This is a preliminary draft. Please do not cite.

1

Friends or Traders? Do social networks affect the use of market mechanisms by

farmers in India

Kathy Baylis, Ashwini Chhatre, Satya Prasanna and Tisorn Songsermsawas†

Selected Paper prepared for presentation at the Agricultural & Applied Economics

Association’s 2012 AAEA Annual Meeting, Seattle, Washington, August 12-14, 2012

This is a preliminary draft. Please do not cite.

Copyright 2012 by [Kathy Baylis, Ashwini Chhatre, Satya Prasanna and Tisorn Songsermsawas].

All rights reserved. Readers may make verbatim copies of this document for non-commercial purposes by any means, provided

that this copyright notice appears on all such copies.

† Tisorn Songsermsawas ([email protected]) is the corresponding author. Kathy Baylis and Ashwini Chhatre

are Assistant Professors in Department of Agricultural and Consumer Economics and Department of Geography,

and Tisorn Songsermsawas is a PhD student in Department of Agricultural and Consumer Economics at the

University of Illinois, Urbana-Champaign. Satya Prasanna is an independent scholar in India.

2

Friends or Traders? Do social networks affect the use of market mechanisms by

farmers in India

Kathy Baylis, Ashwini Chhatre, Satya Prasanna and Tisorn Songsermsawas†

Abstract

In this paper, we ask when small-scale farmers invest in a long-term contract with a

trader to assist them in selling their cash crops. By investing in a long-term relationship, the

farmer may improve their market information, but they lose the option of selling to other traders

in any specific year. The decision to invest in such a long-term contract may also be affected by

the farmer’s social network: either because social networks can facilitate the communication or

market information to larger number of small farmers, in which case they act as complements

with the long-term contract with a trader, or they can provide an alternate source of market

information and risk pooling, in which case they act as substitutes. We use data from a field

survey which contain detailed information about social networks of 522 households across 17

villages of the Thaltukhod Valley in Himachal Pradesh, India. We first outline how social

networks and long-term contract with a trader might help small-scale farmers in a developing

country market their cash crops. We then estimate the probability of a farmer making such an

investment and find that social networks affect this choice. We find significant peer effects, and

we see evidence that the more central position the household has in the network and the smaller

the network, the higher the probability of such an investment.

This is a preliminary draft. Please do not cite.

† Tisorn Songsermsawas ([email protected]) is the corresponding author. Kathy Baylis and Ashwini Chhatre

are Assistant Professors in Department of Agricultural and Consumer Economics and Department of Geography,

and Tisorn Songsermsawas is a PhD student in Department of Agricultural and Consumer Economics at the

University of Illinois, Urbana-Champaign. Satya Prasanna is an independent scholar in India.

3

Introduction

The transaction costs of marketing cash crops for small agricultural farmers in developing

countries can be substantial (Kirsten and Sartorious 2002, Rujis et al. 2004). Market

intermediaries such as traders can help reduce these transaction costs by disseminating market

information, mitigating risks and supporting commitment (Spulber 1999). Social networks can

also act to reduce transaction costs through facilitating information exchange, risk sharing and

enabling economies of scale (Fafchamps and Minten 1999, Lyon 2000, Fukunaga and Huffman

2009). The purpose of this paper is to investigate the relationship between a household’s social

network and the household’s investment in a long-term contract with a trader to market their

agricultural production in rural northern India. We investigate how the type, diversity and

position in the social network affects the household’s decision to make a specific investment in a

long-term contract with an agricultural trader for their cash crops and whether social networks

act as complements or substitutes to the use of a trader. Further, we consider how both the

traders and social network affect household outcomes.

A primary source of transaction costs is information costs and uncertainty (Coase 1937,

Stigler 1961, 1967). For agricultural farmers in rural India, the sources of transaction costs may

include their lack of access to market information, insufficient knowledge of production

technologies, and the potential decay of unsold cash crops. Traders, on the other hand, can help

the farmers overcome these barriers by facilitating the commercialization of the cash crops and

sharing information about the market (Fafchamps and Minten 1999).

Information can also be shared through social networks. A series of studies on the

adoption of new technologies in agricultural production conducted by Udry and Conley (2001,

2005, 2010) show that social networks can assist the flow of information and subsequently can

4

facilitate adoption of new technologies throughout the network. Although a number of studies

have analyzed the importance of both social networks and traders to the improvement of

economic outcomes in several developing countries, very little quantitative research has been

conducted on the relationship between social networks and trader with regards to the household’s

agricultural commercialization decision. This study seeks to highlight some of the key

components that drive the household’s decision whether to take advantage of their social

networks or to commit to a long-run use a trader to help commercialize their cash crops.

We use social network analysis to derive summary characteristics of social networks,

which encompass several attributes including the size, density and structure of the social network

and household location within the network. Then, using econometric methods informed by

spatial econometrics, we ask what characteristics of the network affect the decision of the

household to establish a long-term commitment to a trader for a specific crop. Thus, we explore

whether the nature of the social network affects the choice investment in a long-term relationship

with a trader and the household outcome. Preliminary results from this analysis indicate that

households with smaller and denser social networks are more likely to have a long-term

relationship with at least one trader. We also find substantial peer effects. The centrality

measure of a household in the network is also a significant indicator of a long-run commitment

to a trader, where the centrality measure indicates how central a household is located within its

network. A central location in a social network leads to an influential role of that node for the

other nodes within the same network. Therefore, if a household that is centrally located commits

to a long-run relationship with a trader, it is more likely that this particular household will be

influential on such commitment of other households within the same network. Moreover,

individual characteristics of households also play an important role in determining the decision

5

to invest in a long-term relationship with a trader. Based on the survey data, there is evidence

that greater wealth, higher education, larger family size and higher altitude of a household might

lead to the decision for households to make a specific investment of establishing a long-run

relationship with a trader to support the sale of their cash crops.

We see three main contributions of this study. To our knowledge, this is the first study to

analyze how detailed information on how each household’s position within a social network

affects their use of a trader. Our measures are not restricted to whether a relationship exists

between two households, but also include the nature, closeness, density and degree of contact

among the households in each village. Although it is natural to assume that a household with

more connected network is more likely to be able to obtain information about a trader and

eventually decides to adopt a trader to help commercialize their crops, we are interested to

investigate this pattern of adoption in greater detail in thus study.

The second contribution is due to the fact that our dataset include all households across

the 17 villages in the Thaltukhod Valley in Himachal Pradesh, India. Since our dataset contains

the entire population in the area of study, we can investigate whether social networks and long-

term contract with a trader act as complements or as substitutes without the limitation of

estimation bias from sampled data.

The third contribution of this study is that this is one of the first studies to apply novel

spatial econometric methods to the study of social networks. We follow the new work by Udry

and Conley (2010) and Banerjee et al. (2011), and apply spatial econometric analysis to

explicitly estimate spatial lags across the social network into account.

2. Literature Review and Conceptual Framework

The theoretical model that we use to develop the conceptual framework for this study

relies heavily on the specific investment model of a firm (Tirole’s 1988). Suppose an agricultural

6

household producer (seller) in northern India has to make a decision on whether to make a

specific investment in a long-term contract (or relationship) with a trader (buyer) to help

commercialize for his cash crops. This specific investment depends on each particular type of

cash crop the producer grows. A long-run relationship with a trader is a specific asset of the

household in that it requires investment of time and trust with a specific trader and does not

directly benefit the household in dealings with other traders (Tirole 1988). In return, a long-term

commitment with a trader ensures the farmer that the trader would regularly supply market

information to him.

The specific long-run relationship between a farmer and a trader comes with both

benefits and drawbacks for both parties. For the farmer, a long-term commitment with a trader

assures consistent demand, eliminates future search cost and guarantees frequent delivery of

market information by the trader. However, such specific relationship prevents the farmer from

taking advantage of occasional price spikes in the market if he were to sell the cash crops on his

own. This situation is referred to as a ‘locked-in’ bilateral trading relationship (Klein et al.,

1978). On the other hand, the specific relationship between a farmer and a trader provides the

trader provides the trader with consistent supply and low price variation. The only cost that the

trader has to incur is only due to the specific investment. This specific investment creates

‘hostage’ by providing incentives for the farmer to stay in this specific relationship (Klein et al.,

1978). Anecdotally, in the Thaltukhod Valley such incentives may include offering rides for the

farmer’s family members to hospitals or assistance with marriage arrangements. These incentives

are extraneous to agricultural production, but they are given by the trader to the farmer in order

to hold the farmer in the specific relationship.

7

The specific relationship generates mutual benefits for the farmer and the trader as long

as they remain active in it. However, if the specific contract is terminated, each party faces

different levels of consequences. The discontinuation of the specific relationship might result in

the farmer’s reduction in demand for cash crop. This might further lead to the risk in crop

spoilage, which is different depending on the type of cash crop grown. On the other hand, the

dissolution of the specific contract would incur extra search cost for the trader to look for a new

trading partner to engage in a long-term contract with. This extra search cost might be very

minimal since there are multiple farmers in Thaltukhod Valley that grow the same type of cash

crop. Therefore, if the long-term specific contract is to be suspended, the party that would always

receive a worse-off outcome is the farmer.

