Variations of the Greenberg unrelated question binary model

Abstract

We explore different variations of the Greenberg Unrelated Question RRT model for a binary response (yes or no). In one variation, we allow m independent responses from each respondent. In another variation, we use inverse sampling where we record the number of responses leading up to the kth yes response. It turns out that for m > 1, the variance (theoretical and empirical) of the multiple independent response model decreases significantly relative to the regular Greenberg et al. (1969) model (m = 1). Furthermore, it was also noticed that for k > 1, the variance (theoretical and empirical) of the inverse sampling model decreases significantly as well relative to the inverse sampling model for k = 1. Thus, it turns out that both variations produce more efficient models. These results have been validated by theoretical comparisons, extensive computer simulations, and a field survey.