Switching Regression

While the p-score comparisons in the above try to compare the performance of contract and non-contract farmers with similar intrinsic characteristics, they cannot correct hidden bias because p-score comparison only controls for observed variables (to the extent that they are perfectly measured). For example, farmers' motivation may be an unobserved covariate affecting both farmers' performance and their choices about joining the contract.

Selection models can be used to address unobservable selection bias in deciding whether to join the contract or not. In this section we use an endogenous switching regression model to account for selection biases. We use the model to examine how farmers' characteristics affect their decisions to join the contract and their performance with or without the contract. We also compare farmers' expected performance under the contract and without the contract.

A. Methodology

Consider the following model that describes farmers' choices about joining the contract and their performance with and without the contract:

If γZ_i + u_i > 0, farmer i chooses to join the contract, which is described by I_i = 1

If γZ_i + u_i ≤ 0, farmer i chooses not to join the contract, which is described I_i = 0

Farmer i's profitability with the contract (I_i = 1) is y_1i = β₁ X_1i + ε_1i;

Farmer i's profitability without the contract (I_i = 0) is y_0i = β₀ X_0i + ε_0i;

In the model, Z_iis a vector of farm characteristics that affect farmers’ decisions to join the contract; X_1i and X_0i are two vectors of farm characteristics that affect farmers’ performance under the contract and without the contract; and y_1i and y_0i are dependent variables measuring farmers’ profitability. γ, β₁ and β₀ are vectors of parameters subject to estimation. u_i, ε_1i, and ε_0i are three random error terms that follow trivariate normal distribution.

After the parameters are estimated, we can calculate

xb_1i = E (y_1i | x_1i) = x_1iβ_l

xb_0i = E (y_0i | x_0i) = x_0iβ₀

yc_{1_1i} = E (y_1i | I_i = 1, x_1i) = x_1iβ₁ + σ₁ρ₁ ∫ (γZ_i) / F(γZ_i)

yc_{0_1i} = E (y_0i | I_i = 1, x_1i) = x_1iβ₀ + σ₀ρ₀ ∫ (γZ_i) / F(γZ_i)

yc_{0_0i} = E (y_0i | I_i = 0, x_0i) = x_0iβ₀ - σ₀ρ₀ ∫ (γZ_i) / [1 - F(γZ_i)]

yc_{1_0i} = E (y_1i | I_i = 0, x_0i) = x_0iβ₁ - σ₁ρ₁ ∫ (γZ_i) / [1 - F(γZ_i)]

xb_1i represents the unconditional expectation of farmers’ performance under the contract; xb_0i represents the unconditional expectation of farmers’ performance without the contract; yc_{1_1i} represents the conditional expectation of contract farmers’ performance under the contract; yc _{0_1i} represents the conditional expectation of contract farmers’ performance without the contract; yc_{0_0i} represents the conditional expectation of non-contract farmers’ performance without the contract; and yc_{1_0i} represents the conditional expectation of noncontract farmers’ performance with the contract. σ₁ and σ₀ are the standard errors of å1i, and ε_0i; ρ₁ is the correlation coefficient between ε_1i; and u_i; and ρ₀is the correlation coefficient between ε_0i and u_i.

B. Indicators for Premiums of Joining the Contract

Based on equations (1) to (6), three indicators can be constructed to compare farmers’ profitability with and without the contract.

(1) Π = xb_1i − xb_0i

According to equations (1) and (2), Π is equal to a general farmer i’s (irrespective of his/her choice of contract farming) expected performance under the contract minus his/her expected performance without the contract. The mean of Π measures farmers’ average profitability premiums from joining the contract.

(2) Π₁ = yc_{1_1i} − yc_{0_1i}

According to equations (3) and (4), Π₁ is equal to a sample contract farmer i’s expected performance under the contract minus his/her expected performance without the contract. The mean of Π₁ measures the sample contract farmers’ average profitability premiums from joining the contract.

(3) Π₀ = yc_{1_0i} − yc_{0_0i}

According to equations (5) and (6), Π₀ is equal to a sample non-contract farmer i’s
expected profitability under the contract minus his/her expected profitability without the
contract. The mean of Π₀ measures the sample non-contract farmers’ average profitability
premiums from joining the contract.

C. Indicators for Farmers’ Relative Performance With and Without the Contract

(4) Λ_{1_1} = yc_{1_1i} − xb_1i and Λ_{0_0} = yc_{0_1i} − xb_0i

According to equations (1) and (3), Λ_{1_1} compares a sample contract farmer i’s average profitability under the contract (measured by i yc_{1_1i} ) to the profitability of a general farmer (with the same characteristics) under the contract. A positive mean of Λ_{1_1} indicates that under the contract, farmers who actually joined the contract tend to have higher profitability than those who did not.

According to equations (2) and (4), Λ_{0_1} compares a sample contract farmer i’s average performance without the contract (measured by yc_{0_1i} ) to the profitability of a general farmer without the contract. A positive mean of Λ_{0_1} indicates that outside the contract, farmers who actually joined the contract would also have a higher profitability than those who did not.