Suppose our theoretical model is such that there are a trader who wants to buy the cash

crop and the producer who grows that cash crop faces a (sunk) transaction cost involved in

getting the good to the trader. Each party has hidden information; pre-harvest, the trader does not

know the true quality or quantity of the cash crop that the producer wants to sell, while the

producer does not know about the market outcome. When there is uncertainty about the quality

of the crop, supply of cash crops or market outcome, the market will not be able to function

efficiently since the trader and the producer have asymmetric information (Akerlof 1970).

Assume that each party is trying to maximize his/her payoffs under uncertainty given the other

party’s hidden information. Processes that can reduce this uncertainty can improve the possibility

of trade, and may increase the price received by the producer (Kherallah and Kirsten 2002). In

other words, a long-term contract between the producer and the trader can provide market

opportunities to the producer and guarantee a regular supply of cash crops to the trader, which he

can use to improve his sale price (Kherallah and Kirsten 2002, Swain 2008). We use this

8

conceptual framework to further develop a model to study the role of traders on the

commercialization of cash crop production in northern India and its implications to the social

networks of an agricultural economy.

The theoretical model that we use as our framework to analyze the long-term specific

relationship between an agricultural farmer in rural northern India is based largely on the model

proposed by Noldeke and Schmidt (1995). First, consider an option contract in which a farmer

and a trader, both risk averse, would together establish a long-term, specific relationship to trade

a cash crop. Before trade, both the trader and the farmer make sunk, specific investments

and . The trader’s valuation of the cash crop is given by and

represents the seller’s cost of production, where represents the state of

nature and is distributed on the region according to the joint density function .

Let and represent the strictly increasing and continuous cost functions for

the specific investments made by the farmer and the trader. Also, let and be

strictly positive and continuous in both arguments. Let denote the level of trade, and

denote the net payment from the trader to the farmer. Then, after trade, the utility levels of the

trader and the buyer can be given as follows.

Following Hart and Moore (1988), this specification of option contract gives the farmer

the right to supply to cash crop to the trader and receive price , or not to supply the cash crop

to the trader and receive market price Intuitively, the farmer will have the incentive to invest

the long-term relationship with the trader if he expects the difference between and to be

smaller than his production cost, and will choose not to invest in such specific investment with

9

the trader and will instead trade at market price if he expects otherwise. The price the

farmer receives from both scenarios can be derived as follows (Noldeke and Schmidt, 1995).

If then

If then .

The expressions for the expected utility of each party can be derived given the farmer’s

investment choice made from the option contract offered to him. Given an option contract

where is the base payment (price received by the farmer if he decides not to make a specific

investment) and is the option price. Moreover, the farmer’s decision to make a

specific investment in a long-term relationship with a trader also depends on the source of

information he chooses to receive market information from. In this scenario, a farmer can either

choose to receive market information from a trader or rely on their social networks for

information Therefore, the expected utility for the trader and the farmer can be given by

Given that trade is the efficient outcome, and the trader receives full marginal return from

his investment, the option contract is one in which the farmer chooses the optimal

investment level to solve the following problem

Although deciding to invest in a specific asset as a long-term commitment to a trader can

generate a substantial level of benefits for the producer, this investment in a specific asset also

10

comes with costs (Williamson 1989, Tirole 1988). A specific investment creates a ‘lock-in’

contract between a seller and a buyer that makes the use of such specific asset invested not as

effective outside the relationship (Williamson 1979). By choosing to invest in this long-term

contract with a trader as a specific asset, the agricultural producer’s outside opportunities are

eliminated due to this specific commitment to a trader. Therefore, a long-term commitment to a

trader provides the agricultural producer with a number of benefits including market information

and consistent flow of demand. However, the specificity of this relationship also burdens the

farmer with costs in a way that it reduces the producer’s market opportunities outside the

relationship.

Most literature on the role of agricultural traders has been concentrated on Africa. Several

studies find that the relationship between traders and small-scale farmers helps to improve

economic outcomes and productivity (Fafchamps 1996, Barrett 1997, Fafchamps and Minten

1999, Fafchamps and Minten 2001, Moser et al. 2005). These outcomes result from a reduction

in the time needed to transport goods and explore market opportunities, better information about

the market, more stable demand and supply, and a reduction in losses from spoilage. The

assistance of traders greatly enhances income opportunities for these small-scale farmers,

reducing uncertainties in the quality and the buyer’s willingness to pay, and in return, given a

fixed amount of transaction cost, the traders with well-connected networks enjoy higher

compensation due to higher sales volume.

In many developing countries, personal relationships play a significant role in daily

economic activities. These social networks are viewed as a form of capital that can foster

cooperation and coordination and generate economic returns (Coleman 1990, Putnam 1993,

Fafchamps 1998, Woolcock 1998, Narayan and Pritchett 1999, Lyon 2000, Fafchamps and

11

Minten 2001, Udry and Conley 2001, Winter-Nelson and Temu 2005, Jackson 2008, Udry and

Conley 2010). As a result, social networks may substitute a market intermediary and reduce the

household’s transaction costs of marketing their production. Conversely, social networks may

enhance the benefit of using a trader, since market knowledge can now spread further, and the

pooling potential of the social network may generate economies of scale for the trader

themselves. Thus, we wish to determine whether social networks and long-term contracts with a

trader are complements or substitutes in the rural Indian farmers’ commercialization process of

their cash crops.

How personal relationships affect economic productivity has been analyzed in recent

years. As discussed extensively in Fafchamps and Minten (1999), familiarity and trust can help

facilitate economic exchange in several regards. Specific functionalities of the social network

can foster better economic outcomes of the small-scale farmers. For example, a social network

can facilitate the transmission of information throughout the network, especially the information

about technology and market opportunities (Kranton 1996, Barr 1997). The broader is the

network, the greater are the sources of information. The adoption of a new production

technology or market mechanism by an individual with a large network is more likely to result in

a significant dispersion of similar adoptions throughout his social networks (Bandiera and Rasul

2006). Due to this crucial role of social network, we are interested in examining the

characteristics of social network with regards to type, diversity and location that could lead to

long-term trader commitment among small-scale farmers in the Thaltukhod Valley.

Despite the various functionalities, social network cannot fully replace a long-term

contract with a trader in some of the roles that traders can perform. Several studies have carefully

analyzed the role of trust between trader and agricultural producers and argue that it greatly

12

fosters cooperation between the parties involved (Coleman 1990, Putnam 1993, Lyon 2000,

Fafchamps and Minten 2002). The adoption of a trader as a middle-person in the transaction of

agricultural products can signal quality of agricultural products to the final buyer because the

trader wants to uphold their reputation with the consumers (Kherallah and Kirsten 2002, Batt

2003, Best et al. 2005). Thus, through reputation, a trader can help reduce uncertainties about

product quality and delivery facing the buyer. The fact that social network, on the other hand, is

a non-market mechanism that relies heavily on personal interactions among small-scale farmers

and likely cannot create the trust and reputation as a trader for the consumer.

Transaction costs play a key role in deciding when to enter a contract (Allen and Lueck

1992, 1999, Canjels 1998). Although social networks can help farmers overcome transaction

costs associated with marketing, there are notable aspects of transaction cost that social networks

alone cannot help reduce. As previously noted, the main sources of transaction costs are

information and uncertainty (Coase 1937). With regards to market information, a trader can only

supply the local farmers with the latest information about the market including policy changes,

technological advances or consumer demand. However, a trader who is in contact with a farmer

who has a large social network can also communicate market information to a number of farmers

through that network. If this is the case, then we have evidence that a long-term contract with

trader and social networks act as complements. On the other land, if a household is in contact

with other households that can obtain information about the market, that household might not see

the need for a long-term contract with a trader. In this scenario, social networks and the long-

term contract with the trader are substitutes.

Transaction cost can also occur from the uncertainty specific to each cash crop and the

supply level of each crop (Goetz 1992, Jayne 1994, Omamo 1998, Key et al. 2000, Winter-

13

Nelson and Temu 2005). There are three main cash crops grown by the local farmers in

Thaltukhod Valley: kidney beans, potatoes and peas. The uncertainty associated with potatoes is

mainly due to disease shocks and storage life. Potato blight largely affects all producers in a

region at the same time, resulting in substantial shocks to supply, causing the price of potatoes

sold to fluctuate greatly over time. The long storage life of the potatoes allows them to be stored

up to two years, allowing retailers to mitigate against these potential supply shocks. A farmer

will neither necessarily know about blight in a neighboring production area, nor how many

potatoes are currently in storage in the primary retail markets. Thus, potato producers do not

observe these key components of expected market price unless informed by a trader.

For peas, the pattern of market price is much more consistent over time than that of

potatoes, with the price being higher in the spring and lower in the summer. The high

perishability of peas implies that they cannot be stored from one season to the next, and disease

shocks are less systematic than in the case of potatoes. However, the high perishability of peas

necessitates the rush to deliver the crop to the market soon after harvest.

