Assessing Multicollinearity via Identification of High Leverage Points in Financial Accounting Data

Inaccurate and invalid statistical inferences in regression analysis may be caused by multicollinearity due to the presence of high leverage points (HLP) in a data set. Therefore, it is important that high leverage point which is a form of outlier be detected because its existence can lead to misfitting of a regression model, thus resulting in inaccuracy of regression results. In this paper, several methods have been proposed to identify HLP in a financial accounting data set prior to conducting further analysis of regression and other multivariate analysis. The Pearson’s correlation coefficient and variance inflation factors (VIF) were used to measure the success of a detection method. Numerical analysis showed that common diagnostics like the twice-mean and thrice-mean rules failed to detect HLP in the given data set whilst robust approaches such as the potentials and diagnostic- robust generalized potentials (DRGP) methods were found to be successful in identifying high leverage point as indicated by lower values of the Pearson’s correlation coefficient and variance inflation factors.