22
individual data, participation in the program decreases the likelihood of an individual working on-farm
by 13 percent, although the point estimate is not significant (Table 6, column 1).
For those that expect that Grain for Green will help to promote off-farm employment, the
results of the basic regression are somewhat encouraging. The signs of the basic DID estimates suggest
that Grain for Green is promoting structural change, although the low t-ratios on some of the estimates
suggest weak confidence in the results. In addition, the nature of the results differs for estimates that use
household-level data and those that use individual-level data.
Effect of Program Intensity
While the positive results from the program-participation models are relatively weak, the results
for estimates of the effect of program intensity are somewhat stronger. To exploit the variation in
treatment intensity across households, the DID strategy can be generalized. Consider the difference
between average off-farm labor participation for Grain for Green participants versus nonparticipants. If
devoting more land to the program led to an increase in available labor time or an increase in liquidity
that households could use to find off-farm jobs, the difference in off-farm labor could be positively
related to the area of land retired by each household. This suggests the following regression:
L0 Gj γ) = μ ÷ fi I time + ∙5D (:. 1) + αP(i, t) + plc (:. O) + s(ij, i)
where P(i, t) denotes the intensity of the program for observation i in year t. P(i, t) is zero for all
observations in year 1999 and positive only for participants in year 2004. As before, all specifications
control for the interaction term for the Yangtze River basin dummy variable times the year 2004 dummy
variable and for household size and total land holdings. In the model, we include (1) the ratio of program