(5) Λ_{0_0} = yc_{0_0i} − xb_0i and Λ_{1_0} = yc_{1_0i} − xb_1i

According to equations (2) and (5), Λ_{0_0} compares a sample non-contract farmer i’s average profitability outside the contract (measured by yc_{0_0i} ) to the profitability of a general farmer (with the same characteristics) outside the contract. A positive mean of Λ_{0_0} indicates that outside the contract, farmers who did not join the contract tend to have higher profitability than those who did.

According to equations (1) and (6), Λ_{1_0} compares a sample non-contract farmer i’s average performance outside the contract (measured by yc_{1_0i} ) to the profitability of a general farmer outside the contract. A positive mean of Λ_{1_0} indicates that under the contract, farmers who did not join the contract tend to have higher profitability than those who did.

Λ_{1_1} , Λ_{0_1} , Λ_{0_0} , and Λ_{1_0} measure farmers selection bias on contract farming. There are four patterns.

(1) Λ_{1_1} > 0 ; Λ_{1_0} < 0 and Λ_{0_1} > 0; Λ_{0_0} < 0

This situation indicates that the sampled contract farmers tend to have higher profitability no matter whether they are under the contract or outside the contract. That is, better farmers tend to choose to join the contract.

(2) Λ_{1_1} > 0 ; Λ_{1_0} < 0 and Λ_{0_1} < 0; Λ_{0_0} > 0

This situation indicates that the sampled contract farmers tend to have higher profitability under the contract but lower profitability outside the contract. That is, farmers who have a comparative advantage in contract farming tend to choose to join the contract, while those who have a comparative advantage outside the contract tend to choose to stay outside the contract.

(3) Λ_{1_1} < 0 ; Λ_{1_0} > 0 and Λ_{0_1} > 0; Λ_{0_0} < 0

This situation indicates that the sampled contract farmers tend to have lower profitability under the contract but higher profitability outside the contract. This is an unlikely scenario because it implies that farmers who do not have a comparative advantage in contract farming tend to choose to join the contract, while those who do have a comparative advantage in contract farming nevertheless tend to choose to stay outside the contract.

(4) Λ_{1_1} < 0 ; Λ_{1_0} > 0 and Λ_{0_1} < 0; Λ_{0_0} > 0

This situation is the exact opposite of the first one. It indicates that the sampled contract farmers tend to have lower profitability whether they are under the contract or outside the contract. That is, better farmers tend to choose to stay outside the contract.

D. Comparison of Contract Farmers’ and Non-Contract Farmers’ Profitability in Commercial Rice Farming

Based on the above switching regression model, we use the “movestay” module (Lokshin and Sajaia, 2004) in the STATA program to evaluate factors that affect farmers’ decisions to join the contract and their performance with or without the contract. We measure farmers’ performance according to their profits per hectare in their commercial operations.

In the selection model we include the following variables:

• The rice price and input prices (i.e., seed, wage, chemical fertilizer, compost, irrigation, and machinery) under contract and without contract. For contract (or noncontract) farmers, the prices without contract (or under contract) are unobservable. We estimate such counterfactual prices by using farmers’ geographical locations and their land sizes as two regressors.
• Household characteristics including the age, gender, and education level of the household head, family size, and the ratio of females in the household.
• Farm characteristics, including the size of own land, the value of production assets, the value of consumption assets (e.g. TV), the distance from the farm to the market, the distance from the farm to the highway, the total number of plows and pumps, and the number of motorbikes.
• Three province dummies to identify farmers from four different provinces

In the profit functions, we include the rice price, the input prices, the size of own land, the value of production assets, and the three province dummies. For the non-contract profit function, we also include a dummy to differentiate former-contract and never-contract farmers.

Table 8 [ PDF 21KB | 1 page ] shows the estimation results for the selection function, which suggest the following:

• Households with less asset value are more likely to join the contract.
• Households with younger household heads are more likely to join the contract.
• Households with more educated household heads are more likely to join the contract.
• Households with larger family size are more likely to join the contract.
• Households closer to the highway are more likely to join the contract.

Table 9 [ PDF 22.9KB | 1 page ] shows the estimation results for the profit functions with and without contract; based on which we can estimate the sample farmers’ profits under contract and outside contract. With the estimated results we can then calculate contract and non-contract farmer’s premiums from joining the contract and compare their profitability under contract and outside contract. The results are summarized in Table 10 [ PDF 22.9KB | 1 page ].

• For all the sample farmers on average, joining the contract would tend to raise profit by 0.43 million riel.
• For the sample contract farmers, joining the contract would raise their average profit by nearly one million riel.
• For the sample former-contract farmers on average, had they joined the contract, their profits would have been 0.18 million riel lower than their actual profits.
• For the sample never-contract farmers on average, had they joined the contract, their profits would have been increased by 0.17 million riel, but the difference is not statistically significant. Note that the small sample size (27 never-contract farmers only) may be a factor affecting the significance level.
• Under contract, the sample contract farmers on average have higher profits than the sample former-contract and never-contract farmers; their average profit under contract is 0.03 million riel above the average of all the sample farmers.
• Outside contract, the sample former-contract farmers on average have higher profits than the sample contract and never-contract farmers; their average profit outside the contract is 0.51 million riel above the average of all the sample farmers.

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

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