Here are a few ways to deal with the outliers:
- Remove them, or set them to NA. This can make sense if you do not have too many of them, and it will make your model easier to explain. The trade-off is that you may be removing important data, so use with caution.
- Use a transformation to reduce the variability. Choose the appropriate transformation based upon the skewness of the data, or try a Box-Cox Power Transformation. One advantage to this is that the correct transformation can reduce the influence of the extreme observations.
- Bring them down to a pre-controlled level. This can be accomplished by using a trimmed or Windsorized mean. This is done in the actuarial profession, since some risk can be capped if the ...