Irrigation and Water Resource Management 2011-2012

To implement DID with ex-ante matching, households should be matched before the survey. This requires a detailed enumeration in the treatment and comparison areas to collect a set of variables that can match treatment and comparison households. In the spring of 2012, the Agence Nationale de la Statistique et de la Démographie (ANSD) conducted extensive enumeration in the Delta area, including the Saint Louis and Dagana departments, for a total of about 11,600 households. IMPAQ used the enumeration files to sample and match households. The target sample sizes for the Delta were 1,306 treatment households and 1,306 matched comparison households. To achieve this sample size, IMPAQ provided slightly larger samples to ANSD to allow for some non-response because the survey effort may not have reached 100 percent of the sample. Specifically, we selected 1,637 treatment and 1,637 comparison households (about 25 percent more than the proposed sample sizes).

From the enumeration file, we randomly sampled 1,637 treatment households. We then matched each treatment household with a comparison household identified as the most similar in relevant pre-treatment characteristics.

In the spring of 2012, ANSD completed an extensive enumeration in the Podor area. Specifically, 1,617 households were enumerated in the treatment area and 585 were enumerated in the comparison area. For the impact evaluation, it is important that the treatment group include households that will actually get the treatment (irrigation and land). However, at the time of sampling, we could not find a clear way to identify which households would receive irrigated land in Podor. Because the enumeration data includes 1,617 households in the treatment area and we needed to identify the 440 households that would receive land, a random sample would not ensure that we select enough households that actually receive land.

Given the urgency of selecting the samples and proceeding with the survey, MCA-S and IMPAQ agreed to survey all households (1,617) in the enumerated treatment area. This approach ensured that our survey sample would capture the households that would receive land (treatment group). In addition, we agreed to survey a random sample of 440 households in the Podor comparison area (out of a total of 585 households in the enumeration). Moving forward quickly was very important to avoid wasting time and resources. Furthermore, waiting until spring 2014 (when land decisions are expected to be finalized) could have jeopardized our ability to have useful baseline data because the intervention may take place before the baseline data collection.