Regression Analysis Using Survey Data

There is an increasing tendency to perform regression analyses using survey data. Often such data is the product of a complex sample design reflecting the need to collect data as efficiently as possible within cost constraints. In social surveys it is comparatively unusual to find surveys which cover an area of any appreciable size which do not use cluster or multistage sampling. The observed effect when estimating population means and totals is that such clusters exhibit a positive intracluster correlation (units within a cluster are more alike) and the overall effect is to increase the variance of the estimator compared to the variance that would have been aclhieved under simple random sampling. Kish and Frankel (1974) extend empirical results of this kind to the estimation of more complex statistics such as differences between domain means, correlation coefficients and regression coefficients. This paper is concerned with regression analysis.