Impacts of Agricultural Cooperatives on Farmers ’ Revenues in Cambodia : A Case Study of Tram Kak District , Takeo Province

Agricultural cooperatives in Cambodia have been promoted with the aim of increasing agricultural production and farmers’ revenues. The objectives of this study are to identify factors influencing farmers’ decision on membership in agricultural cooperatives, and to assess the impact of being a member in those cooperatives on farmers’ revenues from paddy, livestock and farm. Cross-sectional data from interviews of 242 households in Tram Kak District, Takeo Province were used. The probit model and propensity score matching were employed to achieve the objectives. The results show that farmers who sold their paddy and had been contacted by extension workers from the government agency and non-governmental organizations (NGOs) are more likely to join the cooperatives while male-headed household farmers and farmers who have high off-farm income are less likely to become members of the cooperatives. Moreover, the results of propensity score matching reveal that agricultural cooperatives have no impact on paddy yields and paddy revenue due to the fact that agricultural cooperatives do not provide sufficient training to their members, and members did not actively attend those trainings provided. Also, the cooperatives have failed to provide members better prices for their paddy. There are positive impacts on their livestock and farm revenues through increasing livestock and other crop production when agricultural cooperatives provide livestock and other crop training to their members. However, there is no impact on non-members if they join the cooperatives as they have higher off-farm income, less paddy land size and fewer laborers that are not favorable to taking on other farming activities.


Introduction
The population of Cambodia was estimated at 14.68 million in 2013 (National Institute of Statistics [NIS], 2013).Among the total 3.16 million households, 2.5 million households lived in rural areas (Asian Development Bank [ADB], 2014).Agriculture shared more than 30% of the gross domestic product (GDP) (Ministry of Agriculture, Forestry and Fisheries [MAFF], 2015), and it employed approximately 45.3% of the total workforce in 2014 (MAFF, 2016).Cultivated areas for agriculture in Cambodia were estimated at 3.7 million hectares, of which three quarters were used for paddy, a staple crop and source of income for most farmers, and the remaining areas were used for producing other food and industrial crops (Food and Agricultural Organization [FAO], 2014).Due to the significance of agriculture in Cambodia, MAFF has initiated programs to promote the agricultural cooperative movement in the country.These programs are intended to boost agricultural production, diversify crop production, create income-generating activities through business development and also expand markets for commercializing all kinds of agricultural products produced by the cooperative members (MAFF, 2008).The development of agricultural cooperatives has been in focus in order to ease the development of agriculture sector, to collectively link with private sectors, to gain technology and credit, to stabilize the food supply to local and international markets, and especially to develop agricultural cooperatives as rural agricultural enterprises with the purpose of improving rural socio-economics (MAFF, 2016). in agricultural cooperatives on yield, paddy revenue, livestock revenue and farm revenue in Cambodia.This study has two objectives: 1) to identify factors influencing farmers' decision on membership in agricultural cooperatives, and 2) to assess the impact of being a member in agricultural cooperatives on farmers' revenues from paddy, livestock and farm.

Study Site and Data Collection
The data collection was conducted in September and October 2016 in Tram Kak District, Takeo Province.A total of 242 farmers (99 members from 10 agricultural cooperatives and 143 non-members) were randomly selected and interviewed using face-to-face structured interviews.Qualitative interviews were also conducted with directors of those agricultural cooperatives in order to understand more about the situations and problems they have faced.Takeo Province is located in the southern part of Cambodia, and it is one of the most important paddy-producing provinces in the country.According to MAFF annual report 2016, this province has 88 agricultural cooperatives, the largest number of agricultural cooperatives among various provinces in Cambodia.Takeo Province has 10 districts and, based on data obtained from the Cambodian MAFF, Tram Kak District has the largest number of agricultural cooperatives in this province with a population of 181,258 (National Committee for Sub-National Democratic Development [NCDD], 2010).All agricultural cooperatives having paddy business in this district were selected.In addition, these agricultural cooperatives also had some agricultural training such as paddy, livestock and other crop production training.

