Food Policy 75 (2018) 24–36

Contents lists available at ScienceDirect

Food Policy
journal homepage: www.elsevier.com/locate/foodpol

Does cooperative membership improve household welfare? Evidence T

from a panel data analysis of smallholder dairy farmers in Bihar, India
⁎
Anjani Kumara, , Sunil Saroja, P.K. Joshia, Hiroyuki Takeshimab
a
International Food Policy Research Institute, South Asia Office, New Delhi, India
b
International Food Policy Research Institute, Washington, DC, USA

1. Introduction
The promotion of cooperatives is widely viewed as the most im-
portant institutional arrangement for spurring dairy development in
India. The Indian government has made efforts to accelerate the sys-
tematic promotion of dairy cooperatives through financial and policy
support, especially following the launch of Operation Flood in 1970.1
Indeed, much of the success of the White Revolution in India—the
greatly accelerated growth in the country’s milk production since the
1970s—is attributed to the cooperative framework of these dairy de-
velopment strategies. The network of dairy cooperatives has expanded
considerably since then, and in 2015–16 about 16 million dairy farmers
(approximately 20 percent of India’s dairy farmers) were associated
with 180,000 dairy cooperative societies (NDDB, 2016). Nonetheless,
the spread of dairy cooperatives in India has been uneven, and its po-
tential in the context of globalized markets is often questioned. Earlier
studies in India suggest that farmers’ participation in dairy cooperatives
led to significant increases in milk production and yield; a decrease in
the cost of milk production; a reduction in the transaction costs of ac-
cessing inputs, information, technology, and markets; and a realization
of higher prices and profits (Birthal et al., 2017, 2009, 2007; Kumar
et al., 2011a,b; Kumar, 2010; Stockbridge et al., 2003; Berdegue, 2001;
Singh and Pundir, 2000; Shukla and Brahmankar, 1999; Candler and
Kumar, 1998; Singh, 1996; Mergos and Slade, 1987; Singh and Das,
1984; Lele, 1981). A few studies indicate that cooperative membership
also has had a significant impact on compliance with food safety
measures (FSM) (Kumar et al., 2011b,a, 2013; Vandeplas et al., 2013;
Gupta and Roy, 2012; Narrod et al., 2009; Roy and Thorat, 2008).
However, in examining the impact of dairy cooperative membership,
most of these studies did not take measures to correct for sample se-
lection bias and thus may give biased results. For instance, Gupta and
Roy (2012), Kumar et al. (2013), and Kumar et al. (2017a) studies are
all based on cross-sectional data.
This paper aims to contribute to the growing literature on the role,
which dairy cooperatives play in agricultural development by identi-
fying the factors that influence dairy farmers’ decisions to join a dairy
cooperative, and by estimating the impact of dairy cooperative mem-
bership on milk yield,2 net returns per liter, and compliance with FSM.
The study uses panel data from 148 dairy farmers in Bihar, one of the
most important milk-producing states in India, for empirical analysis.
It is important to understand the factors that affect farmers’ deci-
sions to join the cooperatives, and to incorporate such mechanisms into
estimations of the impact of cooperatives. As Section 2 of this paper
describes, despite the findings of positive impacts in the aforemen-
tioned studies, Indian dairy cooperatives’ shares of milk production and
marketing have remained modest. Information on the returns to joining
cooperatives may have been limited, discouraging certain farmers from
joining existing cooperatives or forming new ones. The findings in this
study partly help to fill such gaps in understanding.
This study models a farmer’s choice to join a dairy cooperative as a
selection process, where the expected higher net returns received by the
cooperative members drive farmers’ decisions to join. It uses an en-
dogenous switching regression approach to account for sample selec-
tion bias (Lokshin and Sajaia, 2004). This approach analyzes both the
determinants of cooperative membership and the impact of member-
ship on farm performance indicators such as milk yield, net returns, and
compliance with FSM. The results show that dairy cooperative mem-
bership has a positive and significant effect on milk yield, net returns,
and compliance with FSM. Further, the income gains produced through
dairy cooperative membership are greater for small-scale farmers, in-
dicating the pro-poor tilt of dairy cooperatives.
Section 2 of this study gives an overview of the dairy sector and
dairy cooperatives in Bihar. Section 3 presents the data and the corre-
sponding descriptive statistics. Section 4 presents the analytical
methods, and Section 5 discusses the empirical results. Section 6 pro-
vides conclusions and policy implications of the study.

⁎
Corresponding author.
E-mail address: anjani.kumar@cgiar.org (A. Kumar).
1
Operation Flood was launched in 1970 to accelerate dairy development in India and to ensure that the country could meet the demand for milk and milk products by linking
production to marketing. Operation Flood introduced a three-tiered model of cooperative milk production: (1) milk society, (2) milk union, and (3) milk federation. The operation was
instrumental in accelerating the growth in India’s milk production, to such an extent that this growth was deemed the “White Revolution”.
2
For better comparison, this paper converts milk yield into fat-corrected milk (FCM). The milk from cross-bred cattle, indigenous cattle, and buffalo was converted into 4 percent FCM,
following Hemme et al. (2003). The formula for FCM is as follows: FCM = (milk production × fat (%) × 0.15) + (milk production × 0.4).

https://doi.org/10.1016/j.foodpol.2018.01.005
Received 6 September 2017; Received in revised form 8 January 2018; Accepted 15 January 2018
Available online 22 February 2018
0306-9192/ © 2018 Elsevier Ltd. All rights reserved.
A. Kumar et al. Food Policy 75 (2018) 24–36

Table 1
Key indicators of dairy development in India and Bihar.
Source: Government of India (2016).

India Bihar

2000–2001 2015–16 2000–2001 2015–16

Milk production (’000 metric tons)

Crossbred cows 14,131(17.5) 41,931 (27.0) 99(4.0) 2579 (31.1)
Local cows 18,826(23.4) 31,714 (20.4) 651(26.2) 2302 (27.8)
Buffalo 43,428(53.9) 76,459 (49.2) 1460(58.7) 3210 (38.7)
Goats 4222(5.2) 5378(3.5) 279(11.2) 198(2.4)
Total 80,607 (100.0) 146,314 (100.0) 2489 (100.0) 8288 (100.0)

In-milk animals (’000 s)

Crossbred cows 6011(9.3) 15,412 (16.8) 55(1.9) 1104 (20.3)
Local cows 26,872(41.4) 31,753 (34.7) 1093(37.1) 1941 (35.8)
Buffalo 29,371(45.2) 41,190 (45.0) 1194(40.6) 2059 (37.9)
Goatsa 2674(4.1) 3246(3.5) 602(20.4) 322(5.9)
Total 64,928 (100.0) 91,601 (100.) 2944 (100.0) 5426 (100.0)

Yield (kg/in-milk animal)

Crossbred cows 6.4 7.5 4.9 6.4
Local cows 1.9 2.7 1.6 3.2
Buffalo 4.1 5.1 3.4 4.3
Goats 0.4 0.5 0.1 0.2

Animal health and breeding infrastructure

Veterinary institutions (number)* 52,757 62,422 2259 2717
Number of veterinary institutions per 1000 livestock population 5.4 4.8 10.2 7.0
Artificial insemination (’000 s) 21,324 120,745 22 2624
Percentage of artificial insemination to total livestock population 7.5 40.4 0.10 13.81

Number of dairy processing plants

Cooperatives 212 263 6 10
Private sector 403 765 1 2
Others 63 37 0 0
Total 678 1065 7 12

Processing capacity (’000 l/day)

Cooperatives 28,394 43,251 485 700
Private sector 32,415 73,252 120 400
Others 12,170 3046 0 0
Total 72,979 119,549 605 1100

Note: Percentage figures are given in parentheses.

a
Goats converted into equivalent cattle unit; 10 goats = 1 cattle unit.
* Veterinary institutions include hospitals/polyclinics, dispensaries, and stockmen centers.