Transportation cost plays an important role determining the level of transaction cost

associated with each crop (Eswaran and Kotwal 1986, Key et al. 2000). As potatoes are larger

and heavier than kidney beans and peas, potatoes growers bear higher transaction cost when they

are trying to sell them. Due to the fact that farmers growing each crop face different sources

uncertainty, it is reasonable to investigate the decision of a farmer to make a specific investment

by establishing a long-term relationship with a trader in light of lowering transaction costs and

sharing of risk (Fukunaga and Huffman 2009), and analyze its implications of such specific

investment to his social networks.

14

Few papers have studied the relationship between traders and social networks. Most of

the earliest studies in this area belong to a series of papers by Marcel Fafchamps and Bart Minten

from surveys in African countries (Fafchamps 1998, Fafchamps and Minten 1999, Fafchamps

and Minten 2001). These papers demonstrate that more successful traders are those with larger

social networks. Of the various functionalities can accommodate, these studies have found the

consistency of supply and demand, and the sharing of risk are observed to be most crucial. Social

networks also play an important role in shaping and fostering economic development in rural

Ghana, where the spread of information, capital and influence that determine economic decisions

rely heavily on the social networks of local farmers (Conley and Udry 2001, Conley and Udry

2010). Our paper complements the existing literature by considering the combined effect of

traders and social networks on farmers at the household level. Specifically, we explore how the

context and structure of social networks influence a household’s decision to commit to a long-

term contract with a trader to commercialize their agricultural products.

3. Area of Study and Data

The study area is Thaltukhod Valley, an area of 17 villages and 522 households located in

the Indian Himalayas (as shown in Figure 1). There is a considerable level of variation in the

livelihood strategies of households in the valley. Most households make their living from a

combination of subsistence agriculture, commercial crop cultivation, livestock rearing, and civil

service jobs. The forests that adjoin each village also make significant contributions to livelihood

strategies, as households depend (to varying degrees) on forest products like fuel wood, grazing

area, fodder, timber, fencing, biomass, and medicinal plants. Due to the mountainous landscape

of Thaltukhod Valley, the agricultural areas do not generally span large, continuous areas. Each

village has between two and seven agricultural land units that vary in size and altitude. Within

15

each land unit, there are clear, legally-recognized delineations of what land belongs to each

household in the village, with landholding sizes varying greatly among households. While the

land is privately-owned by individual households, the close physical proximity of properties in

the same land unit incentivizes households to make cooperative crop management decisions.

In 2008, a comprehensive survey was administered to households in these villages.

Households were asked detailed questions about their livelihood activities for the previous four

years (2004-2007), and ten years ago (1998). The survey also collected detailed social networks

of the all households and whether the household has a long-term relationship with a trader and

for which crop. The survey also contains detailed crop information for each household.

According to the data from the survey, all farmers in this region grow one or more cash crops of

potatoes, kidney beans and peas.

Within each village, we observe both households who have a long-term relationship with

a trader and households who do not. Moreover, certain villages are in contact with up to three of

the six traders operating in the region. All of the traders are of higher caste. Thus, we see

evidence that the decision to use of a trader is not purely determined by geography. Traders also

operate in all three crops, implying the choice to use a trader is not driven solely by crop choice.

In this study, we analyze four social networks characteristics of the households all the

villages in Thaltukhod Valley. The variables of interest are degree, k-step reach, average

reciprocal distance (ARD) and eigenvector. The degree, k-step reach and average reciprocal

distance variable can help explain how much information can flow within a network due to its

size, spread, and closeness but they do not fully capture a household’s influence on other

households within the same network. The eigenvector variable specifically captures influence of

a node with respect to all other nodes within the same network since it is a measure of network

16

centrality. To provide a better understanding of the network measurements discussed earlier,

consider the two maps of social networks in village 6, Tegar and village 14, Bhumchayan,

presented in Figures 2 and 3. As a comparison, compare household number 5 in village 6

(labeled as HH5 in Figure 2) and household number 16 in village 14 (labeled as HH16 in Figure

3). Both households are circled in red in the village network maps. Although both households

appear to be centrally located within each network, they have very different values of

eigenvector and two-step reach variables. For the eigenvector variable, the eigenvector value of

household 5 in village 6 is 0.411 whereas that of household 16 in village 14 is 0.226. The

explanation of such difference in eigenvector values is that as the network of village 16 is much

larger than the network of village 6 (because there are more households in village 14 than in

village 6), the degree of influence a central household has on all the other households in a bigger

network is less than that of a central household that belongs to a smaller network. The two-step

reach of these two households is also different. The two-step reach of household 5 in village 6 is

0.371. This figure indicates that within two steps, this household can reach 37.1% of all the

households in this network. On the other hand, in a much denser network as in village 14, the

two-step reach variable of household 16 is 0.969. This means that almost all of the households

within this network can be reached from this household within two steps. To summarize the

difference between the two network variables, we can think of the two-step reach variable as a

measure of pure information flow within a social network. However, the eigenvector variable

mainly captures the influential effects of a node on the other nodes within the same network.

The summary statistics of the social network variables (presented in Table 1) clearly

indicate that households with higher network eigenvectors are more likely to establish a long-

term contract a trader for a long-term contract to help them sell their cash crops. The degree

17

variable, which measures the average number of links, or network contacts, a household has, is

slightly higher among households dealing with a trader. The “k-step reach” variable, which in

this study uses k=2, measures the number of nodes within the network reachable within 2 steps,

has a mean of 0.55 among the households that don’t use a trader and 0.60 among those that do.

This statistic means that 55 or 60 percent of the network are friends of friends.

The average reciprocal distance variable is a measure of closeness of centrality. It

indicates the average shortest possible path length between a node in network and any other

nodes is in the network. We observe almost no difference in this category between those that

don’t use a trader and those who use a trader (0.49 to 0.51). The last network variable of interest

is the eigenvector variable. The eigenvector defines centrality by indicating how connected one

household is to all the other households within the same network. Put differently, the eigenvector

is an indicator of how important a node is in the entire network. Due to this feature, this

measurement can help describe the degree of influence a node has on its neighboring nodes.

Households that are in contact with at least one trader have an average eigenvector of 0.22,

which is only slightly higher than those who do not (0.19). Therefore, given the statistics of

these network variables, there is some evidence that the structure of the social networks has an

impact on the household’s decision to adopt a trader to help commercialize their agricultural

produce.

The summary statistics of the individual characteristics of the households who invest in a

long-run contract a trader and those who do not are presented in Table 1. First, the average

household income of the households that do have a long-run contract with a trader is 22,162

rupees. An average household that works with a trader owns a total land of 8.42 bhigas and a

total livestock of 1.94 units whereas households that do not have a long-run contract with a

18

trader on average own a total land of 8.18 bhigas and a total livestock of 1.82 units. The average

household sizes (head count) between the ones having a long-run contract with a trader (5.75)

and those do not (5.65) are not significantly different. The caste dummy variable also sees a

significantly higher average among the households which are in a long-run contact with a trader

(0.89 as compared to 0.80).

Households that have a long-run commitment to a trader have on average 0.54 stall-fed

cattle while those do not have a long-run have on average a lower amount of stall-fed cattle

(0.46). They also consume on average more purchased energy (LPG and kerosense) than the

households that do not have a long-term contract with a trader (0.67% as compared to 0.26%).

Finally, household that has a long-term commitment to a trader consume own-produced food

slightly fewer months a year (2.95 as compared to 3.02) than households that do not choose to

invest in a long-run commitment to a trader.

Table 2 illustrates the summary statistics of the households classified by the type the cash

crop they grow. Most of the household and social networks characteristics are similar across

households that grow each type of crops, except that there are a few notable differences. Farmers

who grow peas on average own more land than those who also grow kidney beans and potatoes.

Moreover, households that grow potatoes and peas on average own higher number of stall-fed

cattle, purchase more energy (LPG and kerosene), and consume more food from their own

production. Finally, the proportion between farmers of higher caste to those of lower caste is the

highest among those who grow peas. This is not a surprising result since peas were the latest

crop to be introduced to the farmers in Thaltukhod Valley. Since all of the traders working in the

area are of higher caste, farmers that belong to the higher caste have more access to obtain

market information about peas from the traders.

19

4. Estimation Method

In our study to analyze the role of social networks on the long-term adoption of trader in

rural northern India, we first construct a weights matrix to determine the links of the households

within each village. We will then have a total of 17 matrices and will subsequently use these

matrices to generate the network variables for each household.

To construct the weights contiguity matrix used in this study, we use two questions that

were asked regarding the social network within each of the 17 villages in Thaltukhod Valley.

