The 2003 household survey used a stratified sampling strategy designed to collect data on a
random sample of 359 households in the program area. From the three provinces (Sichuan, Shaanxi and
Gansu) that had been participating in Grain for Green since 2000, we selected two counties from each
province and then randomly selected three townships from each county. In each township, we randomly
selected two participating villages and randomly chose ten households from each village. The data
includes information on at least one program-participating household for each village. Two of the 36
villages had only participating households. The survey in 2003 collected information on 2002 and 1999.
The survey in 2005 was nearly identical to the earlier wave and included 348 households. Of the 359
households surveyed in 2003, we were able to track 270 of them in 2005, 230 of which were
participating households. Of the 230 households, 27 had entered the program since 2003. The attrition
rate (from the survey) was 24 percent for households participating in the program and 32 percent for
nonparticipating households. The households not included in the 2005 survey were not systematically
different from households that were included in both surveys and were dropped from further analysis.10
Among the program participants, there is variability in the number of years that they
participated; the extent of their participation (in terms of absolute cultivated area and share of the
household’s cultivated area) varied widely across the sample (Figures 1 and 2). A third of the households
in the sample started to participate during the initial year of the program (Figure 1). The share of land
that each household retired from cultivation also varied among participating households and ranged
from less than 5 percent of total cultivated land holdings to 100 percent (Figure 2). When considering
program impacts, it is reasonable to expect that the effect of the program will vary depending on how