2. Dairy sector in Bihar
With a population of about 104 million, Bihar is the third-most
populated state in India, after Uttar Pradesh (200 million) and
Maharashtra (114 million) (Government of India, 2011) It has one of
the lowest per capita incomes and highest incidences of poverty. Dairy
is an important income-generating activity for the rural poor in Bihar; it
comprises about 31 percent of the state’s agricultural gross domestic
product (AGDP) and contributes about three-fourths of livestock gross
domestic product (GDP) (Hoda et al., 2017). Currently, Bihar is one of
the major milk-producing states in India, contributing about 5.3 percent
of national milk production. Production of milk in the state increased
from 2.4 million metric tons in 2000/01 to 8.3 million metric tons in
2015/16, registering a compound annual growth rate of 6.5 percent.
Milk production in Bihar is dominated by smallholder dairy farms.
The average herd size is small, with a few buffalo, dairy cows, or both,
and the herd is reared in a system that is closely integrated with crop
production. In 2013, about 93 percent of milk producers were marginal
or small—having fewer than 2 ha of land—and together contributed
about 90 percent of the total milk production in the state (Kumar et al.,
2017b). Cattle and buffalo are the main milch species, contributing
about 98 percent of the state’s milk production, with the remaining
2 percent contributed by goats (Table 1). Buffalo compose 38 percent of
in-milk animals in Bihar and contribute 39 percent of state milk pro-
duction. Crossbred cattle compose 20 percent of the in-milk animal
population and contribute 31 percent of milk production. Indigenous
cattle, with a 36 percent share of the population, contribute 28 percent
of Bihar’s milk output. The average milk yield of buffalo and cattle in
Bihar is low but has shown significant improvement in recent years.
The Bihar State Milk Cooperative Federation (COMFED) has played
a pivotal role in developing the state’s dairy sector. It was established in
1983 as the implementing agency of the Operation Flood program in
Bihar. COMFED’s cooperative model is a three-tiered structure similar
to the pattern used in Amul Model in Gujarat. The milk producers’
cooperative society is positioned at the village level, the milk union is at
the district level, and the milk federation is at the state level. The
number of dairy cooperative members increased in Bihar from 65,600
in 1987/88 to 629,200 in 2011/12. Milk procurement by COMFED
increased from 71 metric tons per day in 1987/88 to around 1070
metric tons per day in 2011/12, recording a compound annual growth
rate (CAGR) of 13.3 percent (COMFED, 2017a). Despite this robust
growth in milk procurement, at present COMFED procures and pro-
cesses only about 10 percent of the milk produced in Bihar.
Milk marketing has remained a largely informal and highly frag-
mented sector in Bihar, involving local milk vendors, wholesalers, re-
tailers, and neighboring consumers. These informal milk market agents
handle about 74 percent of marketed milk in the state (Kumar, 2010;
Kumar et al., 2017b). Expansion of dairy cooperatives and entry of
formal private sector have bolstered the processing capacity of the or-
ganized dairy sector, but these advancements are inadequate to handle
the marketed surplus generated in the state. There is considerable scope
for the organized sector to participate in milk markets in the state.

25
A. Kumar et al.
Fig. 1. Map showing geographic location of the selected blocks in Patna district, Bihar, India.
3. Data and descriptive statistics
The panel data for this analysis are based on the primary data col-
lected at the farm level in two rounds (2007 and 2015) in Bihar. In both
rounds, we collected data using a thoroughly pretested questionnaire
and trained enumerators. The Patna district of Bihar was selected spe-
cifically for this study because it is well known for milk production and
has several different milk marketing regimes, including dairy co-
operatives, informal traders, and private sector corporate dairy pro-
ducers (Kumar et al., 2011a). Patna district contributed about 6 percent
of the total milk production of Bihar in 2014 (COMFED, 2017b). In
2007, three blocks3 were selected randomly from Patna district (Fig. 1
presents the geographical locations of selected blocks). From each se-
lected block, three villages were randomly selected and 75 dairy
households from each block were selected for the survey (Appendix
Table A.1). At the village level, we randomly selected sample house-
holds in proportion to the village population through the random walk
method. Through this method, a total sample of 225 dairy farmers were
selected in 2007.
The sample households were first stratified into two groups: mem-
bers and nonmembers of dairy cooperatives. Hereafter, to simplify the
discussion, the households are referred to as member and nonmember
households based on their association with dairy cooperatives. In 2015,
we revisited the same 225 households but could trace only 148, as the
remaining households had either migrated or left dairy farming. The
data gathered covered a wide range of information on household
characteristics, milk production, and marketing practices, as well as
compliance with FSM in milk production at the farm level. Appendix
Table A.2 presents summary statistics of variables used in the empirical
model, along with the difference in each explanatory variable for
member and nonmember dairy farmer households.
In 2007, out of 148 dairy farmers, only 21 (14.2%) were members of
dairy cooperative societies. The dairy cooperative membership among
sample households increased to 60 (40.5%) in 2015, a more than
100 percent expansion in cooperative membership. Forty-three dairy
farm households acquired cooperative membership between 2007 and
2015, and four dairy farmers left the cooperatives. To examine the
dynamic effects of dairy cooperative membership further, we per-
formed a descriptive analysis of the 43 households that became
3
A block is an administrative unit within a district.
26
Food Policy 75 (2018) 24–36
members during the study period (Table 2). The analysis assumed that
these households had randomly decided to be cooperative members
based on unobservable characteristics, but the small sample size pre-
cluded any econometric analysis to support this assumption. We found
that milk production appeared to increase significantly after farmers
became dairy cooperative members; their net return per liter more than
doubled, due to higher milk yield.
The average household milk production increased from 3.2 l per day
in 2007 to 6.6 l per day in 2015, mainly due to a more than twofold
increase in milk yield. The share of dairy in household income also
increased from 23 percent in 2007 to 29 percent in 2015, indicating the
increasing role of dairying in small holders’ household income. Further,
the greater extent of marketed milk surplus and the increased partici-
pation of producers in milk markets indicate the enhanced commer-
cialization of these households after becoming cooperative members.
The adoption of FSM also increased from 45.3 percent in 2007 to
52.2 percent in 2015.
4. Methodology
4.1. Adoption of food safety practices at the farm level
To assess the adoption status of FSM at the farm level, this study
identified 42 measures being practiced to ensure milk safety. These
measures were related to the control of chemical and microbiological
hazards. To make a comparative assessment of the adoption status of
FSM compliance, we developed an index of adoption of food safety
practices (FSI) based on weighted scores that align with different
components of food safety.
Weighted scores for the FSI were computed as follows. First, the 42
practices being followed by the dairy farmers were grouped under four
categories: animal health, hygienic milking, hygienic storage, and
maintenance of hygienic premises (including the surrounding en-
vironment). These four categories were accorded weights of 0.25, 0.35,
0.20, and 0.20, respectively, based on their relative importance in en-
suring milk safety and in consultation with scientists working on milk
safety issues (Kumar et al., 2011b, 2017a). The number of practices
being followed in each category was multiplied by the respective
weight and summed across all categories to obtain a weighted score
reflecting the adoption level of FSM. The FSI for ith farm was re-
presented by Eq. (1):
A. Kumar et al. Food Policy 75 (2018) 24–36

Table 2
Descriptive statistics of transition from nonmembers in 2007 to cooperative members in 2015.

Particulars Nonmember households in 2007 Nonmember households in 2007 who became members in 2015 t-value

Number 43 43 –
Age of household head (years) 51.3 56.3 0.0441ns
Household headed by male (%) 97.6 95.3 −0.6417ns
Education of household head (years) 5.7 6.7 1.5360ns
Household size (number) 9.7 9.8 0.4888ns
Dependency ratio 43.8 42.6 −0.0875ns
Size of land holding (hectare) 1.0 0.8 −1.6600*
Herd size (number) 1.5 1.6 0.4301ns
Adoption of crossbred animal (%) 18.7 77.5 8.1151***
Kutcha sheda (%) 51.1 18.6 5.9332***
Pucca/semipucca shedb (%) 48.9 81.4 5.9332***
Annual household income at current price (’000 INR) 80.7 177.4 5.4387***
Annual household income at constant price of 2004 (’000 INR) 70.2 100.1 1.7349*
Annual per capita income at current price (’000 INR) 9.1 19.8 7.4474***
Annual per capita income at constant price of 2004 (’000 INR) 7.8 11.2 1.9985**
Experience in dairying (years) 21.9 28.9 4.1865***
Contribution of dairying to household income (%) 22.7 28.6 3.1748***
Training in dairy farming (%) 0.0 7.0 1.4339ns
Animal value at current price (’000 INR) 22.6 36.7 0.0293ns
Animal value at constant price of 2004 (’000 INR) 19.6 20.7 0.2693ns
Average milk production (liters/day) 3.2 6.6 4.0379***
Milk price at current price (INR) 13.0 31.8 2.2002***
Milk price at constant price of 2004 (INR) 11.5 12.7 41.9283***
Milk-producing households selling milk (%) 30.2 76.7 5.7672***
Share of marketed milk (%) 22.7 56.5 3.5390***
Average milk sold (liters/day/hh) 1.3 5.8 3.7741***
Milk yield (liters/animal/day) 2.1 4.3 6.0560***
Net return per liter at current price (INR) 4.9 10.2 6.3584***
Net return per liter at constant price of 2004 (INR) 4.2 5.8 0.9250ns
Food safety index 0.45 0.52 0.1297ns

Note: hh: household. INR: Indian rupee.