First, each household is asked to name three households within the same village that they are in

contact with most frequently. Further, each household is also asked to name another two

households that they specifically talk about cash crops most frequently. Thus, each household

can list up to the maximum of 5 different households within the same village that they are most

connected with. Due to the fact that in many cases a household nominated the same household

for both questions, we only count whether a household considers another household to be a

neighbor once. The matrix representing social network in each of the 17 villages is a square

binary matrix with elements 0 and 1, with a dimension equal to the number of households in that

village. Each element corresponding to a particular row and column that takes the value of 1

indicates that a household is a neighbor of another household (in terms of trader network),

otherwise 0. Then, after obtaining the weights matrices for all the villages, we obtain the relevant

network variables from these weights matrices.

The household characteristics that we use to create another set of independent variables

in our study can be divided into four categories, economic and wealth conditions, social status,

household size and elevation. The first category, economic and wealth conditions, is represented

by five variables namely ownership of land and livestock and stall-fed cattle, dependence on

20

self-grown food and reliance on purchased energy sources. The first three factors determine a

large proportion of income of the households in the entire Thaltukhod valley and can also

represent economic responsibilities of the household. Land represents the total area of land each

household owns in bighas (1 bigha is approximately 0.2 hectares). Due to the skewed distribution

in the land and livestock variables, these two indicators are transformed into the natural

logarithmic scale for more efficient estimation results. The other two variables, consumption of

self-grown food and dependence on purchased energy reflects the connectivity to the market of a

household. We assume that the more connected a household is to the market, i.e. buy more

energy and food from the market, the more likely that a household would choose to invest in a

long-run contract with a trader to help them commercialize their cash crops.

The second category of variables illustrates the social status of a household through two

variables: education and caste. In construction the education variable, we use the number of

household members who receive education for more than ten years. As discussed in Agrawal and

Gupta (2005), we have also included the variable on the total number of members in a household

in our analysis as well since family size is likely to be correlated to the education level, which is

the variable in the third category. The caste variable is a dummy variable that indicates whether

the household belongs to either a lower caste or a higher caste (0 = lower caste, 1 = higher caste).

The caste system is deeply rooted in the Indian society since the ancient times and still plays a

critical role in determining the important decisions in the way of lives of people in India. As a

result, we are interested whether the caste which a household belongs to has a significant impact

on its decision to engage a trader to help sell the cash crops.

The fourth category is the elevation level of a household. There two main reasons that

support the inclusion this variable into our model. The first is that elevation level can be used as

21

a proxy for transportation costs. If a household is located at a high elevation level, then that

household would have much greater difficulty transporting the cash crops to sell at the market

due to the higher transportation costs and the more time needed to reach the main town area

located at a lower elevation. Due to this reason, these households carry a greater responsibility of

taking care of these larger land areas for agricultural production purposes and as a result keep

them from taking the time to transporting the crops to sell by themselves.

Some of the major data problems that we are likely to encounter when performing any

econometric analyses are multicollinearity, and heteroskedasticity. Regarding heteroskedasticity,

we first run a general OLS regression on all of the regressions presented in Tables 2 and 3, and

compared them with a robust OLS regression to control for heteroskedastic errors. From these

two sets of OLS regressions, we observe that there is no significant difference between the

parameter estimates. To test for multicollinearity problem of the dataset, we ran the variance

inflation factor (VIF) test and the test result suggests that multicollinearity is not a concern in any

of our models.

The econometric model we use to test the likelihood that a household decides to engage a

trader to help them commercialize their products as a function of individual households’

characteristics and their network characteristics is a logistic regression model. Specifically, the

estimation model is of the form:

where represents the average household ’s decision whether to make a long-run investment in

a trader to help commercialize their cash crops or not, is a vector of household level

individual characteristics and is a vector of household’s network characteristics (included

both separately and all together). Additionally, we also estimate the likelihood of a long-term

22

specific contract between a farmer and a trader using the weights matrix, where represents the

k-nearest neighbor (using k=4) weights matrix based on geographical distance between the

households.

Additionally, we suspect that there might be unobserved effects of each crop that might

significantly result in the investment in a long-run contract to employ a trader. Therefore, we

introduce the village fixed effects to the logistic model:

where all other components of model are the same as in equation (1) and represents the crop

fixed effects for each crop (kidney beans, potatoes and peas).

Also, we are also interested in investigating if the unobserved effects within each village

would effect a household’s decision to commit to a trader in the long-run. Therefore, we

introduce a dummy variable for each village into the model. These unobserved effects can

include the existence of paved roads leading from a village to the main town area, among others.

where all other components of model are the same as in equation (1) and (2) and represents

the crop fixed effects for each village ( ).

Further, we are also interested to see if a household’s decision to invest in a long-term

contract depends on the decisions of their peers within the same social networks. To investigate

this hypothesis, we need to implement the spatial econometric procedures which have been

discussed extensively in the existing literature on the applications of spatial econometric

procedures (Paelinck and Klaassen 1979, Anselin 1988). Anselin (2002) highlights some of the

most widely used model specifications in spatial econometric regressions for empirical studies.

23

Two of the most common model specifications are the spatial lag and the spatial error models.

The spatial lag model can be formulated as follows.

where is the vector of observations on the response variable that represents the household ’s

decision whether to make a long-run investment in a trader to help commercialize their cash

crops, is the spatial weights matrix that illustrates the response variables of the neighbors, is

the vector of the independent variables of the households in the model, is the spatial

autoregressive parameter, is the regression coefficient vector and is the vector of error terms.

Moreover, the spatial error model can be formulated as follows.

where all other components of model are the same as in equation (5), and is the spatial

autoregressive parameter. It is necessary to note that if then our model is reduced to the

standard OLS model. However, if there is sufficient evidence for , the results from OLS

estimates are unbiased and consistent, but the error term is wrong and the parameter estimates

are inefficient.

5. Results

The regression results are shown in Tables 3 through 8. The first set of regressions is the

logit regressions of the likelihood of a farmer’s specific investment in a specific relationship with

the trader, which can be seen in Tables 3 and 6. Tables 4, 5, 7 and 8 present the spatial

autocorrelation regressions using the social network and geographical distance (using k=4

nearest neighbors) weights matrices. As with any other binary response regression, the most

useful approach to interpret the estimation results is to interpret the marginal effects of the

model. The marginal effects reported at the mean of the data calculated from the logit regression

24

described in equation (1) are presented in Table 3. In Table 4, we present results from the linear

spatially weighted lag model as defined in equation (5) using the social network weights matrix,

and in Table 5, the geographical distance weights matrix is used. Both Models (1) in Tables 3, 4

and Table 5 are the estimated without any fixed effects. Model (2) in Tables 3, 4 and 5 presents

the econometric estimation with the addition of village fixed effects. Model (3) presents the

results when the crop fixed effects are introduced, and finally in Model (4) we include both crop

and village fixed effects. After we introduce the fixed effects, we observe that the specification

without any fixed effects is not appropriate for our study. The introduction of village fixed

effects allows us to analyze the effects of all other variables in the model on the choice to invest

in a long-term commitment to the trader by removing all the unobserved variations across all the

villages (e.g. size of village). Similarly, the inclusion of crop fixed effects eliminates all the

unobserved variations across the three cash crops.

For the specification of the spatially weighted models, we run the LM specification test

for both spatial lag and spatial error models and the result of the LM test indicates that the spatial

lag model is more likely to be the appropriate specification. We also introduce village and crop

fixed effects to both the logit regression and the spatially weighted regression models. It is

important to note that we are aware of the limitations of the model specification of the linear

spatial lag model. Although our response variable is a binary variable, using the linear spatial lag

model will still yield unbiased coefficient estimates but the error term is inefficient.

The first main result is that social networks affect the decision to invest in a long-term

relationship with a trader. Specifically, different characteristics of a household’s social networks

affect the likelihood of investing in a long-term relationship with the trader in different ways.

The second important finding is that the one’s decision to invest in a long-term commitment with

25

a trader depends on the decisions made by their peers within the same village network. This

result is confirmed by the statistical significance of the spatial autoregressive parameter ( in

the spatial lag regressions. The spatial autoregressive parameter ( is most significant in the

specification that includes both the village and crop fixed effects, as shown in model (4) in Table

4. Thus, we clearly observe that the household’s choice to enter in a long-term contract with a

trader is highly dependent on the crop choice and the actions of their peers in the same network

within the same village. However, in Table 5, the spatial autoregressive parameter is not

statistically significant when the geographical distance weights matrix is used instead of the

social network weights matrix. Third, we observe the effects of the household individual

characteristics that affect the decision to invest in a long-term relationship with the trader namely

ownership of livestock and stall-fed cattle, caste, reliance on purchased energy and consumption

of self-grown food. And finally, the specific characteristics of the choice of crop grown and of

the village largely determine the decision to make a long-term investment in a trader.

Our study investigates the effect of four social network characteristics of households at

the village level across the Thaltukhod Valley. The variables of interest are degree, two-step

reach, ARD and eigenvectors. The degree variable, which measures the number of connections

within the same village network a household has, is statistically significant in some model

specifications, particularly when both the village and crop fixed effects are included. As the signs

of the average marginal effects in the logit regression and the coefficient estimates in the spatial

lag regression are both negative, this indicates that larger number of connections a household

has, the less likely a household would make an investment in a long-term contract with a trader.

