Considering the Effects of Nonresponse on Correlations Between Surveyed Variables

Abstract

Nonresponse to organizational surveys is a vexing problem. Although scholars have a reasonable understanding of how systematic nonresponse influences estimates of population means, they are likely to have less context for understanding how it impacts correlations. We present the results of a simulation study designed to provide a frame of reference for understanding the extent to which systematic nonresponse can bias observed correlations between surveyed variables. Although there were cases where notable levels of bias were observed in the correlations, the majority of situations simulated resulted in mean observed correlations that were within .05 of the true correlation. The variety of situations simulated allows researchers to evaluate the external validity of observed correlations in their data – a critical goal for survey researchers in practice and academics.