Estimation of Small Area Total with Randomized Data

Abstract

In social surveys involving questions that are sensitive or personal in nature, respondents may not provide correct answers to certain questions asked by the interviewer. The impact of this nonresponse or inaccurate response becomes even more acute in the case of small area estimation (SAE) where we already have the problem of small sample size coming from the small area. To obtain a truthful response, we use randomized response techniques in each small area. We assume that a non-sensitive auxiliary variable, highly correlated with the study variable, is available. We use the word model in two senses — one in the context of population models, i.e. the relationship between the study variable and the auxiliary variable; and second, the scrambled response model. We focus on the problem of estimating small area total and examine its performance both theoretically and numerically.