Outliers Linear regressions can be greatly influenced by outliers—atypical cases. Outliers can pull the equation away from the general pattern, and unduly sway the regression output. But what is the appropriate way to deal with an outlier? In some situations it is reasonable to simply delete an outlier from the analysis. For example, perhaps the outlier was a mistake – a data entry or data recording error. Another example is when there are special circumstances surrounding a specific case. For example, Nevada has a very high divorce rate. Because of its laws, it attracts people from outside the state for “drive-by divorces Because Nevada’s divorce rate does not accurately measure the rate of divorce for Nevada residents, as do the divorce rates in other states, it maybe reasonable to completely remove Nevada from any analysis of divorce rates. In other situations, outliers should remain in the analyses, as an outlier maybe the most important observation. In these circumstances, it is a good idea to run the analysis twice, first with the outlier in the regression and second with it excluded. If the outlier is not exerting an undue influence on the outcomes, both models should reasonably coincide. If, however, the results are vastly different, it is best to include both results in the text of the report.