Empirical Models
For the first objective, a probit model was used to identify factors influencing farmers' decision on membership in agricultural cooperatives.Age, gender, education of household head, household size, paddy land size, paddy sale, off-farm income, TV, car, contact with extension workers and access to a good road were used as independent variables (Table 1).For the second objective, the propensity score matching, (PSM) using the single nearest neighbor matching, was employed to assess the impact of being a member in agricultural cooperatives on paddy yield, paddy revenue per hectare, livestock and farm revenues per year (Table 1).
In order to acquire a realistic estimation of adoption impact, we needed to set a control group with similar attributes as much as possible similar to those of the treated group (Monteiro, 2010).According to Rosenbaum and Rubin (1983), PSM has become the common approach used in impact evaluation as it can control the observable characteristics of the control group as a resemblance of the treated group, that is to say it is a method that could establish a counterfactual condition and reduce possible selection bias involved with observable characteristics.
PSM is a two-step procedure (Becker & Ichino, 2002).In first step, the probit model is used for the decision to become a member of an agricultural cooperative, and this will provide a propensity score for each observation.Propensity scores of farmers were calculated by estimating the probability model in the probit model, specified as: where, y is a dependent variable (1 = member of agricultural cooperative; 0 = non-member), β is the regression coefficient to be estimated, and X is an independent variable to be explained.X 1 is the age of household head, X 2 is the gender of household head, X 3 is the years of education of household head, X 4 is the number of household members, X 5 is paddy land size, X 6 is paddy sale, X 7 is annual income of household head from off-farm job, X 8 is household having TV, X 9 is household having car, X 10 is having contact with extension workers related to agricultural cooperatives, and X 11 is access to good road in village (Table 1).
After estimating the probability model, we estimate the propensity score based on the following equation: where, (β 0 + β 1 X 1 + β 2 X 2 + … β n X n ) was used in the probit model as shown in Equation (1).
Propensity score was defined as the conditional probability of receiving treatment given a vector of observable covariates (Rosenbaum & Rubin, 1983).After the propensity score is estimated, each member of an agricultural cooperative was matched with non-members with similar propensity score values with the aim of estimating the average treatment effect on the treated (ATT), which is notated, where, D is an indicator variable equal to 1 if the farmer is a member, Y 1 is the members' outcomes, Y 0 is the non-members' outcomes and x is a vector of the control variables.Outcome variables used this study are paddy yield, paddy revenue, livestock revenue and farm revenue (Table 1).
After matching, a balancing test is required to verify that the differences in the control variables between member group and non-member group have been eradicated, in which the matched comparison group could be regarded as a credible counterfactual (Ali & Abdulai, 2010).Even though there are many kinds of balancing tests, the most commonly adopted is the mean absolute standardized bias (MASB) method.Therefore, we used the MASB approach as recommended by Rosenbaum and Rubin (1983), in which the standardized difference should be smaller than 20% to prove the success in the matching procedure.

Description of Data Variables
Table 1 shows the variables used in this study, and it describes the variable names, definition and unit of each variable.Farmer status was used as the dependent variable while age, gender, education of household head, household size, paddy land size, paddy sale, annual income from off-farm jobs, TV, car, contact with extension workers and access to good road were used as independent variables in the probit model to identify factors influencing membership in agricultural cooperatives.Moreover, paddy yield, paddy revenue, livestock revenue and farm revenue were used as outcome variables in PSM.

