Propensity Score Matching Analysis

As the above comparisons do not control for farmers' characteristic differences, the mean differences in farming performance between contract and non-contract farmers may be caused by farmers' characteristics rather than their contract or non-contract states. In the following we use the “propensity score matching” (p-score) method (Becker and Ichino, 2002) to conduct a more refined comparison by controlling for farmers' characteristic differences.

The first step of the p-score approach is to estimate farmers' propensity scores based on their basic characteristics (i.e., characteristics that are not affected by the choice of contract). The propensity score of each farmer measures his/her tendency to join the contract. The magnitude of a propensity score is between 0 and 1; the larger the score, the more likely the farmer would be to join the contract.

After farmers' propensity scores are estimated, the second step is to divide farmers into groups. Farmers in each group have similar propensity scores. In addition, each group should be balanced in the sense that the basic characteristics of the farmers in it are not significantly different.

After the balanced groups are formed, we can compare different types of farmers in each group. As such comparisons control for farmers' characteristic differences, the performance differences between contract and non-contract farmers are more likely to be caused by contract farming rather than by farmers' basic characteristics.

The above p-score comparison method is usually called “stratification” comparison in that the two groups under comparison are stratified into one-to-one matching sub-groups for comparison. Besides the stratification comparison, another comparison method called the “nearest neighbor” comparison is to compare each contract farmer to the non-contract farmer with the most similar p-score (Becker and Ichino, 2002).

In this paper we use the stratification comparison as the main approach and the nearestneighbor comparison as an additional approach to enhance the robustness of the comparisons. For example, if both comparison approaches indicate that contract farmers have higher profits than never-contract farmers, and the differences are statistically significant, we would have the confidence to conclude that contract farming tends to improve profitability. If both approaches indicate that contract farmers have higher profits, and the difference is statistically significant under one approach but not under the other, the conclusion that contract farming improves profitability would still be sound but less robust than in the first situation. The most troublesome situation would be where one approach indicates that contract farmers have significantly higher profits while the other approach indicates the exact opposite. Fortunately, we do not encounter such situations in this study.

We include the following variables in the p-score estimation: 1) the size of own land; 2) the value of production assets; 3) the value of consumption assets; 4) the age of the household head; 5) the gender of the household head; 6) the educational level of the household head; 7) the number of adult family members; 8) the female ratio in the family; 9) the distance from the farm to the market; 10) the distance from the farm to the highway; 11) a dummy variable identifying province 2; 12) a dummy variable identifying province 3; and 13) a dummy variable identifying province 4.

We use the p-score approach to conduct three comparisons. One is to compare contract farmers and never-contract farmers' performance in their entire operations (including both commercial and self-consumption operations); another is to compare contract farmers and former-contract farmers' performance in their entire operations; and the last one is to compare contract farmers and former-contract farmers' performance in their commercial operations.

A. Contract Farmers vs. Never-Contract Farmers (Entire Operations)

Table 5 [ PDF 19.8KB | 1 page ] shows the results of the p-score comparison of contract farmers and never-contract farmers' performance in their entire operations.

Since contract farmers (as the treatment group) are compared to different never-contract farmers (as the control group) under the stratification approach and the nearest-neighbor approach, the results based on the two approaches may not be consistent. As mentioned above, we use the nearest-neighbor comparisons to examine the robustness of the results from the stratification comparisons.

The ideal situation would have been to compare the commercial operations of contract and never-contract farmers. Unfortunately, as never-contract farmers have very limited areas for commercial purposes, there are only 27 never-contract farmers reporting their commercial operations (compared to 170 contract farmers), which makes the p-score comparisons highly imbalanced and uninformative. Therefore, we use the p-score approach to compare contract and never-contract farmers' performance in their entire operations only. It should be noted that since the sizes of consumption fields operated by contract farmers differ widely, the combined impacts may dilute the findings on the impact of commercialization.

