Treatment of Influential Observations and Outliers in Regression Analysis
Share
In this article, Dr. Narsid Golic uses a case study to illustrate that the visual inspection of data plots alone should not be used as a substitute to implementing proper statistical testing for detecting and removing outliers from the data sample when estimating a regression model. Although regression analysis has found widespread acceptance throughout the judicial system, the results of regression analysis have not gone unchallenged by opposing experts. The opposing expert may argue that the results of regression analysis are unreliable if they are based on sample data that includes obvious outliers. He argues that to support an opinion for exclusion of an outlier from the sample used for regression estimation one is required to implement proper statistical testing commonly used to detect outliers in data; and make a statistical distinction between an outlier (an outlier in a dependent variable) and an influential observation (an outlier in an independent variable).