The distinction between eliminating and identifying outliers

When Prism 5 fits a curve with nonlinear regression, it can identify outliers using the new ROUT method. There are two ways to choose this method in Prism 5.

• Check the "Automatic outlier elimination" option on the the first (Fit) tab of the nonlinear regression dialog. Prism first identifies any outliers, and then reruns the fit with those outliers excluded. The main table of results includes the number of outliers excluded, and these values are tabulated on a separate table of outliers.

• Check the "Count the outliers" option on the last (Diagnostics) tab of the nonlinear regression dialog. Prism identifies and tabulates the outliers, but does not eliminate them. The nonlinear regression results are still influenced by the outliers. You can then go back and look at the data and experimental notes, and decide what to do.
  

In both cases, the aggressiveness of the outlier hunt is determined by the ROUT coefficient Q, which you enter in the Weights tab of nonlinear regression. We recommend leaving Q set to its default value of 1%. If you increase the value, Prism will find more outliers, but will also mistakenly identify more 'good' points as outliers. If you lower the value, Prism will find fewer outliers.

The first step of the ROUT method is to fit the curve using a robust method of nonlinear regression, little affected by outliers. This is then used as a baseline from which to identify outliers. You can choose to report the results of the robust fit  by checking an option on the Fit tab of the nonlinear regression dialog. The advantage of the robust method is that there is no strict threshold between outliers and 'good' points. As values get further from the curve, they have less and less impact on the fit. But this method has disadvantages. Robust nonlinear regression does not generate standard errors or confidence intervals of the best-fit values, and cannot be used to compare two models.