• Both the stratification and nearest-neighbor comparisons indicate that contract farmers have a higher average rice price than never-contract farmers in their entire operations, but the difference is not statistically significant under either approach.
• Both the stratification and nearest-neighbor comparisons indicate that contract farmers have higher average revenue than never-contract farmers in their entire operations; and the difference is statistically significant under both approaches.
• Both the stratification and nearest-neighbor comparisons indicate that contract farmers have a higher average yield than never-contract farmers in their entire operations; the difference is significant under the nearest-neighbor comparison but not under the stratification comparison.
• Both the stratification and nearest-neighbor comparisons indicate that contract farmers have a higher average cost in terms of per hectare of rice field than nevercontract farmers in their entire operations; and the difference is statistically significant under both approaches. Both comparisons indicate that contract farmers also have a higher average cost in terms of per kg of rice production than never-contract farmers in their entire operations; and the difference is statistically significant under the stratification approach but not under the nearest-neighbor approach.
• Both the stratification and nearest-neighbor comparisons indicate that compared to never-contract farmers, contract farmers have a higher average cash cost in terms of per hectare or per kilo of rice production in their entire operations, but the difference is not statistically significant under either approach.
• Both the stratification and nearest-neighbor comparisons indicate that contract farmers have a lower average profit than never-contract farmers in their entire operations. The difference is statistically significant under the stratification approach but not under the nearest-neighbor approach.
• Both the stratification and nearest-neighbor comparisons indicate that contract farmers have a higher average cash profit than never-contract farmers in their entire operations; and the difference is statistically significant under both approaches.

B. Contract Farmers vs. Former-Contract Farmers (Commercial Operations)

Table 6 [ PDF 20.1KB | 1 page ] shows the results of the p-score comparison of contract farmers and former-contract farmers' performance in their commercial operations.

• Both the stratification and nearest-neighbor comparisons indicate that contract farmers have a higher average rice price than former-contract farmers in their commercial operations; and the difference is statistically significant under both approaches.
• Both the stratification and nearest-neighbor comparisons indicate that contract farmers have lower average revenue than former-contract farmers in their commercial operations. The difference is statistically significant under the nearestneighbor approach but not under the stratification approach.
• Both the stratification and nearest-neighbor comparisons indicate that contract farmers have a lower average yield than former-contract farmers in their commercial operations; and the difference is statistically significant under both approaches.
• Both the stratification and nearest-neighbor comparisons indicate that compared to former-contract farmers, contract farmers have a lower average cost (or cash cost) in terms of per hectare of rice field in their commercial operations, but the difference is not statistically significant under either approach. Both comparisons indicate that compared to former-contract farmers, contract farmers have a higher average cost or cash cost) in terms of per kilo of rice production in their commercial operations, but the difference is only statistically significant for the average cost under the nearest neighbor approach. The cost comparisons indicate that former-contract farmers tend to farm more intensively (i.e., higher cost per hectare of rice field); and the higher intensity tends to increase their efficiency in input use (i.e., lower cost per kilo of cost production).
• Both the stratification and nearest-neighbor comparisons indicate that contract farmers have a lower average profit than former-contract farmers in their commercial operations, but the difference is not statistically significant under either approach. Both comparisons indicate that contract farmers also have a lower average cash profit than former-contract farmers in their commercial operations; and the difference is statistically significant under the nearest neighbor approach but not under the stratification approach. According to the profit comparisons, former-contract farmers seem to be the most progressive farmers. Their experience in contract farming with AKR may have helped them become independent commercial farmers who are able to explore their own markets. Without the constraints imposed by contract farming, these farmers are able to adopt more profitable farming practices.

C. Contract Farmers vs. Former-Contract Farmers (Entire Operations)

Table 7 [ PDF 12.4KB | 1 page ] shows the p-score comparisons of contract and former-contract farmers' performance in their entire operations. The results are mostly similar to the comparisons of their commercial operations. One exception is that the stratification comparison shows that contract farmers' profit in their entire operations is significantly higher than former-contract farmers'.

Download this Discussion Paper [ PDF 167.1KB| 31 pages ].

Post a Comment

We welcome your feedback on this publication. Post a comment. ADBI is not obliged to acknowledge or publish comments and may abridge or edit them before web posting.

Comment(s)

There are [1] comment(s) for this entry. Post a comment.

1. Anthony M. Zola
   (posted 02 July 2008 / 12:54:22 AM)
   
   This is an excellent contribution to the current debate about contract farming in mainland Southeast Asia.
   
   I have conducted research for the ADB, much less sophisticated than this study, and had similar results. Smallholder farmers were better off if they were engaged in contract farming than when they sold daily labor to local concessions / nucleus estates.
   
   My congratulations to the research team. It is not an easy topic on which to conduct research.
   
   Anthony Zola
   Bangkok, Thailand
   & Vientiane, Lao PDR