Comparing models only makes sense when both models 'see' exactly the same set of data. That makes it tricky to combine outlier elimination with model comparison. Prism handles this situation automatically.

If you select automatic outlier elimination (on the Method tab), Prism first fits each model separately to determine which points (if any) are outliers. Then it removes all the points that are defined as outliers from either of the models, and fits both models again. It uses the sum-of-squares and df from these fits to perform the comparison. This approach ensures that exactly the same points are fit to both models, and that outliers from either model are excluded.