zi in
such a way that ui = Zi + wi > 0, where represents a vector of parameters to be
estimated, and wi is distributed as N (0,

2w ), which is obtained by truncation from below

where the point of truncation occurs at Zi , or wi > Zi .

5
The sample districts were Hafizabad and Narowal in wheat-rice zone, Sargodha and Okara districts in
mixed-cropping zone, Pakpattan and Khanewal districts in wheat-cotton zone, Muzaffargarh and Layyah in
low-intensity zone, and Jhelum and Attock in barani zone.
6
Mouza is the smallest administrative unit under the revenue department which may consist of one big
village or few small villages. Punjab province has 23385 mouzas with an average of 600 mouzas in each district.

67

Milk Supply Chain and Technical Inefficiency

The start is taken with the translog specification for the stochastic production
frontier,7 which offers the advantage of being a second-order Taylor series expansion to
an arbitrary technology, written as

ln yi 0

ln xi 0.5 ij ln xi ln x j vi ui

(2)

where the technical inefficiency effects, ui, are assumed to be defined by a linear function
of explanatory variables given by

ui

j 1

Zij k wi

(3)

where y and x are the indicators of output and inputs for the ith dairy farm, and the CobbDouglas technology is nested within the translog production technology, i.e., when all ij
= 0. Moreover, Zij is a set of environmental or managerial variables influencing technical
inefficiency, ui, of dairy farms, while k captures unmeasured determinants of ui that are
fixed within a district (district fixed-effects).
4. THE DATA AND VARIABLES
Table 1 presents descriptive statistics of the relevant variables. The dependent variable
in the production function is the estimated gross value of milk,8 and other dairy products sold
during the year. The value of milk income is calculated at the price quoted by the dairy farms.
The average value of production of milk and other dairy output is Rs 88,520 per farm, which
translates into around Rs 243 per day per farm. Based on the size, dairy production varies
across dairy farms ranging from only Rs 900 to around Rs one million.
Seven input variables used in the frontier production function are (1) shed and
structure capital, (2) animal capital, (3) fodders, (4) straws and concentrates, (5)
molasses, (6) feed water, and (7) hired and family labour. Shed and structure capital
measures the user cost of sheds, structures and electricity costs, etc. The average shed
and structure capital is Rs 5,713, which is highly variable ranging from only Rs 20 to Rs
66,000 because subsistence farms do not use shed or structures for their dairy animals.
The animal capital variable is calculated by taking user cost of each animal worked out
on the basis of price and remaining life-span of the dairy animals. Prices of dairy cattle
and buffaloes significantly vary depending upon, among other things, on their breed,
genetic endowments and age, etc. Animal capital turns out to be a major component of
dairy cost with an average amount of Rs 12,583 per farm. Two other major inputs in
dairy production are fodders, and straw and concentrate with average use of 0.81 acres
for fodders and 2,520 kg (63 40 kg) of straw and concentrate.
7
For a recent review of studies that have used the stochastic frontier model in farming sector, see
Bravo-Ureta, et al. (2007).
8
Due to long recall period (i.e., one-year), milk production reported by dairy farms is subject to large
measurement error. To avoid the obvious measurement problem in a key variable, we adopt a procedure, due to
Khan (1997, 2000), and predict daily milk production of each dairy animal in our sample. We obtain estimates
of daily milk production by using the parameter estimates from Khan (2000) for the respective lactation length
of each animal separately for first calves, later calves, and for the summer and winter months together with (i)
the reported milk production for each animal on the interview day, and (ii) reported peak time daily milk
production of each animal.

68

Burki and Khan

Table 1
Descriptive Statistics for the Variables of the Frontier Production
Function and Inefficiency Model
Variables
Mean
Frontier Production Function
Output
Milk Production and other Dairy Outputs (Rs)
88517.9
Inputs
Shed and Structure Capital (Rs)
5713
Animal Capital (User Cost)
12583
Fodders (Acres)
0.81
Straws and Concentrates (40kg)
62.81
Molasses (Yes=1, No=0)
0.025
Feed Water (No. of Times Feed Water to Animals)
2.34
Family and Hired Labour (Hours)
2097
Technical Inefficiency Model
Farm Characteristics
Herd-size (Number)
3.51
Head Age (Years)
49.25
Depression (if SRQ8=1, Otherwise=0)
0.119
Head Literate (Yes=1, No=0)
0.447
Location Variable:
Distance Pucca Road (km)
0.861
Milk Supply Chain
Milk Supply Chain (Yes=1, No=0)
0.525
No Player (No Industry Player in Mouza, Yes=1, No=0)
0.425
One-player (One Player in Mouza, Yes=1, No=0)
0.250
Two-players (Two Players in Mouza, Yes=1, No=0)
0.225
Three-players (Three Players in Mouza, Yes=1, No=0)
0.10
District
Sargodha (Yes=1, No=0)
0.1
Narowal (Yes=1, No=0)
0.1
Hafizabad (Yes=1, No=0)
0.1
Pakpattan (Yes=1, No=0)
0.1
Okara (Yes=1, No=0)
0.1
Muzafargarh (Yes=1, No=0)
0.1
Layyah (Yes=1, No=0)
0.1
Khanewal (Yes=1, No=0)
0.1
Jhelum (Yes=1, No=0)
0.1
Attock (Yes=1, No=0)
0.1
Sample Size
800
Source: LUMS Survey of Dairy Households in Rural Punjab, 2005.