This leads us to imply that households are less likely to invest in a long-term contract if their

peers have market information, so there is less need to use the service of the trader. The two

26

other statistically significant variables in our estimation models are the ARD and eigenvector

variables. Although these two variables are not consistently significant throughout all the

specifications, there is enough evidence to conclude that the higher level of closeness (measured

by ARD) and influence (measured by eigenvector) a household is to other households within the

same social networks at the village level, the more likely a household is to invest in a long-term

commitment to a trader for the commercialization of their agricultural produce. This results

presents some evidence that social network characteristics can either be complements (e.g. ARD,

eigenvector) or substitutes (e.g. degree) with the long-term contract with a trader.

Household characteristics also play a significant role in determining the household’s

choice to invest in a long-term contract with a trader. Specifically, these characteristics are the

agricultural responsibilities and the reliance on the market. The ownership of livestock and stall-

fed cattle variables reflect the level of a household’s agricultural responsibilities. Given that

these two variables are statistically significant, though not across all specifications, they indicate

that the household’s opportunity cost of working on their own farm is high. Rather than spending

time to market their cash crops and negotiate prices with different traders, households whose

opportunity cost of own agricultural labor are high would rather choose to invest in a long-term

relationship with a trader. The variables that reflect a household’s reliance on the market are the

proportion of energy purchased from the market (LPG and kerosene) and the consumption of

self-grown food. Although these two variables are not statistically significant for all the models

considered, they highlight that households that have higher level of dependence on the market

more are more likely to make a long-term investment in a contract with a trader. The statistical

significance of the caste variable notes the evidence that households that belong to the higher

27

caste might have better access to the traders since all of the six traders working in Thaltukhod

Valley also belong to the higher caste.

The inclusion of the crop fixed effects yield a very strongly significant result, and

confirms our hypothesis about each crop. The sign of the kidney bean dummy variable is

negative, indicating that farmers who grow kidney beans are less likely to choose to invest in a

long-term commitment with a trader because kidney beans have the least hidden information

among the three crops grown in the region. Potato is the crop with highest transaction cost

among the three, mainly due to its volatile production shocks, market supply and high

transportation costs. Therefore, we see that the average marginal effects in Table 3 and the

coefficient estimates in Tables 4 and 5 of the dummy variable for potatoes to be the most

positive. This indicates that a potato farmer is likely to commit to a long-term contact with a

trader in order to obtain market information about potatoes. Since peas, despite their high chance

of spoilage, have a very consistent price pattern throughout the year, peas growers see less need

of establishing a long-term contract with a trader.

As potatoes and peas are the two cash crops with considerable level of uncertainty

associated, we estimate the likelihood of investment in a long-term contract with a trader among

potato and pea growers. The average marginal effects from this set of logit regressions are

presented in Tables 6, and the coefficient estimates of the spatial lag model are presented in

Table 7 and 8 based on social network and distance weights matrices. Among the potato farmers,

a household’s influence within the network (measured by eigenvector) is the most significant

factor to commit to a long-run contract with a trader. The average marginal effect of the

eigenvector is statistical significant at the 5% level, but the coefficient estimate of this variable in

the spatial lag model is not significant at the 5% level. As for the household characteristics, the

28

statistical significant factors that influence the decision to invest in a long-term contract with a

trader are ownership of livestock, proportion of purchased energy and consumption of own food.

However, the spatial autoregressive parameter in the spatial lag model is not significant for

potatoes growers. This means that there is no strong evidence that peer effects matters in terms

of the investment in the long-term contract with a trader among potato farmers.

Pea growers show a much stronger peer effects of investing in a long-term relationship

with a trader, as shown in the statistical significance of the spatial autoregressive parameter in

Table 6. In this spatial lag model, there is enough evidence to conclude that the ownership of

livestock also affects the decision to commit to a trader in the long-run. Other factors that might

also affect such decision are the ownership of stall-fed cattle and the proportion of purchased

energy used, but they are not statistically significant at the 10% level. The results indicate that

peas grower experience a very strong learning effect from their peers in light of setting up a

long-term contract with a trader. This is a reasonable outcome since peas were introduced to

local farmers in the Thaltukhod Valley recently and farmers might not have sufficient knowledge

and market information about the production of peas.

6. Conclusion

This paper investigates the decision at the household level of small-scale farmers to

invest in a long-term contract with a trader to help them commercialize their cash crops. The data

used in this study comes from a survey conducted of 522 households in 17 villages in

Thaltukhod Valley in Himachal Pradesh, India. We put together a dataset containing the

household level individual characteristics that capture economic conditions, social status,

education level and elevation from the sea level. We also construct network variables that

indicate the type, diversity and position of each household within the social networks of each

29

village from the weights matrix that determine the links between the households. Then, we

perform econometric estimates of the likelihood that each household decides to establish a long-

term contract with a trader to help them sell their agricultural produce based on the individual

household’s characteristics and network indicators.

The main findings from this study can be summarized as follows. First, different

characteristics of a household’s village social network can either perform as complements

(influence, clustering) or substitutes (size) with a long-term contract with a trader as these

network characteristics reflect different levels of exposure to market information for each

agricultural household in each village of Thaltukhod Valley. Second, households’ individual

characteristics such as caste, opportunity cost of agricultural labor and the dependence on the

market consumption have positive correlation with the investment in a long-run contract with a

trader. And most importantly, farmers make the decision to commit to a trader given their crop

choice. Potatoes are more likely to be commercialized through a long-term contract with a trader

due to its high level of uncertainty, while kidney beans are least likely to be marketed through a

trader since they contain the lowest level of hidden information.

The results presented in this paper highlight the importance of social networks that could

potentially lead to the reduction in transaction costs that small-scale farmers in rural India have

to face. Although certain household and network characteristics as described in section 5 of this

study are likely to be more important in determining the decision to commit to a long-run

contract with a trader, we would like to further evaluate to what extent these factors matter.

Moreover, the regression results from the spatial econometric specification also indicate that the

one’s decision to make a specific investment depends on the decisions of farmer’s peers, rather

30

than the geographic distance. In other words, peer effects dominate geographical effects for a

rural Indian agricultural farmer in deciding to invest in a long-term relationship with a trader.

For future work, we plan to examine other household characteristics that could affect the

decision to use a trader. Some of the factors that we are considering to include in our estimation

models include the household level demand for labor in different production activities (forest,

agricultural, construction, and cultural purposes), the access to various local governance

institutions (both governmental and non-governmental), the geographic variables (distance to the

main town, slope and elevation of production plots) and the source of information regarding the

agricultural markets and technical knowledge for each household. We also plan to instrument for

the social network using household geographic proximity and agricultural plot-sharing to address

the network endogeneity. And finally, we plan to use a limited dependent spatial lag model to

obtain more efficient estimation results.

31

References

Agrawal, A. and K. Gupta (2005). Decentralization and pariticipation: The governance of

common pool resouces in Nepal’s Terai. World Development, 33(7), 1101-1114.

Akerlof, G. (1970), “The Market for Lemons: Qualitative Uncertainty and the Market

Mechanism,” Quarterly Journal of Economics, 84(3): 488-500.

Allen, D. and D. Lueck (1992), “Contract Choice in Modern Agriculture: Cash Rent Versus

Cropshare,” Journal of Law and Economics, 35: 367-426.

Allen, D. and D. Lueck (1999), “The Role of Risk in Contract Choice,” Journal of Law,

Economics, and Organization, 15:704-736.

Anselin, L. (1988), Spatial Econometrics: Methods and Models, Dordrecht, The Netherlands:

Kluwer Academic Publishers

Bandiera, O. and I. Rasul (2006), “Social Networks and Technology Adoption in Northern

Mozambique,” The Economic Journal, 116(514): 869-902.

Banerjee, A. et al. (2011), “The Diffusion of Microfinance,” MIT Working Paper.

Barr, A. M. (1997), Social Capital and Technical Information Flows in the Ghanian

Manufacturing Sector, Center for the Study of African Economies, Oxford University,

Oxford, UK.

Barrett, C. (1997), “Food marketing liberalization and trader entry: Evidence from Madagascar,”

World Development, 25(5): 763-777.

Batt, P. J. (2003), “Building Trust in a Filipino Seed Potato Industry,” Journal of Food &

Agribusiness Marketing, 13(4): 23-41.

Berry, S. (1993), No condition is permanent; The social dynamics of agrarian change in sub-

Saharan Africa. Madison, WI: University of Wisconsin Press.

Best, R., S. Ferris and D. Sautier (2005) “Building linkages and enhancing trust between small-

scale rural producers, buyers in growing markets and suppliers of critical inputs,” draft

paper presented at the Beyond Agriculture: Making Markets Work for

the Poor conference, 28 Feb.–1 Mar., London, UK.

Bramoulle, Y., H. Djebbari and B. Fortin (2009), “Identification of peer effects through social

networks,” Journal of Econometrics, 150: 41-55.

