Design Weighted Quadratic Inference Function Estimators of Superpopulation Parameters

Abstract

Using information from multiple surveys to produce better pooled estimators is an active research area in recent days. Multiple surveys from same target population is common in many socioeconomic and health surveys. Often all the surveys do not contain same set of variables. Here we consider a standard situation where responses are known for all the samples from multiple surveys but the same set of covariates (or auxiliary variables) is not observed in all the samples. Moreover, in our case we consider a finite population set up where samples are drawn from multiple finite populations using same or different probability sampling designs. Here the problem is to estimate the parameters (or superpopulation parameters) of underlying regression model. We propose quadratic inference function estimator by combining information related to the underlying model from different samples through design weighted estimating functions (or score functions). We did a small simulation study for comprehensive understanding of our approach.