Std. Dev

Min

Max

87053.1

900.2

958176

5486.3
10709
0.7693
118.797
0.156
0.51
1380.70

19.6
720
0.0085
5.13
0
1
104

66220.8
131850
9.1882
2811.50
1
4
7488

2.73
13.58
0.324
0.497

1
17
0
0

30
95
1
1

1.06

0.499
0.495
0.433
0.418
0.300

0
0
0
0
0

1
1
1
1
1

0.300
0.300
0.300
0.300
0.300
0.300
0.300
0.300
0.300
0.300

0
0
0
0
0
0
0
0
0
0

1
1
1
1
1
1
1
1
1
1

Feeding molasses to dairy animals is expected to have a positive impact on

productivity. Molasses is a dummy variable that equals one for farms who feed molasses
and zero otherwise. Only 2.5 percent of farms feed molasses to their animals. It is generally
believed that if milching animals are fed sufficient water they yield more milk. But
conventionally, most cows and buffaloes are tied all day due to which they are not free to
drink water at will. Therefore, to gauge the effects on productivity, the frequency of
feeding water to animals is used, which ranges from 1 to 4 times per day with mean value
of 2.34. The labour input includes hired and family labour expressed in hours. The average
use of family and hired labour is 2097 hours, which translates to 40 hours per week ranging
from only 2 hours per week to 144 hours per week. In one sense this is hardly a surprising

Milk Supply Chain and Technical Inefficiency

69

result for a country like Pakistan where small dairy households rarely employ full-time
dedicated workers for day-to-day management of dairy animals. Therefore, family and
hired labour is measured in hours worked per day rather than person-days. In this way, the
likely underemployment of family labour is also discounted for.
Several features of the technical inefficiency model in Equation (3) should be
highlighted. The milk supply chain is the variable of interest, which reflects the status of a
dairy farm and is equal to 1 if the farm is located in the milk supply chain region of the
processing industry, and 0 otherwise. It is noted that 52.5 percent of the sample area is located
in the milk supply chain. In the rest of the sample area, the processing industry is not present
due to which only traditional milk collecting agents are buying farmer milk. The coefficient
on milk supply chain identifies the differential effects of farm location in the milk supply
chain and the non-milk supply chain district on technical inefficiency of the dairy farms.
Another set of important explanatory variables included in the specification of the
technical inefficiency model captures the differential effects on technical inefficiency
attributable to the buyer side market structure. The number of milk processors competing for
farmer milk in a village indicates the extent of imperfect competition in farmer milk market.9
To this end, four dummy variables are introduced. No-player is a dummy variable indicating
that no industry player is present in the mouza due to which the traditional milk collecting
agent (dodhi) enjoys the monopsony power in buying farmer milk. In the study data, 42.5
percent of the respondents sell milk directly to dodhi or other traditional milk collecting agent.
One-player, two-players and three-players indicate presence of one, two or three industry
players (or their agents), respectively competing in a village for the farmer milk. Roughly, 25
percent of the respondents are located in mouzas where one-player is present, 22.5 percent
where two-players are present and 10 percent where three-players are present.
The variable, distance from pucca (metalled) road, is taken as an indicator of
location of mouza. The average distance of dairy farms from pucca road is 0.86 km
where the maximum distance from a farm is 8 km. Because distance from pucca road is
roughly common to all dairy farms in a mouza/village, it also captures some locationspecific unobserved heterogeneity in the sample. Two interactive terms are incorporated
in the model i.e. (milk supply chain distance pucca road, and milk supply chain herdsize) to capture additional effects on technical inefficiency associated with presence of
milk supply chain with distance from pucca road, and herd-size.
Control variables are also introduced to capture variation in technical inefficiency
across farms on account of differences in farm characteristics. Here the relevant variables are
herd-size, head age, depressive disorder and head literate. For the measure of depressive
disorder, an index of depressive disorder is used. The psychiatric epidemiological studies
show that anxiety and depressive disorder is not only common occurrence in Pakistan, but is
also associated with disability [Mirza and Jenkins (2004)]. It is expected that farmers with
major depression to operate at much less than their full potential. Therefore, the degree of
long-term major depression is measured from the number of yes answers to the 20 questions
in WHOs self-reporting questionnaire (SRQ-20). In the present sample, 12 percent of dairy
farmers suffer from major depression measured by 8 or more yes answers to SRQ-20.
9

The market structure is said to be a monopsony when there is a single buyer of fresh milk, e.g.,
traditional rural milk collecting agent. This monopsony market structure closely resembles the picture
prevailing in the non-milk supply chain in Pakistan. When there are two buyers of fresh milk a duopsony is said
to exist; if there are several buyers oligopsony is the proper title.