Brueckner, J. K. (2003), “Strategic Interactions among Governments: An Overview of Empirical

Studies,” International Regional Science Review, 26(2): 175-188.

32

Canjels, E. (1998), “Risk and Incentives in Sharecropping: Evidence from Modern U.S.

Agricultture,” New York: New School for Social Research, Schwartz Center for

Economic Policy Analysis, Working Paper No. 4.

Chandrasekhar, A. G. and R. Lewis (2011), “Econometrics of Sampled Networks,” MIT

Working Paper.

Coase, R. H. (1937), “The Nature of the Firm,” Economic, 4(16): 386-405.

Coleman, J. (1990), Foundations of social theory. Cambridge, MA: Harvard University Press.

Eswaran, M. and A. Kotwal (1986), “Access to Capital and Agrarian Production Organization,”

Economic Journal, 96: 482-498.

Fafchamps, M. (1998), “Market emergence, trust and reputation,” Stanford, CA: Stanford

University.

Fafchamps, M. and B. Minten (1999), “Relationships and traders in Madagascar,” Journal of

Development Studies, 35(6): 1-35.

Fafchamps, M. and B. Minten (2001), “Property rights in a flea market economy,” Economic

Development and Cultural Change, 49 (2): 229-266.

Fafchamps, M. and B. Minten (2002), “Social capital and the firm: evidence from agricultural

traders in Madagascar,” in C. Grootaert and T. van Bastelaer (eds.), The Role of Social

Capital in Development: An empirical assessment. Cambridge, UK: Cambridge

University Press.

Fukunaga, K. and W. Huffman (2009), “The Role of Risk and Transactions Costs in Contract

Design: Evidence from Farmland Lease Contracts in U.S. Agriculture,” American

Journal of Agricultural Economics, 91(1): 237-249.

Goetz, S. (1992), “A Selectivity Model of Household Food Marketing Behavior in Sub-Saharan

Africa,” American Journal of Agricultural Economics, 73(2): 496-501.

Jackson, M. (2008), Social and economic networks, Princeton University Press.

Jayne, T. (1994), “Do High Hood Marketing Costs Constrain Cash Crop Production? Evidence

from Zimbabwe,” Economic Development and Cultural Change, 42(2): 387-402.

Key, N., E. Sasdoulet and A. De Janvry (2000), “Transactions Costs and Agricultural Household

Supply Response,” American Journal of Agricultural Economics, 82(1): 245-259.

Kherallah, M. and J. F., Kirsten (2002), “The new institutional economics: applications for

agricultural policy research in developing countries,” Agrekon: Agricultural Economics

Research, Policy and Practice in Southern Africa, 41(2): 111-133.

33

Kirsten, J. and K. Sartorious (2002), “Linking agribusiness and small-scale farmers in

developing countries: Is there a new role for contract farming?” Development Southern

Africa, 19(4): 503-529.

Klein, B., R. Crawford and A. Alchian (1978), “Vertical Integration, Appropriable Rents,

and the Competitive Contracting Process," " Journal of Law and Economics, 21:297-326.

Kranton, R. E. (1996), “Reciprocal Exchange: A Self-Sustaining System,” American Economic

Review, 86(4): 830-851.

Lyon, F. (2000), “Trust, networks and norms: The creation of social capital in agricultural

economics in Ghana,” World Development 28(4): 663-681.

Manski, C. (1993). “Identification of endogenous social effects: The reflection problem,”

The Review of Economic Studies, 531-542.

Narayan, D. and L. Pritchett (1999), “Household Income and Social Capital in Rural Tanzania,”

Economic Development and Cultural Change 47(4): 871-897.

Omamo, S. (1998), “Transport Costs and Smallholder Cropping Choices: An Application to

Siaya District, Kenya,” American Journal of Agricultural Economics, 80(1): 116-123.

Paelinck, J. and L. Klaassen (1979), Spatial Econometrics, Farnborough, UK: Saxon House

Putnam R. (1993), Making Democracy Work: Civil Traditions in Modern Italy. Princeton, NJ:

Princeton University Press.

Rujis, A., C. Schweigman, and C. Lutz (2004), “The impact of transport-and transaction-cost

reductions on food markets in developing countries: evidence for tempered

expectations for Burkina Faso,” Agricultural Economics, 31(2-3): 219-228.

Spulber, D. (1999), Market Microstructure: Intermediaries and the theory of the firm. New

York, NY: Cambridge University Press.

Stigler, G. (1961) “The Economics of Information,” Journal of Political Economy, 69: 213-225.

Stigler, G. (1967) “Imperfections in the Capital Market,” Journal of Political Economy, 75: 287-

292.

Tirole, J. (1988), The Theory of Industrial Organization. Cambridge, MA: The MIT Press.

Udry, C. and T. Conley (2001), “Social Learning through networks: The adoption of new

agricultural technologies in Ghana,” American Journal of Agricultural Economics 83(3):

668-673.

34

Udry, C. and T. Conley (2005), “Social networks in Ghana,” in C. Barrett (ed.) The Social

Economics of Poverty: Identities, Groups, Communities and Networks, London, UK:

Routledge.

Udry, C. and T. Conley (2010), “Learning about a new technology: Pineapple in

Ghana,” American Economic Review 100: 35-69.

Williamson, O. E. (1979), “Transaction-Cost Economics: The Governance of Contractual

Relations,” Journal of Law and Economics, 22(2): 233-261.

Williamson, O. E. (1989), “Transaction Cost Economics,” in Handbook of Industrial

Organization, R. Schmalensee & R. Willig (eds.), Vol. 1, Elsevier.

Winter-Nelson, A. and A. Temu (2005), “Impacts of Prices and Transaction Cost on Input Usage

in a Liberalizing Economy: Evidence from Tanzanian Coffee Growers,” Agricultural

Economics, 33: 243-253.

Woolcock, M. (1998), “Social capital and economic development: Towards a theoretical

synthesis and policy framework,” Theory and Society, 8: 53-111.

35

Table 1: Descriptive Statistics (By Trader Use)

Households Housholds

not working with

trader working with trader

Variable Mean Std. Dev. Mean Std. Dev.

Degree 0.2027 0.1464 0.2162 0.1416

Two-step reach

0.5529

0.2788

0.6006

0.2632

Average reciprocal distance

0.4938

0.1412

0.5157

0.1309

Eigenvectors

0.1892

0.1544

0.2208

0.1424

Elevation (meters)

Income (rupees)

Land (bhigas)

Livestock (units)

Stall-fed cattle (number)

2036.13

22161.61

8.1896

0.4882

0.4645

191.771

22259.96

9.7174

0.8472

0.8121

2049.21

25048.17

8.4260

0.6323

0.5498

199.055

50574.21

5.7224

1.1003

0.8132

Purchased energy (%)

0.2559

2.0727

0.6688

4.1038

Own-food consumption (months)

3.0185

1.334

2.955

1.1985

Family size (head count)

5.654

2.3982

5.7621

2.3171

Caste (0=lower, 1=higher)

0.7962

0.4038

0.8842

0.3204

Observations 311 211

Source: Field Household Survey, 2008.

36

Table 2: Descriptive Statistics (By Crop Choice)

Kidney beans Potatoes Peas

Variable Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.

Degree 0.2628 0.1453 0.2111 0.1436 0.2576 0.1529

Two-step reach 0.6909 0.2435 0.5821 0.2701 0.6767 0.2554

Average reciprocal distance 0.5624 0.1281 0.5072 0.1352 0.5548 0.1356

Eigenvectors 0.2386 0.1401 0.2084 0.1479 0.2349 0.139

Elevation (meters) 2158.43 152.89 2044.37 195.98 2170.2 155.28

Income (rupees) 19486.36 18163.75 23892.67 41555.61 20819.13 19293.23

Land (bhigas) 9.1374 8.8335 8.3369 7.5927 9.9746 9.065

Livestock (units) 0.5199 1.0618 0.5739 1.0071 0.5797 1.1205

Stall-fed cattle (number) 0.4437 0.7527 0.5163 0.8135 0.5109 0.8112

Purchased energy (%) 0.2848 2.7483 0.4645 3.3292 0.4783 4.0366

Own-food consumption

(months)

2.7483 0.953 2.9864 1.2485 2.9388 1.0199

Family size (head count) 5.7682 2.3112 5.7159 2.3501 5.7355 2.2944

Caste (0=lower, 1=higher) 0.8808 0.3246 0.8503 0.3571 0.9601 0.196

Observations 302 521 276

Source: Field Household Survey, 2008.