, , and * indicate 1, 5 and 10 percent levels of significance, respectively, and ns represents “not significant.”
*** **

a
A kutcha animal shed is made from bamboo and straw and is considered to be of poorer quality than concrete sheds.
b
A pucca shed is made from concrete and considered to be a good quality shed; a semipucca shed is part concrete and part bamboo/straw.

4
That is, a dairy farmer will opt to be a cooperative member (Mi = 1) ,
FSIi = ∑ wj pj ,
if M ∗ > 0, where M ∗ represents the expected benefits of being a co-
j=1 (1)
operative member compared to not being a member.
where wj is the weight assigned to the jth FSM category, and pj is the Here, X is a vector of variables that determine a dairy farmer’s as-
proportion of FSM practiced in each category. These weights sum to sociation with a cooperative. These variables include the farmer
one, and the FSI ranges between 0 and 1. household’s characteristics, which include age, gender, education,
household size, land size, herd size, dependency ratio, experience in
dairy farming, share of dairy income, per capita income, animal value,
4.2. Impact of dairy cooperatives on farmers’ performance: Endogenous
condition of animal shed, training on livestock, and share of crossbred
switching regression
animals. In the second step, based on the results of the selection func-
tion, two regime equations are specified explaining the outcomes of
Our variable of interest, membership in dairy cooperatives, is based
interest: milk yield, net return per liter, and adoption of FSM. The re-
on individual selections and may be correlated with unobservable
lationship between a vector of explanatory variables X and the outcome
characteristics—such as a farmer’s ability to join or the extent of their
Y can be represented by Y = f(X ) . Specifically, the two regimes are
motivation to join—that also could affect their performance in dairy
represented as follows:
farming. A simple comparison of members and nonmembers may pro-
vide misleading results that attribute differences to the performance of Regime 1: Y1i = X1i β1 + ε1i if Mi = 1
cooperative and independent producers. To account for both en-
dogeneity and sample selection, we use an endogenous switching re- Regime 2: Y2i = X2i β2 + ε2i if Mi = 0, (3)
gression (ESR) framework to estimate the parameters. We account for
where Yi is the outcome of interest (i.e., milk yield, net return per liter,
the endogeneity of the association with cooperative members by esti-
or FSI) in regimes 1 and 2, and Xi represents a vector of the explanatory
mating a simultaneous equations model with endogenous switching by
variables discussed above. Finally, the error terms are assumed to have
full information maximum likelihood (FIML). The decision to be a co-
a trivariate normal distribution, with zero mean and covariance matrix.
operative member is voluntary and may be based on individual self-
If the estimated covariance between δ and ε ’s ( ρ1 and ρ2 , respectively)
selection. The dairy farmers associated with cooperatives may have
are statistically significant, then the farmer’s cooperative membership
inherently different characteristics from the nonmembers, and may
and the farm’s milk yield, net return per liter, or FSI are correlated. The
have decided to be a cooperative member based on the expected ben-
ρ1 and ρ2 are the transformation of the correlation between the errors
efits of membership. We specify the selection equation for association
from Eq. (3). Using this method, we found evidence of endogenous
with a cooperative as
switching and rejected the null hypothesis that sample selectivity bias
1 if Mi∗ > 0 was absent. This model is defined as a “switching regression model with
Mi∗ = Xi α + δi with Mi = ⎧ . endogenous switching” (Maddala and Nelson, 1975).
⎨
⎩ 0 otherwise (2)
It is important to use selection instruments in this model. The

27
A. Kumar et al. Food Policy 75 (2018) 24–36

Table 3 Table 4
Treatment and heterogeneity effect—Decision stage. Impact of dairy cooperatives on milk yield.

Transitional Decision stage Treatment Endogenous switching regression Panel fixed

heterogeneity effects effect model
Member Nonmember Member = 1 Member = 0 (Member = 1,
(Dairy farmers (Dairy farmers otherwise = 0)
Member (a) E (Y1i |Mi = 1) (c) E (Y2i |Mi = 1) TT are are not
Nonmember (d) E (Y1i |Mi = 0) (b) E (Y2i |Mi = 0) TU cooperative cooperative
Heterogeneity effects BH1 BH2 TH members) members)
Dependent Milk yield Milk yield Membership Milk yield
Note: (a) and (b) represent observed expected outcome indicators; (c) and (d) represent variable→
counterfactual expected outcome indicators. Mi = 1 if dairy farm households are co- Column 1 2 3 4
operative members; Mi = 0 if dairy farm households are not cooperative members. Y1i:
Membera 1.445**
Outcome indicators if dairy farm households are cooperative members; Y2i: Outcome
(0.681)
indicators if dairy farm households are not cooperative members. TT: the effect of the
Age in years 0.095 −0.008 −0.008 0.023
treatment (i.e., dairy farm households are cooperative members) on the treated (i.e., dairy
(0.071) (0.024) (0.010) (0.025)
farm households are not cooperative members). TU: the effect of the treatment (i.e.,
Male-headed 0.688 0.649 −0.448 −0.162
cooperative members) on the untreated (i.e., not cooperative members). BHi: the effect of
house- (5.472) (1.391) (0.492) (1.268)
base heterogeneity for dairy farm households that are cooperative members (i = 1), and holda
not associated with cooperative members (i = 2). TH = (TT − TU) (i.e., transitional Education in 0.140 −0.039 −0.018 0.033
heterogeneity). years (0.148) (0.049) (0.023) (0.060)
Household 0.210 0.057 −0.048* 0.058
selection of instrumental variables should directly affect the selection size (0.157) (0.072) (0.027) (0.067)
Dependency 1.396 −1.348 −0.586 −0.605
variable but not the outcome variable. In this study, we used caste as a
ratio (3.044) (0.985) (0.521) (1.015)
selection instrumental variable. We established the admissibility of the Land holdings −0.588* 0.601*** 0.072 −0.000
instruments by performing a simple falsification test: if a variable is a in (0.340) (0.195) (0.063) (0.253)
valid selection instrument, it will affect the households of dairy farmers hectares
that are associated with a dairy cooperative but will not affect the Herd size 2.653*** 2.360*** 0.129 3.011***
(0.621) (0.457) (0.091) (0.403)
outcome variable of the households of dairy farmers that are not as-
Share of −0.021 0.003 0.007** 0.004
sociated with a dairy cooperative (Appendix Table A.3). crossbred (0.030) (0.008) (0.003) (0.007)
In addition to using the endogenous switching regression model, we animals
calculated the dairy farmers’ conditional expectations for milk yield, Experience in −0.050 −0.012 −0.015 −0.023
dairy (0.056) (0.018) (0.010) (0.018)
net return per liter, and adoption of the FSM in the four cases presented
farming
below: in years
In addition to using the endogenous switching regression model, we Share of dairy 0.052 0.034*** 0.002 0.041***
calculated the dairy farmers’ conditional expectations for milk yield, income in (0.052) (0.011) (0.005) (0.013)
net return per liter, and adoption of the FSM in the four cases presented total
income
below:
Training in −4.978 0.401 0.695 −1.768
livestocka (3.247) (1.982) (0.819) (1.742)
⎡ ⎤ 2.758*** −0.208 0.844**
a. E (Y1i |Mi = 1) = ⎢ ∑ (X1i β1 + σ1n γ1i ) ⎥ N1 Log of per 0.258
⎣ Mi = 1 ⎦ (4) capita (0.955) (0.344) (0.156) (0.373)
income
Log of animal 1.104 0.495 0.212 0.646*
⎡ ⎤
b. E (Y2i |Mi = 0) = ⎢ ∑ (X2i β2 + σ2n γ2i ) ⎥ N0 values (1.143) (0.356) (0.138) (0.361)
⎣ Mi = 0 ⎦ (5) Year dummya −2.776 2.596*** 1.452*** 1.401**
(2.541) (0.723) (0.318) (0.683)
General caste 1.368***
⎡ ⎤
c. E (Y2i |Mi = 1) = ⎢ ∑ (X1i β2 + σ2n γ1i ) ⎥ N1 (0.288)
⎣ Mi = 1 ⎦ (6) Constant −36.600* −8.861** −1.036 −15.950***
(20.249) (4.112) (1.734) (5.350)
Block fixed Yes Yes
⎡ ⎤
d. E (Y1i |Mi = 0) = ⎢ ∑ (X2i β1 + σ1n γ2i ) ⎥ N0. effect
⎣ Mi = 0 ⎦ (7) Observations 296 296 296 296
σi 1.706*** 1.116***
N1 and N0 are the number of observations with Mi = 1 and Mi = 0 , re- (0.230) (0.019)
spectively. ρj −0.750 0.207
Cases (a) and (b) in Table 3 represent the actual expectations ob- (1.009) (0.178)