70

Burki and Khan

Dairy farms located in various districts differ in many characteristics (e.g.

differences in climate, soil conditions, temperature, rainfall and water availability). These
factors might independently affect relative technical inefficiency of dairy farms across
districts and thus bias the estimate of the coefficients in this study. Therefore, a complete
set of all district dummy variables is also taken to control for district fixed-effects.
5. ESTIMATION RESULTS
The frontier production function, Equation (2), and the inefficiency effects,
Equation (3), models are simultaneously estimated using the procedure in computer
programme FRONTIER 4.1 [Coelli (1996)]. The hypothesis testing regarding functional
forms and specifications is conducted on the basis of generalised likelihood ratio tests,10
which have approximately a 2 distribution, except cases where the null hypothesis also
involves the restrictions of = 0. In such cases, the asymptotic distribution of the
likelihood ratio test statistic is a mixed 2 distribution and therefore the appropriate
critical values are drawn from Kodde and Palm (1986). The hypothesis tests are
conducted on the basis of empirical specification in model 1.
An important null hypothesis of interest is whether the Cobb-Douglas production
frontier is an adequate representation of the dairy sector data versus the translog
production frontier model. Table 2 presents the results of the hypothesis test, which
shows that the translog production frontier is rejected in favour of the Cobb-Douglas
production frontier at the 1 percent level of significance. Table 2 also reports the
generalised likelihood ratio test that technical inefficiency effects are absent, or = 0
== 19 = 0, which is strongly rejected at the 1 percent level of statistical significance; it
confirms that most of the dairy farms are operating below the production frontier due to
which the estimated inefficiency of these farms is high. Continuing, the null hypothesis,
= 0, implies that the inefficiency effects are not stochastic, which is rejected at the 1
percent level of statistical significance. Finally, the null hypothesis, H0 : 0 = 1 == 19
= 0, entails that all the explanatory variables in the inefficiency model are jointly zero is
also rejected. This result suggests that the linear explanatory variables accounting for the
sources of technical inefficiency are significant even though the individual parameters of
some variables may not be significant.
Table 2
Generalised Likelihood Ratio Hypothesis Tests
Critical Value
Test
Statistics
( = 0.01)
Null Hypothesis
Decision
H0 : Cobb-Douglas vs. Translog Production
30.58
21.79
Fail to Reject H0
41.02a
512.7
Reject H0
H0 : = 0 = 1 == 19 = 0
6.63a
281.21
Reject H0
H0 : = 0
40.29
315.42
Reject H0
H0 : 0 = 1 == 19 = 0
a
Critical values are taken from Table 1 of Kodde and Palm (1986) using 1 percent level of significance.
10
The generalised likelihood-ratio test is defined by LR= 2 {ln[L(H0)/L(H1)]}= 2{ln[L(H0)]
ln[L(H1)]} where L(H0) and L(H1) denote the values of the likelihood function under the null and alternative
hypothesis, respectively [Coelli, et al. (1998)]. Under the null-hypothesis the test statistic has approximately
chi-square distribution with parameters equal to difference between the parameters involved in the null and
alternative hypothesis.

71

Milk Supply Chain and Technical Inefficiency

(a) Production Frontier Results

The estimated parameters of the stochastic frontier and the technical inefficiency
effects models are presented in Table 3. We begin with model 1 as a parsimonious model
Table 3
Estimation Results for the Frontier Production Function and Inefficiency Model
Variables
Frontier Production Function
Constant
Shed and Structure Capital
Animal Capital
Fodders
Straws and Concentrates
Molasses (Yes=1, No=0)
Feed Water (No. of Times)
Family and Hired Labour
Technical Inefficiency Model
Constant
Herd-size (Number)
Head Age
Head Age2
Depression (if SRQ8=1, Otherwise=0)
Head Literate (Yes=1, No=0)
Distance Pucca Road (km)
Milk Supply Chain (Yes=1, No=0)
Milk Supply Chain Distance Pucca Road
Milk Supply Chain Herd-size

Model 1

Model 2

Model 3

2.933***
(13.32)
0.003
(0.29)
0.886***
(30.35)
0.042**
(2.01)
0.039*
(1.76)
0.053
(0.94)
0.029
(1.23)
0.010
(0.56)