Descriptive Results before and after Matching
Table 2 shows the characteristic differences between members and non-members before and after matching.Before matching, household size, paddy land size, paddy sale and contact with extension workers were significantly different between members and non-members.On average, household size of members was 4.68 while the household size of non-members was 3.83.Moreover, members had paddy land size of 0.97 hectare, and this is 0.19 hectare bigger than non-members.In addition, 0.82% of members sold their paddy, which was 0.19% higher than non-members.Based on the unmatched results, 0.87% of members been in contact with extension workers compared to only 0.08% of non-members having had contact with extension workers.Outcome variables including paddy yield, paddy revenue, livestock revenue and farm revenue are also presented in Table 2. Livestock revenues of members was US$421.61 per year, which is US$219.41significantly higher than non-members.Also, members got farm revenues of US$1,291.26per year, US$322.83statistically more than non-members.From simple comparison, results suggest that members obtained higher livestock revenue and farm revenue than non-members before matching.However, the differences in outcomes before matching may be caused by characteristics differences rather than being a member.It may lead to biased conclusion if we do not control these differences.Thus, we employed PSM to control these differences of characteristics in order to get unbiased results.
The mean absolute standardized bias was 17.1% and as Rosenbaum and Rubin (1983) suggested that the mean absolute standardized bias should be smaller than 20%, this confirms the success in the matching process.After matching, the differences between members and non-members were reduced.Only education and household size were still significant after we conducted matching process.Source: Own survey (2016).

Determinants of Membership in Agricultural Cooperatives
As the results of coefficients in the probit estimation could not be interpreted directly, the marginal effects of independent variables of becoming a member of agricultural cooperatives were used and are shown in Table 3, and the units of those marginal effects are the same as the units of measurement for the explanatory variables (Greene, 2013).According to the probit estimates in Table 3, paddy sale and having contact with extension workers are positively associated with the decision to become members of agricultural cooperatives, while a male-headed household and off-farm income are negatively associated.For gender of household heads, the result of marginal effects shows that if the household heads are males, the probability of becoming a member of agricultural cooperatives decrease by 0.11 (holding all other variables constant) compared to female household heads.This may be due to the fact that male household heads mostly have off-farm jobs, so they do not want to join.On the other hand, female-headed households are generally poor, so they want to join the cooperative to receive agricultural techniques and other services from the cooperatives.This is contrary to the finding of Bernard and Spielman (2009), and Abebaw and Haile (2013) who found that woman-headed households were less likely to join the cooperatives in Ethiopia.Also, Mayoux (1999) mentioned that females in Africa have a limited chance of joining in collective activities such as cooperatives.For paddy sale, the probability of becoming a member in agricultural cooperatives of farmers who sell their paddy increases by 0.09 (holding all other variables constant) compared to farmers who did not sell their paddy.This is because they want to acquire rice-growing techniques and want to sell their paddy for better prices.Based on the results of marginal effects, with one percent increase in off-farm income, the probability of becoming a member of agricultural cooperatives decreases by 0.06 (holding all other variables constant).Farmers who have higher off-farm income are less likely to join the cooperatives because they are busy with off-farm jobs, and rice is not their main source of income.This is consistent with the finding of Nugusse, Huylenbroeck, and Buysse (2012), who found that households with special skills other than farming were less likely to join the cooperatives in Northern Ethiopia.Moreover, farmers who have been in contact with extension workers are more likely to join the cooperatives because they had got the information on the benefits of the cooperatives, and their probability of becoming a member of an agricultural cooperative increases by 0.46 holding all other variables constant.This result is in line with Debeb and Haile (2016), who found that access to information on the benefits of agricultural cooperatives encouraged farmers to join the cooperatives.Source: Own survey (2016).