37

Table 3: Marginal Effects at the Mean of the Data of Logit Regressions on the Long-term Contract with a Trader

(1) (2) (3) (4)

No FE Village FE Crop FE Village + Crop

FE

Degree -0.326 -1.552** -0.439 -0.161**

(0.293) (0.651) (0.314) (0.671)

Two-step reach 0.192 -0.276 0.072 -0.226

(0.219) (0.352) (0.235) (0.368)

ARD -0.192 1.870 0.283 1.802

(0.591) (1.230) (0.632) (1.261)

Eigenvector 0.263 0.276 1.495 0.257

(0.168) (0.197) (0.178) (0.204)

Elevation -0.0003***

-0.0001

-0.00003 -0.0008

(0.0001) (0.0071) (0.0001) (0.0073)

Land 0.017 0.046 0.049 0.051

(0.038) (0.043) (0.040) (0.046)

Livestock 0.189** 0.211** 0.273*** 0.253***

(0.083) (0.087) (0.092) (0.095)

Family size 0.004 0.004 0.003 0.006

(0.007) (0.007) (0.007) (0.007)

Caste 0.160*** 0.123** 0.130*** 0.145**

(0.048) (0.073) (0.050) (0.068)

Purchased fuel 0.003

0.003 0.004 0.004

(0.004) (0.004) (0.004) (0.004)

Own-food -0.011 -0.027 -0.025 -0.027

consumption (0.017) (0.019) (0.017) (0.020)

Stall-fed cattle -0.070 -0.086* -0.106** -0.106**

(0.047) (0.049) (0.052) (0.053)

Village 2 0.432*** 0.383**

(0.136) (0.188)

Village 3 0.066 -0.085

(0.197) (0.179)

Village 4 0.245 -0.195

(0.263) (0.182)

Village 5

0.132

-0.270*

(0.303) (0.142)

38

Village 6 0.237

0.243

(0.168) (0.192)

Village 7 0.278 -0.189

(0.289) (0.216)

Village 8

0.142

0.357*

(0.206) (0.661)

Village 9

-0.011

0.027

(0.098) (0.108)

Village 10

0.037

0.009

(0.103) (0.108)

Village 11 0.011* 0.272*

(0.076) (0.140)

Village 12 0.057 -0.035

(0.085) (0.738)

Village 13 0.134 0.083

(0.135) (0.094)

Village 14 -0.043 -0.123

(0.125) (0.112)

Village 15 0.077 -0.117

(0.171) (0.141)

Village 16

Village 17

Kidney beans

Potatoes

0.299

(0.196)

-0.010

(0.249)

-0.335***

(0.004)

0.238***

(0.052)

-0.116

(0.183)

-0.166

(0.186)

-0.334***

(0.034)

0.251***

(0.039)

Observations 1099 1099 1099 1099

Standard errors in parentheses * p < 0.1,

** p < 0.05,

*** p < 0.01

Dummy variables for Village 1 and peas are automatically omitted.

39

Table 4: Coefficient Esimates of Spatial Lag Regressions on the Long-term Contract with a Trader based on Social

Network Weights Matrix

(1) (2) (3) (4)

No FE Village FE Crop FE Village + Crop

FE

Degree -0.319 -1.243** -0.378 -0.175**

(0.281) (0.565) (0.255) (0.515)

Two-step reach 0.194 -0.215 0.086 -0.175

(0.212) (0.309) (0.193) (0.282)

ARD -0.213 1.159 0.122 0.298

(0.570) (1.037) (0.517) (0.946)

Eigenvector 0.252 0.237 0.129 0.229

(0.164) (0.188) (0.0455) (0.171)

Elevation -0.0003

0.0002

0.00008 -0.0005

(0.0001) (0.0006) (0.0001) (0.0006)

Land 0.019 0.072* 0.055 0.071

(0.037) (0.043) (0.034) (0.039)

Livestock 0184** 0.197** 0.210*** 0.196***

(0.037) (0.082) (0.073) (0.075)

Family size 0.004 0.004 0.003 0.004

(0.007) (0.007) (0.006) (0.007)

Caste 0.161*** 0.092 0.095* 0.118

(0.055) (0.079) (0.050) (0.073)

Purchased fuel 0.003

0.005 0.005 0.005

(0.047) (0.004) (0.003) (0.004)

Own-food -0.012 -0.036* -0.029* -0.033

consumption (0.017) (0.019) (0.015) (0.017)

Stall-fed cattle -0.065 -0.068 -0.071* -0.067

(0.046) (0.047) (0.037) (0.042)

Village 2 0.378** 0.248

(0.189) (0.173)

Village 3 0.115 -0.050

(0.184) (0.168)

Village 4 0.407 -0.066

(0.262) (0.241)

Village 5 0.320

(0.283)

-0.167

(0.261)

40

Village 6

0.278*

0.246

(0.167) (0.152)

Village 7 0.427 -0.077

(0.297) (0.274)

Village 8

0.046

0.224

(0.192) (0.176)

Village 9

0.237*

0.210

(0.861) (0.850)

Village 10

0.009

0.019

(0.095) (0.088)

Village 11 0.242* 0.258**

(0.125) (0.113)

Village 12 0.067 0.079

(0.079) (0.072)

Village 13 0.157 0.222*

(0.127) (0.116)

Village 14 0.051 -0.028

(0.125) (0.113)

Village 15 0.268 0.071

(0.167) (0.153)

Village 16

Village 17

Kidney beans

Potatoes

Rho

Intercept

0.007

(0.011)

0.961

(0.269)

0.490**

(0.217)

0.159

(0.238)

0.040***

(0.011)

-0.757

(1.469)

-0.264***

(0.037)

0.258***

(0.035)

0.029***

(0.010)

0.213

(0.254)

0.065

(0.200)

0.010

(0.218)

-0.263***

(0.037)

0.255***

(0.036)

0.039***

(0.010)

0.762

(1.344)

Observations 1099 1099 1099 1099

Standard errors in parentheses * p < 0.1,

** p < 0.05,

*** p < 0.01

Dummy variables for Village 1 and peas are automatically omitted.

41

Table 5: Coefficient Esimates of Spatial Lag Regressions on the Long-term Contract with a Trader based on

Geographical Distance Weights Matrix

(1) (2) (3) (4)

No FE Village FE Crop FE Village + Crop

FE

Degree -0.343 -1.333** -0.389 -0.127**

(0.285) (0.567) (0.258) (0.516)

Two-step reach 0.173 -0.190 0.054 -0.155

(0.214) (0.310) (0.194) (0.282)

ARD -0.140 1.457 0.258 1.291

(0.574) (1.038) (0.521) (0.946)

Eigenvector 0.262 0.272 0.151 0.262

(0.164) (0.186) (0.146) (0.170)

Elevation -0.0003***

-0.00007

0.00003 -0.0005

(0.0001) (0.0004) (0.0001) (0.0006)

Land 0.016 0.045 0.043 0.045

(0.037) (0.042) (0.034) (0.038)

Livestock 0188** 0.202** 0.224*** 0.201***

(0.080) (0.082) (0.063) (0.074)

Family size 0.004 0.004 0.003 0.005

(0.007) (0.007) (0.006) (0.007)

Caste 0.169*** 0.132* 0.124* 0.158**

(0.054) (0.078) (0.049) (0.071)

Purchased fuel 0.003

0.003 0.004 0.003

(0.004) (0.004) (0.004) (0.004)

Own-food -0.011 -0.026 -0.085 -0.023

consumption (0.017) (0.018) (0.041) (0.017)

Stall-fed cattle -0.069 -0.083* -0.071** -0.082*

(0.046) (0.046) (0.041) (0.042)

Village 2 0.413** 0.282*

(0.188) (0.171)

Village 3 0.086 -0.080

(0.183) (0.167)

Village 4 0.271 -0.199

(0.258) (0.238)

Village 5 0.164

(0.279)

-0.319

(0.257)

42

Village 6

0.238

0.206

(0.167) (0.152)

Village 7 0.300 -0.200

(0.294) (0.270)

Village 8

0.127

0.300*

(0.192) (0.175)

Village 9

-0.081

0.032

(0.096) (0.087)

Village 10

0.044

0.009

(0.096) (0.087)

Village 11 0.199 0.216*

(0.124) (0.113)

Village 12 0.059 0.070

(0.080) (0.073)

Village 13 0.115 0.183

(0.127) (0.116)

Village 14 0.023 -0.102

(0.123) (0.112)

Village 15 0.901 -0.102

(0.159) (0.146)

Village 16

Village 17

Kidney beans

Potatoes

Rho

Intercept

-0.013

(0.029)

0.918***

(0.270)

0.321

(0.211)

0.038

(0.235)

0.012

(0.030)

-0.014

(1.453)

-0.268***

(0.037)

0.245***

(0.034)

0.005

(0.002)

0.141

(0.255)

-0.100

(0.195)

-0.110

(0.215)

-0.264***

(0.037)

0.252***

(0.036)

0.008

(0.027)

1.352

(1.329)

Observations 1099 1099 1099 1099

Standard errors in parentheses * p < 0.1,

** p < 0.05,

*** p < 0.01

Dummy variables for Village 1 and peas are automatically omitted.