2.899***
(11.81)
0.003
(0.29)
0.892***
(29.89)
0.044**
(2.13)
0.031*
(1.17)
0.052
(0.96)
0.033
(1.40)
0.012
(0.70)

2.93***
(13.49)
0.003
(0.30)
0.885***
(29.32)
0.039*
(1.85)
0.045*
(1.74)
0.048
(0.87)
0.027
(1.19)
0.009
(0.55)

2.246**
(4.41)
0.156***
(13.15)
0.071***
(3.32)
0.001***
(2.83)
0.629***
(4.17)
0.035
(0.38)
0.169***
(3.83)
0.515***
(3.41)

1.901***
(3.76)
0.050
(1.52)
0.063***
(3.12)
0.000**
(2.48)
0.611***
(3.42)
0.038
(0.43)
0.200***
(3.57)
0.054
(0.34)
0.262**
(2.22)
0.117***
(3.51)

2.283***
(4.14)
0.160***
(13.75)
0.075***
(2.99)
0.001***
(2.63)
0.620***
(3.78)
0.038
(0.40)
0.188***
(3.05)

0.751***
(3.26)
Two-players (Yes=1, No=0)

0.115
(0.66)
Three-players (Yes=1, No=0)

1.304***
(2.94)
District Fixed-effects
Yes
Yes
Yes
0.882***
0.769***
0.903***

(5.85)
(5.20)
(4.35)
0.962***
0.958***
0.963***

(126.139)
(116.89)
(107.01)
Log-likelihood
253.57
250.69
249.93
Sample Size
800
800
800
*, ** and *** indicate statistically significant at the 90 percent, 95 percent and 99 percent confidence level, respectively.
One-player (Yes=1, No=0)

72

Burki and Khan

in which the milk supply chain is included as a key variable along with control variables
included in all models. In model 2, it is shown how technical inefficiency of farms
participating in milk supply chain is influenced when they are located in remote areas,
i.e., interaction term milk supply chain distance pucca road, or they have large herdsize, i.e., milk supply chain herd-size. Model 3 explores how increased competition
among the buying networks affects technical inefficiency of dairy farms. The extent of
competition is introduced by four dummy variables ranging from no industry player to
three players present in mouza/village.
The estimated coefficients of the Cobb-Douglas frontier production function
model indicate that all input elasticities possess expected signs and the estimated
coefficients are similar in magnitude in all the specifications. Animal capital, fodder, and
straw and concentrate continue to be the most important determinants of raising output in
smallholder dairy operations, while molasses, feed water, family and hired labour, and
shed and structure capital do not significantly increase dairy output. To illustrate, the
coefficient of animal capital is large, positive and statistically significant indicating that
every 1 percent increase in the value of animal capital results in about 0.89 percent
increase in dairy output.
Similarly, dairy output is statistically significantly correlated with fodder and straw
and concentrate. The estimated fodder, and straw and concentrate elasticities are
relatively much smaller (at approximately 0.042 and 0.039, respectively) and marginally
significant suggesting that these inputs are not much of a limitation. By contrast, shed and
structure capital, molasses, feed water and family and hired labour are not a constraint in
raising dairy production, as suggested by their statistically insignificant coefficients.
While the observed pattern for family and hired labour is explained by disguised
unemployment of family labour, these results suggest that excess supply of straws and
concentrate, and family labour can be used more productively by further expanding the
capacity of the dairy farms (e.g., by purchasing more dairy animals). The policy makers
can help by devising simpler and dairy-friendly credit policies, which may have
substantial potential for dairy development in the country.
The estimated scale elasticity is measured by the sum of all the input elasticities.
The estimated returns to scale is less than one (0.998), and the null hypothesis of constant
returns to scale by using the Wald test is not rejected. In other words, a proportionate
increase in the use of all inputs brings about a proportionate growth in dairy output.
(b) Milk Supply Chain Effects on Dairy Inefficiency
In the technical inefficiency model (Table 3), the dependent variable is measured
in units of inefficiency ranging over the (0, ) interval so that a score of zero indicates
full efficiency and scores of more than zero indicate inefficiency. Likewise, coefficients
with positive signs indicate increase in inefficiency, and vice versa. The estimated
relationships between technical inefficiency and its correlates are qualitatively similar
and robust in all regressions.
It may be noted that model 1 takes milk supply chain as a combined variable
capturing milk supply chain effects plus other control variables. The estimate for
parameter is significantly greater than zero, which suggests that the production frontier
model is a significant improvement over the standard OLS regression model. In model 1,