served in the sample. Cases (c) and (d) represent the counterfactual
Note: Robust standard errors in parentheses.
expected outcomes. However, following Heckman et al. (2001), we a
Denotes dummy variable.
calculated the effect of the treatment “cooperative members” on the *** p < .01.
treated (TT) as the difference between (a) and (c), which represents the ** p < .05.
impact of members on the outcome variable of the dairy farm house- * p < .1.
holds that are cooperative members. Similarly, we calculated the dif-
ference between (d) and (b) as the effect of the treatment on the un-
treated (TU) for the dairy farm households that did not become
cooperative members. (Carter and Milon, 2005).
We also defined the “the effect of base heterogeneity” for the group Finally, we examined the transitional heterogeneity (TH)—namely,
of dairy farm households that decided to become cooperative members whether the effect of a cooperative membership on the outcome vari-
as the difference between (a) and (d). For the group of dairy farm able is larger or smaller for the dairy farm households that are co-
households that decided not to become cooperative members, the effect operative members than for those that are not cooperative members in
of base heterogeneity was defined as the difference between (c) and (b) the counterfactual case (i.e., the difference between TT and TU).

28
A. Kumar et al. Food Policy 75 (2018) 24–36

Table 5 Table 6
Impact of dairy cooperatives on net return per liter. Impact of dairy cooperatives on adoption of FSM.

Endogenous switching regression Panel Endogenous switching regression Panel fixed

fixed effect
Member = 1 Member = 0 (Member = 1, effect Member = 1 Member = 0 (Member = 1, model
(Dairy farmers (Dairy farmers otherwise = 0) model (Dairy farmers (Dairy farmers otherwise = 0)
are cooperative are not are are not
members) cooperative cooperative cooperative
members) members) members)
Dependent Net return per Net return per Membership Net Dependent FSI FSI Membership FSI
variable→ liter liter return variable→
per liter Column 1 2 3 4
Column 1 2 3 4
Membera 0.103**
Membera 0.242*** (0.048)
(0.093) Age in years 0.160 0.070 −0.015 0.001
Age in years −0.005 −0.003 −0.014 −0.002 (0.136) (0.066) (0.009) (0.001)
(0.004) (0.005) (0.010) (0.004) Male-headed −10.157 −1.944 −0.623 −0.114
Male-headed 0.129 0.040 −0.584 0.087 house- (7.499) (2.391) (0.497) (0.083)
house- (0.170) (0.186) (0.547) (0.119) holda
holda Education in 0.430 0.567*** −0.037 0.011***
Education in 0.000 0.021 −0.034 0.009 years (0.403) (0.155) (0.026) (0.004)
years (0.011) (0.016) (0.026) (0.010) Household size 0.685 −0.139 −0.044 0.002
Household size −0.003 0.001 −0.057* −0.002 (0.447) (0.218) (0.027) (0.005)
(0.016) (0.012) (0.030) (0.010) Dependency 8.802 0.100 −0.575 0.027
Dependency 0.114 0.061 −0.474 −0.002 ratio (7.487) (3.350) (0.558) (0.078)
ratio (0.261) (0.217) (0.605) (0.170) Land holdings 0.855 0.356 0.053 0.010
Land holdings −0.034 −0.077 0.066 −0.050* in hectares (0.667) (0.687) (0.051) (0.009)
in hectares (0.023) (0.059) (0.054) (0.026) Share of −0.004 −0.035* 0.008*** −0.000
Experience in 0.002 0.003 −0.008 0.002 crossbred (0.050) (0.020) (0.003) (0.000)
dairy (0.004) (0.004) (0.010) (0.003) animals
farming^ Experience in −0.149 −0.082 −0.012 −0.003**
Share of 0.001 −0.001 0.007** −0.001 dairy (0.112) (0.063) (0.009) (0.001)
crossbred (0.002) (0.002) (0.003) (0.002) farming in
animals years
Share of dairy 0.006*** 0.001 0.002 0.003 Share of dairy −0.049 −0.140*** 0.002 −0.004***
income in (0.002) (0.002) (0.005) (0.002) income in (0.085) (0.037) (0.005) (0.001)
total total
Training in 0.101 0.328 0.860 0.125 income
livestocka (0.183) (0.423) (0.926) (0.141) Training in 9.610 −12.616*** 1.111 0.138
Milk 0.021 0.048*** 0.048** 0.030*** livestocka (7.558) (4.050) (0.850) (0.116)
production (0.014) (0.009) (0.024) (0.009) Log of per −1.390 −0.930 −0.284* −0.015
in liters capita (2.606) (1.068) (0.153) (0.025)
Log of animal −0.045 −0.101 0.152 −0.055 income
values (0.068) (0.087) (0.147) (0.069) Log of animal −3.006 2.744*** 0.169 0.049**
Log of per 0.014 0.043 −0.186 0.043 values (2.479) (0.995) (0.131) (0.022)
capita (0.066) (0.068) (0.173) (0.057) Milk 0.652* 0.767*** 0.040* 0.014***
income produc- (0.348) (0.223) (0.021) (0.003)
Year dummya 0.758*** 0.662*** 1.325*** 0.741*** tion in
(0.180) (0.168) (0.333) (0.138) liters
General caste 1.534*** Year dummya 1.017 −1.474 1.355*** 0.022
(0.296) (5.750) (2.407) (0.286) (0.047)
Constant 1.978 **
2.219 **
−0.211 1.351 General caste 1.599***
(0.986) (1.126) (1.865) (0.894) (0.311)
Block fixed Yes Yes Block fixed Yes Yes
effect effect
Observations 270 270 270 270 Constant 88.096*** 21.475* 0.605 3.564***
σi −1.141*** −0.431** (31.836) (12.098) (1.885) (0.267)
(0.301) (0.182) Observations 296 296 296 296
ρj −0.081 −0.092 σi 2.625*** 2.272***
(1.092) (0.167) (0.159) (0.034)
ρj −0.690 −0.494
Note: Robust standard errors in parentheses. (0.745) (0.378)
a
Denotes dummy variable.
*** p < .01. Note: Robust standard errors in parentheses.
a
** p < .05. Denotes dummy variable.
* p < .1. *** p < .01.
** p < .05.
* p < .1.
5. Results and discussion

Tables 4–6 report the estimates of the ESR model estimated by
FIML. As indicated earlier, the FIML approach jointly estimates both the
selection and the outcome equations. The outcome equations that re-
present the impact of cooperative membership on milk yield, net re-
turns, and compliance with FSM for both members and nonmembers
are given in the second and third columns of Tables 4–6, respectively.
The selection equations that represent the determinants of dairy
cooperative membership are given in the fourth column of Tables 4–6.
Being estimated jointly with the outcome equations through FIML, es-
timated results for selection equations differ slightly across Tables 4–6,
even though both the dependent variable and the set of explanatory
variables are the same. The final column of Tables 4–6 provide the
estimated impact of cooperative membership based on the coefficients
of ordinary least squares (OLS) regressions.

