Limiting Effects of Outliers

[Matt Von]

I am trying to use OLS regression. I have tested my model for normality and it meets the OLS assumption except the outliers. I am wondering what's the best way to fix this without threating the validity of my results. I have researched various methods but cannot come to a conclusion of which is best. Furthermore, I have read it is possible to limit the effect of outliers through the use of robust regression though I not convinced this is the case. The data I am working with is Time Series.

[Nick Cox]

Matt Von/Mon/Linsky: You seem to have several personas on the internet, perhaps more than those. On Statalist we ask that you use a full real name, typically the given name you use most and a full family name.

As you have time series, the validity of various kinds of inference, including regression, will depend on what assumptions you make about serial correlation and whether they are satisfied. Using an outlier-robust or -resistant flavour of regression. will not fix that directly.

Robust regression in Stata is supported in various ways. Although some would deny that it is robust regression, quantile regression default is often worth checking out. That is the qreg command.

The bad and the good news is that there are many implementations of some kind of robust regression in Stata.

 
search verardi, author 
search lindsey, author

is I think the best way into finding out about implementations. I'd start with the paper by Verardi and Croux. I don't have advice on which flavour of robust regression is best, as I am no expert and the experts either disagree or deny that is an answerable question.

Don't use rreg (personal advice; Google "Statalist rreg" for discussion if you need).

[Richard Williams]

As luck would have it, I told somebody today not to use rreg. The evils of rreg are discussed in Appendix B of

http://www3.nd.edu/~rwilliam/stats2/l24.pdf

Mostly it consists of quotes from Nick Cox! The rest of the handout discusses some other alternatives for detecting and dealing with outliers.

[Nick Cox]

Richard's fine document contains a comment from me about the Antarctic ozone hole, which I would like to withdraw given https://www.math.uni-augsburg.de/htd...heim/1990c.pdf

[Matt Von]

Thank You both for your reply. My issues is resolved though slightly confused about the first post.

[Richard Williams]

Richard's fine document contains a comment from me about the Antarctic ozone hole, which I would like to withdraw given https://www.math.uni-augsburg.de/htd...heim/1990c.pdf

Done! You may need to refresh your browser to see the revisionist history version. Thanks.