You are here

Exploring the Goodness of Fit in Linear Models - What Should I Do with the Outlier?

Author(s): 
Scott A. Sinex

All too often, the statement “it’s a lousy datum point - let’s just throw it out” is made in reference to an unusual datum value. This is not the proper way to handle a potential outlier. If you have recently-collected experimental data, you can repeat the measurement or experiment to try to determine if the outlier was a fluke. Is the outlier a real anomalous datum point? Some outliers are simply data recording errors. But, for example, with annual stream discharge (volume/sec), you can have a very wet or dry year that stands out from more typical or average years. It just pays to be cautious with outliers and use methods that deal with them, such as running averages. See Motulsky (2002) for more information on dealing with outliers, including the Grubbs' test for detecting outliers on replicates of measurement.

Scott A. Sinex, "Exploring the Goodness of Fit in Linear Models - What Should I Do with the Outlier?," Convergence (March 2005)