29
A. Kumar et al. Food Policy 75 (2018) 24–36

Table 7 Table 8
Treatment and heterogeneity effects. Impact of dairy cooperative membership on milk yield, net return, and FSM adoption by
farm size.
Member Nonmember Treatment effects
Farm category Mean outcome ATT t-value Change in %
Milk yield
Member (a) 4.09 (c) 2.93 TT = 1.16*** Members Nonmembers
Nonmember (d) 2.07 (b) 1.80 TU = 0.27***
Heterogeneity effects BH1 = 2.02 BH2 = 1.13 TH = 0.89*** Milk yield
Landless 5.2 3.3 1.9 1.7102* 57.6
Net return per liter
*** Marginal 8.3 4.3 4.0 3.3245*** 93.0
Member (a) 7.16 (c) 5.20 TT = 1.96
Small and large 8.5 4.7 3.8 2.5282** 80.9
Nonmember (d) 4.87 (b) 3.52 TU = 1.35***
Heterogeneity effects BH1 = 2.29 BH2 = 1.68 TH = 0.61*** Net return per liter
Landless 8.4 8.0 0.4 0.1176ns 4.7
Adoption of FSM
*** Marginal 12.5 7.0 5.4 2.7300*** 77.1
Member (a) 0.43 (c) 0.39 TT = 0.04
Small and large 10.0 5.3 4.7 1.9171* 89.2
Nonmember (d) 0.45 (b) 0.42 TU = 0.03***
Heterogeneity effects BH1 = −0.28 BH2 = −0.29 TH = 0.01*** Adoption of FSM
Landless 0.44 0.36 0.8 2.1391** 20.4
*** Significant at the 1 percent level. Marginal 0.53 0.44 0.9 3.251*** 20.4
Small and large 0.56 0.49 0.7 2.185** 12.6

In the next section, we first discuss the determinants of dairy co-

operative membership based on the selection equations in Tables 4–6.
We then discuss the impact of cooperative membership on milk yield,
net returns, and compliance with FSM, and highlight our unique find-
ings on how FSM affects the effects of different factors on these out-
comes. Finally, we present the estimates for the average treatment ef-
fects on the treated (ATT) in Table 7.
5.1. Determinants of dairy cooperative membership
The results of the participation equation given in the third column
of Tables 4–6 suggest that the drivers of dairy farming households’
decisions to become cooperative members include herd composition,
milk production, caste affiliation, household size, and per capita in-
come. A higher share of crossbred animals, higher milk production, and
belonging to a general caste category all positively influence house-
holds’ decisions to join dairy cooperatives. But per capita income and
herd size have a negative influence on the probability of becoming
cooperative members. The probability of participation in dairy co-
operatives increased significantly between 2007 and 2015, as indicated
by the positive and significant coefficient of the year dummy variable.
The land size, age, gender, household dependency ratio, and regional
location variables do not significantly affect the probability of be-
coming a dairy cooperative member.
Note: ***, **, and * indicate 1, 5 and 10 percent levels of significance, and ns represents
“not significant.”
Land holding had a positive effect on the milk yield for nonmember
farmers. Herd size, year, and share of dairy as a part of household in-
come also are positively associated with milk yield for nonmember
dairy farmers. For cooperative dairy farmers, herd size and per capita
income is positively associated with the milk yield, whereas land
holding had a negative effect on milk yield.
These differences may arise for multiple reasons. For instance, even
though nonmembers may continue to rely on extensive livestock
rearing, cooperative members may switch to more intensive rearing in
which a higher density of livestock per land holdings may reduce the
costs of monitoring livestock conditions and providing forages and in-
tensive cares. Similarly, for cooperative members, higher per capita
incomes may translate into higher milk yield because their production
may be more capital intensive, whereas for nonmembers production
may be more labor intensive. In the latter group, higher milk yield is
achieved if more family members are engaged in milk production as
opposed to other income-earning activities, so that their opportunity
costs are lower and dairy income is a higher share of total income.
5.3. Net return effects

5.2. Milk yield effects
The fourth column (OLS approach) of Table 4 investigates the effect
of cooperative membership on milk yield. This approach suggests that
there is a significant difference in the milk yield between members and
nonmembers of dairy cooperatives. The coefficient of the member
dummy variable is positive and statistically significant. However, the
OLS approach assumes that membership in a cooperative is exogen-
ously determined, even though such membership is endogenously de-
termined. Therefore, using the OLS approach here will lead to biased
and inconsistent estimates. Further, the OLS-derived estimates do not
account for structural differences between the outcome variables of
cooperative members and their counterparts. To correct for these
weaknesses, the results presented in the first and second columns of
Table 4 provide estimates for endogenous switching in the milk-yield
function. The estimated coefficient of correlation ( ρj ) is not statistically
significant in either function. This finding suggests that we failed to
reject the null hypothesis that sample selectivity bias was absent in both
equations. Nevertheless, we found a difference between the coefficient
of the milk-yield function in members of dairy cooperatives and their
counterparts, indicating the presence of heterogeneity in the sample.
The milk yield is significantly higher for cooperative members than for
nonmembers.
The fourth column (OLS approach) of Table 5 investigates the effect
of cooperative membership on net return. As mentioned earlier, this
approach suggests that there is a significant difference in the net return
between members and nonmembers of dairy cooperatives. The coeffi-
cient of the member dummy variable is positive and statistically sig-
nificant, but the estimated coefficient of correlation ( ρj ) is not statisti-
cally significant in either function. However, we found a difference
between the coefficient of the net return function in members of dairy
cooperatives and their counterparts, indicating the presence of het-
erogeneity in the sample. The scale of milk production and the year had
a positive effect on the net return for nonmember farmers. This may be
the case because nonmembers must have sufficient production scale in
order to attract buyers that are more willing to pay for their products, as
well as to reduce the transactions costs per unit of milk output; co-
operative members, by contrast, can do so through jointly with other
members. In comparison, for cooperative dairy farmers the share of
dairy income and year are positively associated with the net return.
This may be the case because cooperative members may engage more in
quality improvement processes (including FSM, as described in the next
subsection) that may be more labor intensive, and households that have
greater dairy income shares may have more family members available
for such quality improvement activities; nonmembers may be less likely
to be engaged in those activities.