Milk Supply Chain and Technical Inefficiency

73

the parameter for herd-size indicates that, ceteris paribus, keeping one additional milch
animal significantly decreases technical inefficiency of dairy farms. The negative and
positive coefficients for head-age and age-square predict that, on average, technical
inefficiency of farmers continues to decrease until they reach the age of 36 years and
increases thereafter. The significantly positive coefficient on the dummy variable for
depression indicates higher inefficiency of farmers who suffer from severe long-term
depression.
Farms located in remote areas do not face favourable operating conditions. It
makes intuitive sense when it is found that distance from pucca road is positive and
highly significant. For example, in model 1, the parameter (0.169, t = 3.83) indicates that
technical inefficiency significantly increases with an additional kilometre distance of
dairy farm from pucca road. In other words, we detect that remoteness of dairy farms
clearly has unfavourable effect on technical inefficiency.
The primary interest in this paper is to explore the differential impact of milk
supply chain on technical inefficiency of dairy farms, holding all else as constant. It is
clear from the results that the presence of milk supply chain indeed decreases technical
inefficiency of smallholder dairy farms. The milk supply chain variable11 has a negative
estimated coefficient; this effect is statistically significant at the 1 percent level in model
1. The results suggest that it is important to build supply chains in rural areas if the policy
makers are really interested in increasing productivity and growth of smallholder
producers.
In model 2, the results suggest that while distance from pucca road increases
technical inefficiency (0.200, t = 3.57), building of milk supply chain clearly benefits
dairy households in remote mouzas. For example, the negative and statistically significant
coefficient of the interaction term (0.262, t = 2.22) reveals that building of milk supply
chain tends to decrease inefficiency of dairy farms with their increasing distance from the
pucca road. This is an interesting result since remoteness of rural communities remains a
key feature in many developing countries including Pakistan. Given that local population
in remote rural areas is partially or completely excluded from the facilities available to
the rest of the population, building of milk supply chain in these mouzas enables
producers to reap such benefits as fair prices, weekly payments, transparent milk-grading,
and training in farm management. These services, in turn, help dairy producers to
decrease relative technical inefficiency.
The question arises whether location of dairy farms in milk supply chain
influences their technical inefficiency on the basis of small vs. large herds. The
interaction term (milk supply chain herd-size) in model 2 also allows the differential
effects of milk supply chain to vary by herd-size, holding all else as constant. From the
parameter of the interaction term (0.177, t = 3.51) we further predict that the
inefficiency reducing effect of large herd-size becomes even stronger when farms are
located in the milk supply chain, as suggested by the difference in the two delta
coefficients (0.050 0.177), which is 0.227 and in the same direction. The combined
effect of the two interaction terms suggests that milk supply chain benefits sample dairy
11
Here milk supply chain variable accounts for the possibility that if differential effects associated with
milk supply chain are indeed present then predicted inefficiency should vary across farms in milk supply chain
and non-milk supply chain.

74

Burki and Khan

producers disproportionately more when they are located at a distance from pucca road,
and they maintain relatively larger herds.
Finally, as conditions become more competitive with entry of other industry
players, farmers look for better prices, improved dairy extension services, and more
economical ways to manage their dairy farms. To this end, three dummy variables
(one-player, two-players, and three-players) are introduced in model 3 indicating the
number of milk processors competing for fresh milk in a mouza, while no industry
player is the excluded category. With increase in number of industry players,
technical inefficiency of dairy farms decreases in this sample. The estimated
coefficients for one-player (0.751, t = 3.26) and three-players (1.304, t = 2.94)
are large, negative and statistically significant at the 1 percent level, which indicates
that, on average, dairy farms located in mouzas where one industry player and three
industry players are present are relatively less inefficient than the excluded category.
The difference in the estimated delta coefficient (0.751 1.304) is 2.055,
predicting that improvement in technical inefficiency of farms dealing with three
players is much higher than those dealing with one-player. These results clearly show
that increase in the number of industry players tends to decrease technical
inefficiency of dairy farms. It appears that industry players pay higher prices where
they have more competition in villages. While the statistically insignificant
coefficient for two-players (0.115, t = 0.66) is surprising; it may be blamed on high
collinearity between two-players and district fixed-effects.
(c) Cross-sectional Properties of Technical Efficiency
Table 4 reports summary statistics of the predicted mean technical efficiency
scores derived from the stochastic frontier and technical inefficiency effects models. 12 It
is worth noting that the mean and the median technical efficiency in this sample is 73
percent and 81 percent, respectively, which is comparable to the averages presented by
Bravo-Ureta, et al. (2007) for the stochastic frontier models in the dairy sectors of other
countries. This suggests that an average dairy farm loses about 37 percent of dairy output
due to being technically inefficient.
Farms that participate in formal milk supply chain appear to be far more efficient
than those in non-milk supply chain. Moreover, the standard deviation of technical
efficiency is also relatively lower in a milk supply chain. It shows that farms located in
milk supply chain cluster closely to the production frontier than farms in non-milk supply
chain.
Superior efficiency performance of dairy farms in milk and non-milk supply chain
is also indicated in Figure 1 where the empirical cumulative distribution functions of the
estimated technical efficiency scores are plotted. Further insights are provided in Figure 2
where the frequency distribution of mean technical efficiency of dairy farms in milk and
non-milk supply chain is compared. For the milk supply chain sample, a relatively large
number of dairy farms cluster closely to the higher-end of technical efficiency than at
the lower-end, which is in sharp contrast to the efficiency levels of farms in non-milk
12
The relationship between efficiency (Eff) and inefficiency (ui) is given by Eff = 1/(1+u). Thus a score
of 0 on u implies 100 percent or full-efficiency, and a score of 1 means 50 percent efficiency. Alternatively, u =
( Eff)/Eff. In other words, the 70 percent (or 0.70) efficiency entails 42.86 percent inefficiency.

