Bootstrapping the Standard Error of the Mediated Effect

Mediation analysis seeks to go beyond the question of equations, using the notation from MacKinnon, Warsi, whether an independent variable causes a change in a and Dwyer (1995). dependent variable. Mediation addresses the question of how that change occurs. When a third variable is thought to be intermediate in the relationship between two variables, it is called a mediator. In order to test the mediated effect for significance, or to create confidence limits for the effect, the standard error of the effect is needed. Estimates of the standard error can fluctuate widely when sample size is small or when the variables are not normally distributed. Bootstrapping is a method that resamples from an original sample to derive a more accurate estimate than is found through traditional methods. This paper describes a SAS program that estimates the mediated effect of a sample, takes bootstrap samples, and calculates the standard error and confidence limits of the mediated effect. The audience should have an understanding of multiple regression analysis. This program was written with base SAS software and operates in a regular PC environment.