30
A. Kumar et al.
5.4. FSM adoption effects
Similarly, the fourth column of Table 6 investigates the effect of
cooperative membership on FSI score. The coefficient of the member
dummy variable is positive and statistically significant. The estimated
coefficient of correlation ( ρj ) is not statistically significant in either
function. This finding suggests that we failed to reject the null hy-
pothesis that sample selectivity bias is absent in both equations.
Nevertheless, we found a difference between the coefficient of the FSI
function in members of dairy cooperatives and their counterparts, in-
dicating the presence of heterogeneity in the sample. The adoption of
FSM is significantly higher for cooperative members than for non-
members. Years of education has a positive effect on the adoption of
FSM for nonmember dairy farmers. The value of livestock assets and the
scale of milk production is also positively associated with the adoption
of FSM for nonmember farmers. Yet the share of crossbred animals,
share of dairy income, and training in livestock have a negative impact
on FSM adoption for nonmembers. This may be the case because
crossbred animals may require more complex appropriate FSM. Simi-
larly, FSM may be too costly and incur losses for nonmembers, because
unlike cooperative members they may not be able to fetch higher prices
from FSI (particularly if production quantities are small) and thus those
who rely more on dairy incomes for their livelihood may be less likely
to engage in FSI. Relatedly, training in livestock may be affecting the
returns on other types of livestock-related investments, raising the re-
lative costs for investments in food safety. For cooperative dairy
farmers, the scale of milk production is positively associated with FSM
adoption.
These differences may be driven by the fact that, whereas non-
members’ engagement with FSM may be constrained by various factors,
cooperative membership largely may remove these constraints. For
example, the findings suggest that, among nonmembers, the adoption
of FSM is constrained by low education levels and a higher share of
crossbred animals in their herd size.
5.5. Treatment effects
Table 7 presents the expected value of milk yield, net return per liter
of milk, and adoption of FSM under actual and counterfactual condi-
tions. Cells (a) and (b) represent the expected value of outcome vari-
ables. The expected values for milk yield for members of dairy co-
operatives was higher than for nonmembers. This simple comparison,
however, could be misleading in attributing the different values of milk
yield to cooperative membership. The last column of the first panel in
Table 7 presents the treatment effects of cooperative membership on
milk yield at the farm level. In the counterfactual case (c), dairy farmers
who became cooperative members would have lower yield by 1.16 l per
day if had they not become members. The positive mean difference of
(d) and (b) elicits a similar conclusion: nonmember dairy farm house-
holds would have increased the milk yield of their cattle by 0.27 l per
day if they had become members. However, the transitional hetero-
geneity effect for milk yield is positive, meaning that the effect is
greater for cooperative dairy farm households than for nonmember
dairy farming households.
The impact of cooperative membership on net return per liter is also
positive and significant. Milk-producing households gained from being
associated with cooperatives; their mean net return per liter was
37 percent higher than it would have been if they had not become
cooperative members. The positive mean difference of (b) and (d) in-
dicates that the nonmember households would have had higher net
returns per liter (38 percent) if they had become members.
Similarly, the expected level of FSM adoption for members (0.43) of
dairy cooperatives was higher than that for nonmembers (0.39). The
last column of the third panel in Table 7 presents the treatment effects
of cooperative membership on adoption of FSM at the farm level. In the
counterfactual case, dairy farmers who became cooperative members
31
Food Policy 75 (2018) 24–36
would be 4 percent more likely to have lower compliance with FSM if
they had not become members. Nonmember dairy farm households, by
contrast, would only have been 3 percent more likely to adopt FSM if
they had been members.
To gain further understanding of the impact of cooperative mem-
bership on different groups of farmers, we also examined the differ-
ential impact of membership by dividing households into quantiles
based on farm size (Appendix Table A.4–A.6). Table 8 shows that the
marginal farmers (operating on less than 1 ha of land) turned out to be
the greatest beneficiaries of dairy cooperative membership.
6. Conclusion
This paper examines the factors that influence dairy farmers’ deci-
sions to become members of milk cooperative societies, and the impact
of cooperative membership on milk yield, net returns per liter, and
adoption of FSM. The study used farm-level panel data collected from a
randomly selected sample of 148 households. Simple comparisons of
average milk yield, net returns per liter, and FSM adoption revealed
significant differences between members and nonmembers of dairy
cooperatives. As these comparisons are merely descriptive and do not
account for confounding factors that affect the differences, we em-
ployed an ESR model that accounts for both observed and unobserved
factors to consider the issue of selection bias. The results revealed that
sample selection bias would result if the outcome specifications were
estimated without considering the membership decision.
The empirical results showed a positive and significant relationship
between dairy cooperative membership and milk yield, net returns per
liter, and adoption of FSM. In particular, association with a dairy co-
operative society tends to increase milk yield by 40 percent, net return
by 38 percent, and adoption of FSM by 10 percent. The estimates, dif-
ferentiated by farm size, revealed that the income gains brought about
by dairy cooperative membership were higher for small-scale farmers.
This finding suggests that dairy cooperatives can play a significant role
in enhancing the household income of smallholder dairy farmers. The
study also revealed that dairy cooperatives have the potential to en-
hance milk yield and net returns and improve farmers' compliance with
FSM. Therefore, the government should further strengthen and promote
the expansion of dairy cooperatives in India.
Future studies should address a number of issues related to this
research. First, even though the analyses in this study are internally
valid, their external validity is yet to be investigated. For example, milk
prices might have been considerably different during the study period
than in other periods, and milk market structures also have changed
considerably during the past two decades, as was described in Table 1.
Our analyses may not offer insights into what would have happened
during the period or in areas where the changes in these market char-
acteristics were more stable. Future studies must address these issues,
such as by examining similar issues in a broader set of locations across
India or in different time periods. Such investigations are important
also because milk price levels and their growth might have also partly
affected the estimated impact of cooperative membership in ways that
are not always clear. For instance, higher milk prices could lower the
relative margins gained by the quality improvements induced by co-
operative membership, if the price premiums from such improvements
are relatively fixed. However, if price premiums from quality im-
provements were proportional to the milk price, then the benefits of
joining the dairy cooperative would have been greater.
Second, future studies need to investigate more closely the moti-
vations to join cooperatives, as well as perceptions of expected out-
comes and barriers for or costs to joining cooperatives. Such informa-
tion can help researchers and policymakers assess the transaction costs
associated with joining cooperatives, such as the roles of insufficient
knowledge of expected returns. For instance, some studies (for example,
of dairy and fishery cooperatives in the United States) have suggested
that part of the motivation for forming cooperatives was to improve
A. Kumar et al. Food Policy 75 (2018) 24–36

product quality by coordinating quality control across different stages Acknowledgement

in the production process (Sexton and Iskow 1988; Jardine et al. 2014).
Whether this finding holds true for dairy farmers in India as well, fur-
ther research will provide more insights into the understanding of the
role of cooperatives across diverse settings.
Finally, future studies need to investigate the mechanisms of the
observed impact pathways, and how different characteristics of co-
operatives were responsible for impacts such as membership size,
membership policy, methods of finance, and decision-making pro-
cesses. Identifying the mechanisms of such impact pathways will shed
further light on what types of cooperatives (with various structures)
may be particularly effective.
We are grateful to the Indian Council of Agricultural Research
(ICAR) and United States Agency for International Development
(USAID) for extending financial support to conduct this study. We
would like to thank the farmers who provided the needed information
for this study. We also express our sincere thanks to the anonymous
reviewer and editor of the journal for providing valuable comments to
improve earlier version of the paper. Any opinions, findings, conclu-
sions, or recommendations expressed within this paper do not ne-
cessarily reflect the views of the ICAR, USAID, IFPRI or any other or-
ganizations.

Appendix A.

See Tables A.1–A.6.

Table A.1
Distribution of sample households and transition to dairy cooperative membership.

Sample No. Block name No. of Old No. of matched % of matched

villages sample in households in households
2007 2015

1 Bikram 3 75 54 72.0
2 Mokama 3 75 57 76.0
3 Phulwari 3 75 37 49.3
Shariff
Total 9 225 148 65.8

Transition to dairy cooperative membership between 2007 and 2015

2007 2015

Transition Member Nonmember

Member 21 17 (81.0) 4 (19.0)

Nonmember 127 43 (33.9) 84 (66.1)
Total 148 60 (40.5) 88 (59.5)

Note: Percentage figures are given in parentheses.

Table A.2
Summary statistics comparing member and nonmember dairy farmer households.

Particulars 2007 t-value 2015 t-value t-value

M NM All (2007) M NM All (2015) Year

M - NM M - NM 2015–2007

Number 21 127 148 – 60 88 148 – –

Explanatory variables
Age of household head (years) 52.3 49.5 49.9 0.9737ns 57.0 58.1 57.6 −0.5493ns 5.4676***
Household headed by male (%) 95.2 97.6 97.3 −0.6248ns 95.3 96.4 97.3 −0.3389ns − 0.6417ns
Education of household head (years) 9.7 5.2 5.8 3.7160*** 7.8 6.1 6.8 1.9189* 1.5843ns
Household size (number) 8.5 9.1 9.0 −0.5009ns 9.4 8.2 8.7 1.5899ns −0.5897ns
Dependency ratio 32.3 45.5 43.6 −2.756*** 41.1 44.9 43.4 −1.0517 0.0875ns
Size of land holding (hectare) 3.7 0.7 1.1 5.9781*** 1.3 0.3 0.8 4.0234*** −1.5454ns
Herd size (number) 2.3 1.4 1.6 3.2646*** 1.6 1.4 1.5 0.9631ns −0.4264ns
Adoption of crossbred animal (%) 49.0 13.0 18.1 4.3796*** 78.1 41.1 57.1 5.1908*** 7.9653***
Kutcha shed (%) 23.8 54.3 50.0 2.6341*** 16.7 20.5 18.9 0.5744ns 5.9332***
Pucca/semipucca shed (%) 76.2 45.7 50.0 2.6341*** 83.3 79.5 81.1 0.5744ns 5.9332***
Experience in dairying (years) 24.0 20.0 20.5 1.3082ns 18.7 23.1 21.2 −2.0781** 0.4090ns
Contribution of dairying to household income (%) 26.7 24.8 25.1 0.3911ns 27.9 37.1 35.9 −2.3571** 3.1116***
Training in livestock (%) 4.8 0.8 1.4 1.4620ns 8.3 1.1 4.1 2.2004** 1.4339ns
Annual per capita income at current price (’000 INR) 11.8 7.8 8.4 2.9849*** 19.5 17.2 18.1 0.9405ns 7.4474***
Annual per capita income at constant price of 2004 (’000 INR) 10.3 6.8 7.3 2.9849*** 11.1 9.7 10.3 0.9405ns 3.6536***
Animal value at current price (’000 INR) 40.3 21.7 24.3 3.3973*** 37.3 15.7 24.5 4.5429*** −0.0293ns
Animal value at constant price of 2004 (’000 INR) 35.1 18.9 21.1 3.3973*** 21.1 8.9 13.8 4.5429*** 3.3260***
Household belonging to general caste (%) 100.0 28.4 38.5 7.2364*** 65.0 20.5 38.5 6.0790*** –
(continued on next page)