75

Milk Supply Chain and Technical Inefficiency

Table 4
Descriptive Statistics of Estimated Efficiency of the Dairy Farms
Estimated Efficiency of Farms by

Mean

Median

Std. Dev

Min

Max

Milk Supply Chain

0.794

0.846

0.145

0.096

0.949

420

Not in Milk Supply Chain

0.662

0.727

0.228

0.016

0.961

380

No Industry Player

0.658

0.725

0.231

0.016

0.961

340

One Industry Player

0.783

0.840

0.156

0.263

0.949

200

Two Industry Players

0.776

0.839

0.164

0.096

0.933

180

Three Industry Players

0.809

0.852

0.116

0.332

0.939

80

Herd-size 1-2

0.679

0.753

0.219

0.219

0.961

369

Herd-size 3-4

0.757

0.824

0.178

0.016

0.959

243

Herd-size 5-6

0.779

0.848

0.168

0.024

0.949

108

Herd-size 7-10

0.825

0.879

0.123

0.394

0.955

63

Herd-size 11-15

0.805

0.890

0.204

0.200

0.925

12

Herd-size 16 or More

0.907

0.893

0.028

0.885

0.952

With Major Depression

0.681

0.769

0.218

0.016

0.961

95

Without Major Depression

0.738

0.821

0.197

0.024

0.959

705

Full Sample

0.731

0.813

0.200

0.016

0.961

800

Milk Supply Chain Effects

Farm Characteristics
Herd-size

Farmers Long-term Stress Levels

Source: Authors estimations.

Cumulative Frequency
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Efficiency
Non-milk supply chain
Milk supply chain

Fig. 1. Cumulative Distribution Function for Estimated Technical Efficiency

76

Burki and Khan

Non-milk Supply Chain

250
200
150
100
50
0
0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Efficiency

Milk Supply Chain

250
200
150
100
50
0
0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Efficiency

Fig. 2. Frequency Distribution of Mean Technical Efficiency Levels

supply chain sample. Very few dairy farms in milk supply chain have mean technical
efficiency scores of less than 70 percent. On the contrary, a large number of dairy farms
in non-milk supply chain sample have mean efficiency scores in the range of 20 to 70
percent.
Figure 3 presents the distribution of mean efficiency by mouzas or villages where
mouzas are ranked from best performers to worst performers. It can be seen that 15 of the
top 20 mouzas in our sample are from milk supply chain districts, whereas 13 of the
bottom 20 mouzas are from the non-milk supply chain districts. In general, these findings
tend to corroborate the positive contribution and efficacy of milk supply chain districts in
contributing to increased productive efficiency of smallholder dairy producers.

77

Milk Supply Chain and Technical Inefficiency

Mean Technical Efficiency by Mouza
1.000
0.900
0.800
0.700
0.600
0.500
0.400
0.300
0.200
0.100
0.000
1

11

13

15

17

19

21

23

25

27

29

31

33

35

37

39

Mouza number
Non-milk
chain

supply

Milk supply chain

Fig. 3. Mean Technical Efficiency Levels by Mouza

Table 4 also shows that technical efficiency of the dairy farms is positively correlated
with the number of industry players in a mouza. The highest mean technical efficiency is
achieved when market structure resemble oligopsony (three players) while the lowest mean
technical efficiency is achieved when market structure resembles monopsony (no-player).
Furthermore, the difference in mean and median technical efficiency between two-players and
no-player is statistically significant at the 1 percent level, which corroborates the view that
statistically insignificant coefficient for two-players in Table 3 is indeed explained by the
suspected collinearity between two-players and the district fixed-effects.
Table 4 and Figure 4 (panel A) show that in general technical efficiency is positively
correlated with herd-size. Technical efficiency estimates are more fat-tailed for larger farms.
Major efficiency gains occur when we move from herd-size 12 to herd-size 34. Stacked up
against each other, panels B and C depict efficiency estimates by herd-size for farmers who
participate or do not participate in a milk supply chain. In the milk supply chain, the mean
technical efficiency levels for herd-size 12, 34 and 56 are much higher than otherwise.
Fullsample
Sample
Full

