Historical background: Mediation

Mediation refers to the relationships among three variables: an independent variable, a potential mediating variable, and a dependent variable. The question raised in this case is whether the mediating variable accounts for a significant amount of the shared variance between the IV and DV. Researchers have long been interested in understanding the ways in which variables may be associated so various statistical methods have been generated in which one can study this question: raw (zero order) correlations; part- and partial-correlations; multiple regressions; and so forth.

Mediation is a special case of the partial correlation technique in which three variables are examined in a particular fashion. Some of the earliest attempts to explore mediation dates back to early path models (Duncan, 1975; Kenny, 1979), and much has been done with path analysis since. In fact, most researchers who wish to examine relationships among a group of variables (i.e., more than three) typically opt for path analysis in structural equation modelling (SEM). However, there are still many researchers who are interested in the three-variable case.

Again, as with moderation, current statistics programmes do a poor job of allowing the researcher to specifically examine his or her data for mediation. Sobel (1988) has described a useful statistical test that allows one to determine whether significant mediation has occurred, but few statistics programmes include this computation. A web-site created by Preacher and Leonardelli (http://quantpsy.org/sobel/sobel.htm) is used by many researchers to compute the Sobel's test, but it does not provide other basic information deemed important by researchers.

To address this concern, I created MedGraph in 2003. This programme takes statistical output generated from a correlation matrix and two regressions, performs the Sobel test, computes a 95% confidence interval, and generates additional useful information such as effect sizes.