On reuse of clusters in repeated studies - ARCHIVED

Suppose data for a survey with multi-stage design is to be collected in two periods of time. This paper assesses the relative merits of keeping the same clusters in the sample vs. sampling new clusters, under different statistical (correlation between clusters and over time) and logistical (costs of survey) scenarios. The design effect of re-using the same clusters from the master sample over time is of the form "1 - Ap(pi)/n" where "p" is intertemporal correlation of the cluster totals, "n" is the number of clusters, "pi" is the proportion of clusters retained from the previous round, and "A>0" is a fixed constant. As long as the efficiency gains appear to be minor, the value of the designs that reuse the clusters comes from the logistical (cost of the survey) considerations. Empirical demonstration that uses Demographic and Health Survey (DHS) data for Bangladesh, 1996 and 2000, is provided.