Efficiency
MeanEfficiency
Mean

1
0.8
0.6
0.4
0.2
0
1 to 2

3 to 4

5 to 6

7 to 10

Herd-Size
Herd Size

Panel A

11 to 15

16 and
High

78

Burki and Khan

Non-milk Supply Chain

1.0
0.8
0.6
0.4
0.2
0.0
1 to 2

3 to 4

5 to 6

7 to 10

11 to 15

16 and
High

11 to 15

16 and
High

Herd-Size

Panel B

Milk Supply Chain

1.0
0.8
0.6
0.4
0.2
0.0
1 to 2

3 to 4

5 to 6

7 to 10

Herd-Size

Panel C
Fig. 4. Mean Technical Efficiency Levels by Herd-size
That mental depression is a common occurrence in the dairy sector of rural Punjab
is confirmed by the prevalence of long-term depression in 11.8 percent of the sample
respondents, and the estimated efficiency differentials between those with and without
major depression also corroborates how this disability can cause economic adversity.
Table 4 depicts that the mean and median efficiency index significantly falls for farmers
who report major depression (68 percent and 76 percent) as compared with respondents
with no major depression (74 percent and 82 percent). These results suggest that farmers
without major depression cluster much closer to the frontier compared with those with
major depression.

Milk Supply Chain and Technical Inefficiency

79

6. CONCLUSIONS
This paper provides empirical evidence on how formal participation in a milk
supply chain affects smallholder technical inefficiency. This relationship has been
examined on the basis of survey data of 800 smallholder commercial dairy farmers taken
from milk supply chain and non-milk supply chain districts in Punjab, Pakistan. The
frontier inefficiency effects model and the Cobb-Douglas production technology has been
used to examine the differential impact on relative inefficiency of smallholder dairy
producers. The results show that animal capital, fodder, and straw and concentrate
continue to be most important determinants of raising dairy output, while labour, shed
and structure capital, feeding of water and molasses do not significantly increase dairy
output in our sample. The marginal significance attached to hired and family labour is
attributed to the disguised unemployment of family labour. The scale elasticity estimates
in this study show that if the present trends continue, dairy producers are expected to
bring about a proportionate increase in dairy output with proportionate increase in inputs.
While the location of the dairy households is exogenously determined, the
building of milk supply chain network indeed decreases technical inefficiency of
smallholder dairy households in this sample. Evidence in the present case suggests that
dairy farms located in milk supply chain districts employ fewer resources relative to
those located in non-milk supply chain districts to produce the given output levels. In
considering the mechanism through which a milk supply chain affects technical
inefficiency, the results of this study suggest that it benefits disproportionately those
farms more that are located away from pucca road and are relatively large in size. In
general, remoteness of rural communities remains a key feature in Pakistan where local
population is often excluded from the basic facilities. For the same reason, distance of a
farm from pucca road clearly has unfavourable effect on their technical inefficiency.
Likewise, we find that farms away from pucca road are technically more inefficient, but
this disadvantage tends to decrease significantly when farms are located in a milk supply
chain area. Similarly, it is shown that sample farms with larger herds are less inefficient
than those with smaller herds, yet the inefficiency reducing effect of herd-size becomes
stronger when large farms are located in milk supply chain regions. The study also shows
that increase in the number of industry players buying farmer milk in the supply chain
leads to decrease in technical inefficiency of dairy farms. From the results it is concluded
that technical inefficiency is highest where the market structure resembles monopsony
and lowest where the market structure resembles oligopsony.
If policy makers are indeed interested in increasing productivity and growth of
smallholder dairy producers then they should promote building of supply chains in rural
areas. However, efficiency and productivity gains are far greater if the supply chains also
bring into their fold medium and relatively large farmers based in remote rural areas. The
results in this article further suggest that the buyer-side market structure holds the key for
the success or failure of the emerging agro-food supply chain systems in developing
countries. If anything, the advice to policy makers from these results conforms to the
standard economic view that market competition, which is long viewed as key to
economic development, leads to enhanced levels of technical efficiency of smallholder
producers. Without government intervention in the milk supply chain, profit motive alone
provides incentives to dairy farms to move toward greater efficiency.