32
A. Kumar et al. Food Policy 75 (2018) 24–36

Table A.2 (continued)

Particulars 2007 t-value 2015 t-value t-value

M NM All (2007) M NM All (2015) Year

M - NM M - NM 2015–2007

Average milk production (liters/day) 8.1 2.8 3.7 5.1594*** 7.1 5.2 6.1 1.8506* 4.1810***
Annual household income at current price (’000 INR) 96.1 68.7 72.5 1.9151* 162.6 154.1 157.6 0.2795ns 5.4387***
Annual household income at constant price of 2004 (’000 INR) 83.6 59.7 63.1 1.9151* 91.9 87.1 89.0 0.2795ns 2.7422**
Milk-producing households selling milk (%) 66.7 33.9 38.5 2.9247*** 78.1 64.3 70.2 1.8332* 5.7672***
Share of marketed milk (%) 48.1 26.4 29.5 3.8887*** 56.0 45.5 50.0 1.7258* 3.5294***
Average milk sold (liters/day/hh) 6.3 1.3 2.0 5.4023*** 5.8 3.4 4.5 2.0375** 3.7741***
Milk price at current price (INR) 13.0 13.0 13.0 – 31.9 31.7 31.8 0.8786ns 2.9871**
Milk price at constant price of 2004 (INR) 11.5 11.5 11.5 – 12.8 12.6 12.7 0.8786ns 36.5881***

Outcome variables
Milk yield (liter/animal/day) 3.9 1.9 2.3 4.4146*** 4.5 3.7 4.1 1.5961ns 6.2303***
Net return per liter at current price (INR) 6.8 4.0 4.4 2.3964** 12.5 11.0 11.7 0.6445ns 6.2625***
Net return per liter at constant price of 2004 (INR) 5.9 3.5 3.8 2.3964** 6.8 6.4 6.6 0.6445ns 3.9634***
Adoption of FSM 0.46 0.43 0.43 0.1102ns 0.53 0.41 0.46 0.4749*** 0.1866*

Note: NA: Not applicable. M: Member. NM: Nonmember. hh: household. INR: Indian rupee.
, , and * indicate 1, 5 and 10 percent levels of significance, respectively, and ns represents “not significant.”
*** **

Table A.3
Parameters estimates—Validity test of the selection instrument.

Model 1 Model 2 Model 3 Model 4

Parameter Member Milk yield of Net return of Adoption of
estimates (1 = member, 0 nonmember nonmember FSM of
= otherwise) nonmember

General caste 1.265*** 0.410 −0.208 −1.200

(0.169) (0.355) (0.147) (1.871)
Constant −1.199*** 2.557*** 1.965*** 37.612***
(0.122) (0.194) (0.066) (1.653)
Wald test on Chi (2) = F-stat = F-stat = F-stat =
instrumental 55.89*** 1.33 2.00 1.72
variable
Observations 296 215 194 215

Table A.4
Impact of dairy cooperatives on milk yield across farm size.

Dependent variable—Milk yield Farm size

Landless Marginal Small and large

Membera 0.953 1.189 −0.320

(1.299) (0.995) (1.307)
Age in years 0.009 0.037 −0.037
(0.038) (0.044) (0.059)
Male-headed householda 1.056 −2.151** −30.002**
(1.875) (1.006) (12.867)
Education in years −0.168* −0.017 0.023
(0.090) (0.064) (0.161)
Household size 0.121 0.241*** −0.012
(0.141) (0.078) (0.098)
Dependency ratio −2.079* −2.274 3.367
(1.200) (1.482) (3.720)
Experience in dairy farming in years −0.023 −0.043 −0.036
(0.032) (0.033) (0.036)
Share of dairy income in total income 0.022 0.064*** 0.074
(0.017) (0.018) (0.046)
Log of per capita income 0.299 0.820** 2.459***
(0.692) (0.390) (0.905)
Log of animal values 0.075 0.813* 0.739
(0.699) (0.461) (0.778)
Herd size 2.255*** 3.173*** 3.192***
(0.820) (0.336) (0.871)
(continued on next page)

33
A. Kumar et al. Food Policy 75 (2018) 24–36

Table A.4 (continued)

Dependent variable—Milk yield Farm size

Landless Marginal Small and large

Year dummya 1.052 2.789*** −0.705

(1.190) (0.934) (1.585)
Training in livestocka 6.009*** −6.046** −2.197
(1.758) (2.737) (2.359)
Share of crossbred animals 0.003 −0.006 0.026*
(0.013) (0.011) (0.016)
Block fixed effect Yes Yes Yes
Constant −4.127 −16.631*** –
(8.069) (6.398) –
Observations 90 128 78

Note: Robust standard errors in parentheses.

a
Denotes dummy variable.
*** p < .01.
** p < .05.
* p < .1.

Table A.5
Impact of dairy cooperatives on net return per liter by farm size.

Dependent variable—Net return per Farm size

liter
Landless Marginal Small and large

Membera 0.181 0.363*** 0.044

(0.180) (0.133) (0.227)
Age in years 0.001 0.002 −0.006
(0.011) (0.004) (0.008)
Male-headed householda 0.302 −0.026 –
(0.236) (0.137) –
Education in years 0.006 −0.008 0.039*
(0.022) (0.012) (0.021)
Household size −0.031 0.026** −0.011
(0.027) (0.012) (0.017)
Dependency ratio −0.302 −0.142 0.551
(0.365) (0.192) (0.576)
Experience in dairy farming in years 0.005 −0.002 −0.001
(0.006) (0.004) (0.010)
Share of dairy income in total income −0.002 0.005** 0.010
(0.004) (0.002) (0.007)
Log of per capita income 0.056 0.064 −0.060
(0.107) (0.068) (0.140)
Log of animal values 0.113 −0.182** 0.019
(0.105) (0.079) (0.156)
Milk production in liters 0.032** 0.046*** 0.015
(0.015) (0.008) (0.018)
Year dummya 0.813*** 0.425** 0.931***
(0.267) (0.179) (0.279)
Training in livestocka −0.570 −0.150 0.096
(0.503) (0.206) (0.213)
Share of crossbred animals −0.001 0.000 −0.001
(0.002) (0.002) (0.003)
Block fixed effect Yes Yes Yes
Constant −0.265 2.256** 1.416
(1.345) (1.021) (1.914)
Observations 83 115 72

Note: Robust standard errors in parentheses.

a
Denotes dummy variable.
*** p < .01.
** p < .05.
* p < .1.

34
A. Kumar et al. Food Policy 75 (2018) 24–36

Table A.6
Impact of dairy cooperatives on adoption of FSM by farm size.

