17
To address this concern, we use data from nonparticipating households to identify variations in
the outcome variables of interest (e.g., off-farm labor-market participation) that are due to factors other
than the Grain for Green program. The data from both participating and nonparticipating households are
used in a difference-in-differences (DID) estimator that analyzes these types of program effects with
these types of data. In fact, DID has been used extensively in the labor economics literature to assess the
employment effects of a number of different government policies, including the impact on employment
of a raise in the minimum wage (Card and Krueger, 1994) and the effects of temporary disability
benefits on the duration of time off from work after an injury (Meyer, et al., 1995 ).
In short, DID compares outcomes from a policy change on two groups—those affected by the
policy change (program participants) versus those who are not (non-participants of the program -- Meyer,
1995). Formally, DID can be shown by letting t and t′ denote time periods after and before the
program, respectively. The DID estimate is given by
DID = [E(Yt | D = 1) - E(Yt' | D = 1)]-[E(Yt | D = 0) - E(Yt' | D = 0)].
The idea is to correct the simple difference between an outcome before the policy change and after for
the treatment group by comparing the before-after change of treated units with the before-after change
of control units. By doing so, any common trends that show up in the outcomes of the control units and
of the treated units are differenced out (Smith, 2004). The estimator also can eliminate recall bias
inherent in a retrospective survey to the extent that the bias is the same for the two groups.
Use of the DID estimator, however, depends on several key assumptions. The conventional DID
estimator requires that, in the absence of the program, average outcomes for participants and