Impacts of Agricultural Cooperatives on Farmers' Revenues
After matching, each member of the agricultural cooperatives was matched with non-members with similar propensity score values to estimate the average treatment effect for the treated (ATT) and average treatment effect for the untreated (ATU).The results of propensity score matching in Table 4 show that before matching, on average, paddy yields of members and non-members are 2,889.08Kg/ha and 2956.46Kg/ha, and members and non-members have paddy revenues of US$815.57and US$822.22 per hectare respectively.However, there are no significant differences before and after matching.These results suggest that membership in agricultural cooperatives has no impact on paddy yield and revenue as there is no significant difference between members and non-members with and without the matching process.This may be due to the fact that the agricultural cooperatives have not provided sufficient training, and members did not actively attend those trainings that were provided.Furthermore, the cooperatives have failed to provide better prices compared to other traders as they have small capital and the capability of the management committees is limited.This result is consistent with Afolami, Obayelu, Agbonlahor, and Lawal-Adebowale (2012), who also found no significant difference in yields between non-members and members of rice agricultural cooperatives in Nigeria.Similarly, Hoken and Su (2015) also found no significant difference in net income between participants and non-participants in rice-producing cooperatives in China.Being a member in agricultural cooperatives, members sampled could obtain US$219.41 and US$403.42respectively from livestock and farm significantly higher than non-members.These results show the positive impact on livestock and farm revenues for members, according to ATT, but there is no impact for non-members if they join the cooperatives according to ATU.The cooperatives provide training on livestock production and encourage members to raise more livestock compared to non-members who have no or fewer livestock, so this leads to positive impacts.However, there is no impact on non-members if they join the cooperatives, as they have higher off-farm income, less paddy land size, and fewer laborers, conditions that are not favorable for them to undertake other farming activities.Note.*, **, *** significant at 10%, 5%, 1%, respectively; S.E. is standard error.ATT: average treatment effect for the treated; ATU: average treatment effect for the untreated.

Conclusion and Policy Implications
In conclusion, farmers who sold their paddy and farmers who had contact with extension workers are more likely to join the cooperatives.Male farmers and higher off-farm-income farmers are less likely to join the cooperatives.
Agricultural cooperatives have no impact on paddy yield and paddy revenue, but there are positive impacts on livestock and farm revenues for members as they can increase their livestock and other agricultural production when obtaining agricultural training from the cooperatives.However, there is no impact on non-members if they join the cooperatives, as they have higher off-farm income, less paddy land size and fewer laborers, which are not favorable to undertaking other farming activities.
Based on the results of this study, some recommendations can be drawn.The extension workers should disseminate the benefits of agricultural cooperatives to farmers more widely.Moreover, the cooperatives should focus more on farmers who have no or lower off-farm income and female-headed households and encourage farmers to commercialize themselves in rice and other agricultural activities to gain more benefits from agricultural cooperatives.Furthermore, agricultural cooperatives should provide more training on rice production and encourage members to actively join such training.Moreover, the cooperatives should strengthen and expand paddy markets to get better prices for their members through some means such as contract farming with millers.Additionally, the capital of the cooperatives should be increased to sustain their management system and to enable them to compete with other traders.Relevant institutions should provide capacity-building training or study tours to management committees, and management committees should be regularly monitored and advised by technical officers from the government or NGOs.
of agricultural cooperative; 0 = non-member Independent variables (used in probit model) Age Age of household head Year Gender Gender of household head; 1 = male; 0 = female Dummy Education Years of education of household head Year Household size Number of household members Number Paddy land Paddy land size Hectare Paddy sale Farmers who sell their paddy = 1; 0 = otherwise Dummy Off-farm Annual income of household head from off-farm job US $ TV owned Household having TV = 1; 0 = otherwise Dummy Car Household having car = 1; 0 = otherwise Dummy Extension Having contact with extension workers related to agricultural cooperatives = 1; 0 = otherwise Dummy Access to road Access to good road in village = 1; 0 = otherwise Dummy Outcome variables (used in matching of propensity score) Paddy yield Yield per hectare Kg/ha Paddy revenue Total revenue from paddy per hectare US $/ha Livestock revenue Total revenue from animals (pigs and poultry) per year US $ Farm revenue Total revenue from farm activities (paddy, crop, animal, aquaculture) per year US $

Table 1 .
Definition of variables Variables Definition UnitDependent variable (used in probit model)

Table 4 .
Results of propensity score matching