|
Abstract:
|
Prior theoretical work on survey nonresponse adjustment has shown that bias and variance reduction of a respondent mean requires high correlation between an adjustment variable, survey participation, and the variables of interest (Y) (Little and Vartivarian 2005). In practice, multiple auxiliary variables are used whose association with participation and Y varies in strength and direction and may be correlated with each other. In this paper, we will use simulation to examine how multiple influences on response propensity and Y affect nonresponse bias of the unadjusted respondent mean of Y, the correlation between estimated propensity and Y, and bias and RMSE of adjusted means. The simulations also examine the efficacy of nonresponse adjustment when one auxiliary variable is omitted from the response propensity model. The results will be compared to findings from existing surveys.
|
