The role of the partical correlation coefficient in statistical inference

Abstract

The paper analyzes situations when the calculation of partial
correlation coefficients becomes justified and can lead to conclusions that do not
coincide with the initial ones at the stage of calculating the usual Pearson
correlation coefficients. Partial correlations allow identifying significant factors in
conditions of their collinearity, as well as selecting predictors to build the best
regression model. Particular attention is paid to graphical interpretation and issues

of checking the significance of the partial correlation coefficient in conditions of
non-Gaussian samples. Examples of both model and field data are shown where the
use of partial correlations allowed making informed decisions about cause-andeffect relationships.