Inaccuracy of regression results in replacing bivariate correlations

Abstract

This manuscript considers discrepancies between the bivariate correlation and several indices of association estimated from regression results. These indices can be estimated from results typically reported in primary studies. In recent years, many researchers conducting meta-analyses have used these indices in place of, or together with, the bivariate correlation. I illustrate the differences among these indices and the bivariate correlation. I demonstrate the inaccuracy of these indices as replacements for bivariate effects. Thus, I recommend discontinuing the use of these indices and partial effect sizes as replacement for the bivariate correlation.

Keywords: meta‐analysis; partial correlation; partial effect sizes; regression results; semipartial correlation.