Dependent variable—Adoption of Farm size

FSM
Landless Marginal Small and large

a *
Member 6.304 5.860 2.811
(5.525) (3.108) (3.838)
Age in years 0.017 0.087 0.070
(0.095) (0.101) (0.129)
Male-headed householda −3.571 −5.681 92.947***
(3.911) (7.133) (34.319)
Education in years 0.128 0.566*** 0.554
(0.468) (0.199) (0.412)
Household size −0.152 0.231 −0.026
(0.497) (0.362) (0.394)
Dependency ratio 7.113 −0.576 −4.069
(5.266) (4.125) (8.986)
Experience in dairy farming in years −0.186** −0.164 −0.075
(0.085) (0.108) (0.114)
Share of dairy income in total income −0.098* −0.155*** −0.201
(0.056) (0.056) (0.146)
Log of per capita income −1.965 0.637 −7.415***
(2.502) (1.230) (2.663)
Log of animal values 3.069** 1.763 1.969
(1.298) (1.626) (2.843)
Milk production in liters 1.104*** 0.577** 0.561*
(0.350) (0.264) (0.298)
Year dummya 2.430 −0.017 7.361
(4.633) (3.206) (4.691)
Training in livestocka 28.344*** 17.497 −1.264
(6.925) (14.509) (5.429)
Share of crossbred animals −0.031 0.003 −0.022
(0.034) (0.029) (0.041)
Block fixed effect Yes Yes Yes
Constant 34.682* 27.769
(19.639) (18.614)
Observations 90 128 78

Note: Robust standard errors in parentheses.

a
Denotes dummy variable.
*** p < .01
** p < .05
* p < .1.

Appendix B. Supplementary material

Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/j.foodpol.2018.01.005.

References
Berdegue, J.A., 2001. Co-operating to compete: associative peasant business firms in
Chile. PhD thesis, Wageningen University and Research Centre, Department of Social
Science, Communication and Innovation Studies Group, Wageningen, The
Netherlands.
Birthal, P.S., Joshi, P.K., Gulati, A., 2007. Vertical coordination in high-value food
commodities: implications for smallholders. In: Joshi, P.K., Gulati, A., Cummings, R.
(Eds.), Agricultural Diversification and Smallholders in South Asia. Academic
Foundation, New Delhi, pp. 405–440.
Birthal, P.S., Jha, A.K., Tiongeo, M.M., Narrod, C., 2009. Farm-level impacts of vertical
coordination of the food supply chain: evidence from contract farming of milk in
India. Indian J. Agric. Econ. 64 (3), 481–495.
Birthal, P.S., Chand, R., Joshi, P.K., Saxena, R., Rajkhowa, P., Khan, M.T., Khan, A.,
Chaudhary, K.R., 2017. Formal versus informal: efficiency, inclusiveness and finan-
cing of dairy value chains in Indian Punjab. J. Rural Stud. 54 (August 2017),
288–303.
Candler, W., Kumar, N., 1998. India: The Dairy Revolution: Impact of Dairy Development
in India and the World Bank’s Contribution. World Bank Operation Evaluation
Department, World Bank, Washington, DC.
Carter, D.W., Milon, J.W., 2005. Price knowledge in household demand for utility ser-
vices. Land Econ. 81 (2), 265–283.
COMFED (Bihar State Milk Cooperative Federation), 2017a. Unit wise milk procurement.
< http://www.sudha.coop/topics.aspx?mid=Physical%20Performance > (accessed
7 July 2017).
COMFED, 2017b. District wise milk production in Bihar. < http://www.sudha.coop/
DataFiles/CMS/file/Districtwise%20Milk%20Production%20-%20Bihar.
pdf > (accessed 7 July 2017).
Government of India, 2011. Primary census abstract, 2011. Office of the Registrar General
and Census Commissioner, India.
Government of India, 2016. Basic Animal Husbandry and Fisheries Statistics. Ministry of
Agriculture and Farmers Welfare, Department of Animal Husbandry, Dairying and
Fisheries.
Gupta, K., Roy, D., 2012. Gains from coordination in milk-fed dairy in Punjab. J. Agribus
Dev. Emerg. Econ. 2 (2), 92–114.
Heckman, J.J., Tobias, J.L., Vytlacil, E.J., 2001. Four parameters of interest in the eva-
luation of social programs. South. Econ. J. 68 (2), 210–233.
Hemme, T., et al., 2003. IFCN Dairy Report 2003. IFCN Centre, Braunschweig, Germany.
Hoda, A., Rajkhowa, P., Gulati, A., 2017. Unleashing Bihar’s agriculture potential:
Sources and drivers of agriculture growth. ICRIER Working Paper No-336, ICRIER,
New Delhi.
Jardine, S.L., Lin, C.Y.C., Sanchirico, J.N., 2014. Measuring benefits from a marketing
cooperative in the Copper River fishery. Am. J. Agric. Econ. 96 (4), 1084–1101.
Kumar, A., 2010. Milk marketing chains in Bihar: implications for dairy farmers and
traders. Agric. Econ Res. Rev., Conference Issue 469–477.
Kumar, A., Staal, S.J., Singh, D.K., 2011a. Smallholder dairy farmers’ access to modern
milk marketing supply chain in India. Agric. Econ. Res. Rev. 24 (2), 243–253.
Kumar, A., Wright, I.A., Singh, D.K., 2011b. Adoption of food safety practices in milk
production: implications for dairy farmers in India. J. Int. Food Agribus. Mark. 23 (4),
330–344.
Kumar, A., Shinoj, P., Jee, S., 2013. Do dairy co-operatives enhance milk production,
productivity and quality? Evidences from the Indo-Gangetic plain of India. Indian J.
Agric. Econ. 68 (3), 457–468.
Kumar, A., Thapa, G., Joshi, P.K., Roy, D., 2017a. Adoption of food safety measures on
milk production in Nepal: impact on smallholders’ farm-gate prices and profitability.
Food Policy 70, 13–26.
Kumar, A., Mishra, A.K., Shinoj, P., Jha, G.K., 2017b. Structure of Milk-marketing

35
A. Kumar et al.
Channels in India: The Role of Informal Outlets. International Food Policy Research
Institute, South Asia Office, New Delhi.
Lele, U., 1981. Co-operatives and the poor: a comparative perspective. World Dev. 9 (1),
55–72.
Lokshin, M., Sajaia, Z., 2004. Maximum likelihood estimation of endogenous switching
regression models. Stata J. 4 (3), 282–289.
Maddala, G.S., Nelson, F.D., 1975. Switching regression models with exogenous and
endogenous switching. Proc. Am. Stat. Assoc. Bus. Econ. Sect. 423–426.
Mergos, G., Slade, R., 1987. Dairy development and milk cooperatives: The effect of a
dairy project in India. World Bank Discussion Paper No. 15, World Bank,
Washington, DC.
Narrod, C., Roy, D., Okello, J., Avendano, B., Rich, K., Thorat, A., 2009. Public-private
partnerships and collective action in high-value fruit and vegetable supply chains.
Food Policy 34 (1), 8–15.
NDDB (National Dairy Development Board), 2016. Annual Report 2015–2016. NDDB,
Government of India.
Roy, D., Thorat, A., 2008. Success in high-value horticultural export markets for the small
farmers: the case of mahagrapes in India. World Dev. 36 (1), 1874–1890.
36
Food Policy 75 (2018) 24–36
Sexton, R.J., Iskow, J., 1988. Factors Critical to the Success or Failure of Emerging
Agricultural Cooperatives. University of California, Davis, Division of Agriculture and
Natural Resources.
Singh, K., Das, V.M., 1984. Impact of Operation Flood I at the village level. Research
Report No. 1, Institute of Rural Management, Anand, India.
Singh, R.K.P., 1996. Dairy co-operatives: organizations of the poor in Bihar. In:
Rajagopalan, R. (Ed.), Rediscovering Cooperation: Co-operatives in the Emerging
Context. Institute of Rural Management, Anand, India.
Singh, K., Pundir, R.S., 2000. Co-operatives and rural development in India. Research
Paper 17, Institute of Rural Management, Anand, India.
Shukla, R.K., Brahmankar, S.D., 1999. Impact Evaluation of Operation Flood on Rural
Dairy Sector. National Council of Applied Economic Research, New Delhi.
Stockbridge, M., Dorward, A., Kydd, J., 2003. Farmer Organizations for Market Access: A
Briefing Paper. University of London, Wye College.
Vandeplas, A., Minten, B., Swinnen, J., 2013. Multinationals vs. cooperatives: the income
and efficiency effects of supply chain governance in India. J. Agric. Econ. 64,
217–244.