Responsible Partialling

Abstract

Partialling correlations (partial, semipartial or bipartial) is built upon regression equation(s). It is imperative that the assumptions of those models be considered. Those assumptions rest on the OLS residuals of those models (e.g., linearity, homoscedasticity, normality, independence, absence from data with extreme influence). Partial correlations of all types depend on residuals, but consideration of fit of data in respect to those residuals is seldom done. Although this can be accomplished with repeated subsequent analyses, partialling in commercial software does not directly render these assumption diagnostics. As these are mandatory for use of least squares, examples are produced and an Excel program that automatically performs an exhaustive set of diagnostic analyses and plots for partialling is offered.