43

Table 6: Marginal Effects at the Mean of the Data of Logit Regressions on the Long-term Contract with a Trader for

Potato and Pea Farmers

(1) (2) (3) (4)

No FE

Potatoes

Village FE

Potatoes

No FE

Peas

Village FE

Peas

Degree -0.541 -0.844 -0.552 -0.679**

(0.436) (0.819) (0.563) (6.840)

Two-step reach 0.214 0.066 0.675 -0.775

(0.325) (0.461) (0.437) (3.489)

ARD 0.038 0.367 -0.262 0.255

(0.851) (1.443) (1.180) (13.230)

Eigenvector 0.442* 0.486* 0.198 1.086

(0.022) (0.271) (0.413) (2.776)

Elevation -0.0001

-0.0002

-0.0001 -0.005

(0.0002) (0.0009) (0.0002) (0.012)

Land 0.079 0.078 0.008 -0.148

(0.055) (0.063) (0.0076) (0.388)

Livestock 0.304** 0.280* 0.460* 1.413

(0.150) (0.149) (0.255) (0.957)

Family size -0.002 -0.001 0.010 0.062

(0.010) (0.011) (0.014) (0.067)

Caste 0.116 0.168 -0.191 0.778

(0.074) (0.110) (0.137) (1.295)

Purchased fuel 0.029**

0.026* 0.039 -0.032

(0.238) (0.015) (0.028) (0.061)

Own-food -0.049** -0.053** -0.034 0.147

consumption (0.022) (0.026) (0.038) (0.196)

Stall-fed cattle -0.134 -0.128 -0.204 -0.520

(0.083) (0.015) (0.140) (0.522)

Village 2 0.165 0.014

(0.198) (1.377)

Village 3 0.152 -0.143

(0.213) (1.475)

Village 4 0.144 -0.055

(0.295) (0.971)

Village 5 -0.003

(0.382)

-0.005

(0.954)

44

Village 6

0.182

0.314

(0.193) (1.072)

Village 7 0.097 0.102

(0.369) (0.592)

Village 8

0.273*

0.209*

(0.146) (0.661)

Village 9

0.076

0.210

(0.142) (0.850)

Village 10

0.223*

0.710

(0.104) (0.543)

Village 11 0.251* 0.122

(0.120) (0.780)

Village 12 0.007 -0.035

(0.120) (0.738)

Village 13 0.136 0.015

(0.111) (0.887)

Village 14 0.045 -0.028

(0.177) (1.122)

Village 15 0.083

(0.196)

Village 16

Village 17

0.137

(0.231)

0.112

(0.282)

Observations 521 521 276 276

Standard errors in parentheses * p < 0.1,

** p < 0.05,

*** p < 0.01

Dummy variables for Village 1 and peas are automatically omitted. Peas are not grown in Villages 15, 16 and 17.

45

Table 7: Coefficient Esimates of Spatial Lag Regressions on the Long-term Contract with a Trader for Potato and

Pea Farmers based on Social Network Weights Matrix

(1) (2) (3) (4)

No FE

Potatoes

Village FE

Potatoes

No FE

Peas

Village FE

Peas

Degree -0.475 -0.759 -0.554 -0.734

(0.422) (0.759) (0.555) (0.761)

Two-step reach 0.190 0.007 0.630 0.453

(0.311) (0.424) (0.427) (0.480)

ARD 0.020 0.289 -0.196 -0.144

(0.813) (1.300) (1.155) (1.032)

Eigenvector 0.351 0.362 0.222 0.485

(0.237) (0.269) (0.405) (0.466)

Elevation -0.0001

-0.0002

-0.0002 -0.0003

(0.0001) (0.0008) (0.0002) (0.0004)

Land 0.078 0.077 0.010 0.024

(0.052) (0.060) (0.077) (0.077)

Livestock 0.024** 0.219 0.331** 0.355**

(0.119) (0.122) (0.163) (0.169)

Family size -0.001 -0.0005 0.001 0.013

(0.010) (0.010) (0.014) (0.013)

Caste 0.106 0.148 -0.213 -0.150

(0.070) (0.100) (0.198) (0.192)

Purchased fuel 0.013**

0.014** 0.013 0.014

(0.006) (0.007) (0.007) (0.009)

Own-food -0.045** -0.049** -0.030 -0.044

consumption (0.021) (0.024) (0.038) (0.031)

Stall-fed cattle -0.099 -0.093 -0.133 -0.223

(0.067) (0.069) (0.092) (0.087)

Village 2 0.183 0.014

(0.244) (0.175)

Village 3 0.167 0.143

(0.250) (1.475)

Village 4 0.151 0.055

(0.340) (0.971)

Village 5 0.004

(0.367)

0.040

(1.592)

46

Village 6

0.231

0.314

(0.265) (1.072)

Village 7 0.099 0.102

(0.389) (0.592)

Village 8

0.354

0.209

(0.265) (0.661)

Village 9

0.101

0.210

(0.152) (0.850)

Village 10

0.266*

0.710

(0.152) (0.543)

Village 11 0.323 0.122*

(0.204) (0.780)

Village 12 0.159 0.035

(0.131) (0.738)

Village 13 0.159 0.015

(0.199) (0.887)

Village 14 0.583 0.028

(0.181) (1.122)

Village 15 0.991

(0.212)

Village 16

Village 17

Rho

Intercept

0.027

(0.030)

0.592

(0.384)

0.149

(0.273)

0.134

(0.313)

-0.003

(0.051)

0.718

(1.926)

-0.018

(0.075)

0.952*

(0.61)

-0.032*

(0.026)

0.996*

(0.784)

Observations 521 521 276 276

Standard errors in parentheses * p < 0.1,

** p < 0.05,

*** p < 0.01

Dummy variables for Village 1 and peas are automatically omitted. Peas are not grown in Villages 15, 16 and 17.

47

Table 8: Coefficient Esimates of Spatial Lag Regressions on the Long-term Contract with a Trader for Potato and

Pea Farmers based on Geographical Distance Weights Matrix

(1) (2) (3) (4)

No FE

Potatoes

Village FE

Potatoes

No FE

Peas

Village FE

Peas

Degree -0.522 -0.809 -0.469 -0.554

(0.419) (0.758) (0.551) (0.555)

Two-step reach 0.195 0.049 0.605 0.630

(0.312) (0.422) (0.424) (0.427)

ARD 0.075 0.345 -0.189 -0.196

(0.813) (1.297) (1.146) (1.055)

Eigenvector 0.415* 0.472 0.469 0.222

(0.227) (0.259) (0.424) (0.405)

Elevation -0.0001

-0.0002

-0.0002 -0.0002

(0.0001) (0.0008) (0.0002) (0.0002)

Land 0.077 0.071 0.027 0.010

(0.052) (0.060) (0.075) (0.077)

Livestock 0.254** 0.234* 0.360** 0.331**

(0.118) (0.122) (0.163) (0.163)

Family size -0.0008 -0.0007 0.011 0.010

(0.010) (0.010) (0.013) (0.014)

Caste 0.114 0.168 -0.182 -0.213

(0.070) (0.100) (0.197) (0.198)

Purchased fuel 0.013**

0.013* 0.012 0.013

(0.006) (0.007) (0.007) (0.004)

Own-food -0.045** -0.047* -0.039 -0.030*

consumption (0.021) (0.024) (0.037) (0.038)

Stall-fed cattle -0.107 -0.105 -0.150 -0.133

(0.067) (0.069) (0.092) (0.092)

Village 2 0.174 0.014

(0.245) (0.175)

Village 3 0.108 0.143

(1.464) (1.475)

Village 4 0.006 0.055

(1.038) (0.971)

Village 5 0.008

(1.584)

0.040

(1.592)

48

Village 6

0.316

0.314

(1.073) (1.072)

Village 7 0.123 0.102

(0.579) (0.592)

Village 8

0.229

0.209

(0.647) (0.661)

Village 9

0.237

0.210

(0.861) (0.850)

Village 10

0.790*

0.710*

(0.535) (0.543)

Village 11 0.350 0.122*

(0.785) (0.780)

Village 12 0.007 0.035

(0.704) (0.738)

Village 13 0.061 0.015

(0.849) (0.887)

Village 14 0.029 0.028

(1.077) (1.122)

Village 15 0.153

(1.231)

Village 16

Village 17

Rho

Intercept

-0.001

(0.052)

0.620

(0.052)

0.168

(0.274)

0.130

(0.314)

0.044

(0.029)

0.612

(1.930)

-0.039

(0.023)

0.989

(0.598)

-0.032

(0.026)

0.996

(0.784)

Observations 521 521 276 276

Standard errors in parentheses * p < 0.1,

** p < 0.05,

*** p < 0.01

Dummy variables for Village 1 and peas are automatically omitted. Peas are not grown in Villages 15, 16 and 17.

49

Figure 1: The study area is Thaltukhod Valley in the Indian Himalayas,

consisting of 17 villages and 522 households.

Source: Survey Data (2008)

50

Figure 2: Village 6 - Tegar

51

Figure 3: Village 14 - Bhumchayan

52

Figure 4: Agricultural Producers in Thaltukhod Valley based on Cash Crop

53

Figure 5: Traders’ Activity in Thaltukhod Valley based on Cash Crop