80

Burki and Khan

REFERENCES
Ahumada, O. and J. R. Villalobos (2009) Application of Planning Models in the Agrifood Supply Chain: A Review. European Journal of Operational Research 196, 120.
Aigner, D., C. A. K. Lovell, and P. Schmidt (1977) Formulation and Estimation of
Stochastic Frontier Production Function Models. Journal of Econometrics 6, 2137.
Battese, G. E. and T. J. Coelli (1993) A Stochastic Frontier Production Function
Incorporating a Model for Technical Inefficiency Effects. Department of
Econometrics, University of New England, Armidale. (Working Papers in
Econometrics and Applied Statistics No. 69.)
Battese, G. E. and T. J. Coelli (1995) A Model for Technical Inefficiency Effects in a
Stochastic Frontier Production Function for Panel Data. Empirical Economics 20,
325332.
Bravo-Ureta, E. Boris, D. Solis, Victor H. Moreira Lopez, J. F. Maripani, A. Thiam, and
T. Rivas (2007) Technical Efficiency in Farming: A Meta-regression Analysis.
Journal of Productivity Analysis 27, 5772.
Burki, Abid A., Mushtaq A. Khan, and Faisal Bari (2004) The State of Pakistans Dairy
Sector: An Assessment. Pakistan Development Review 43:2, 149174.
Coelli, T. (1996) A Guide to FRONTIER Version 4.1: A Computer Programme for
Frontier Production Function Estimation. Department of Economics, University of
New England, Armidale. (CEPA Working Paper 96/07.)
Coelli, T., D. S. Prasada Rao, and G. E. Battese (1998) An Introduction to Efficiency and
Productivity Analysis. Boston, MA: Kluwer Academic Publishers.
Dolan, C. and J. Humphrey (2000) Governance and Trade in Fresh Vegetables: The
Impact of UK Super markets on the African Horticulture Industry. Journal of
Development Studies 37, 147176.
Dries, L. and J. F. M. Swinnen (2004) Foreign Direct Investment, Vertical Integration,
and Local Suppliers: Evidence from the Polish Dairy Sector. World Development 32,
15251544.
Gorton, M., M. Dumitrashko, and J. White (2006) Overcoming Supply Chain Failure in
the Agri-food Sector: A Case Study from Moldova. Food Policy 31, 90103.
Pakistan, Government of (1999) Pakistan Mouza Statistics 1998. Lahore: Agricultural
Census Organisation.
Gow, H. and J. Swinnen (1998) Agribusiness Restructuring, Foreign Direct Investment,
and Hold-up Problem in Agricultural Transition. European Review of Agricultural
Economics 25, 331350.
Gow, H. and J. Swinnen (2001) Private Enforcement Capital and Contract Enforcement
in Transition Economies. American Journal of Agricultural Economics 83, 686690.
Islam, M. S. (2008) From Pond to Plate: Towards a Twin-driven Commodity Chain in
Bangladesh Shrimp Aquaculture. Food Policy 33, 209223.
Jevsnik, M., V. Hlebec, and P. Raspor (2008) Food Safety Knowledge and Practices
among Food Handlers in Slovenia. Food Control 19, 11071118.
Key, N. and D. Runsten (1999) Contract Farming, Small-holders, and Rural Development
in Latin America: The Organisation of Agroprocessing Firms and the Scale of
Outgrower Production. World Development 27, 381401.

Milk Supply Chain and Technical Inefficiency

81

Khan, M. S. (1997) Predicting Lactation Milk Yield from Last Test Day Yield Buffaloes.
Buffalo Bulletin 16:4, 7880.
Khan, M. S. (2000) Lactation Length Adjustment in Nili-Ravi BuffaloesA New
Procedure. Buffalo Newsletter 14, 45.
Kodde, D. A. and F. C. Palm (1986) Wald Criteria for Jointly Testing Equality and
Inequality Restrictions. Econometrica 54, 12431246.
Meeusen, W. and J. van den Broeck (1977) Efficiency Estimation from Cobb-Douglas
Production Functions with Composed Error. International Economic Review 18, 435
444.
Mirza, I. and R. Jenkins (2004) Risk Factors and Treatment of Anxiety and Depressive
Disorders in Pakistan: Systematic Review. British Medical Journal 328, 794797.
Pinckney, Thomas C. (1989) The Demand for Public Storage of Wheat in Pakistan.
Washington, D.C. (International Food Policy Research Institute Research Report 85).
Pingali, P. (2006) Westernisation of Asian Diets and the Transformation of Food
Systems: Implications for Research and Policy. Food Policy 32, 281298.
Reardon, T. and C. B. Barrett (2000) Agroindustrialisation, Globalisation, and International Development: An Overview of Issues, Patterns, and Determinants.
Agricultural Economics 23, 195205.
Sartorius, K. and J. Kirsten (2007) A Framework to Facilitate Institutional Arrangements
for Smallholder Supply in Developing Countries: An Agribusiness Perspective. Food
Policy 32, 640655.
Stanton, J. V. (2000) The Role of Agribusiness Development: Replacing the Diminished
Role of the Government in Raising Rural Incomes. Journal of Agribusiness 18, 173
187.
Unneveher, L. J. (2000) Food Safety Issues and Fresh Food Product Exports from LDCs.
Agricultural Economics 23, 231240.
Weatherspoon, D. D. and T. Reardon (2003) The Rise of Super Markets in Africa:
Implications for Agrifood Systems and the Rural Poor. Development Policy Review
21, 333356.