10
much of a farm is part of the program and how long the land has been retired. Hereafter, we use these
two variables as measures of the intensity of program participation and as tools for identifying the
effects of the program.
Combining the 2003 and 2005 surveys provides information on nearly all of the same variables
for both before (in 1999) and after (in 2002 and 2004) implementation of Grain for Green. Enumerators
collected information on each household’s production activities on a plot-by-plot basis. The survey also
collected detailed information on each household’s total asset holdings, its demographic make-up and
other income-earning activities involving both on-farm and off-farm activities.
The study relies on information for 1999 that was collected in 2003, and we acknowledge the
potential for problems inherent in recall data, especially regarding the preprogram period. Long-term
recall data are potentially inaccurate, although this issue continues to be debated in the literature.
Unfortunately, the Chinese government’s quick decision to implement Grain for Green and lack of
transparency in the details of its implementation precluded interviews with potential participants at the
program’s onset. We addressed concerns about recall bias through the design of the survey and careful
training and monitoring of the enumerators to ensure that respondents gave their best recollection of past
amounts and activities. We also endeavor to deal with the recall bias by reestimating all of the analyses
using a sample of only 67 households—the 27 households that switched from nonparticipant to
participant status between the two surveys and the 40 nonparticipating households. With this subsample,
we can compare the changes in off-farm labor between 2002 and 2005 to avoid having to rely on the