Currently I am working on a project wherein I need to predict the estimated new Selling price of IPL players based on their current base price and previous years' performance & sold price. I'm planning to use Step-wise Regression but since I have around 20 variable I'm planning to do PCA. What would be the best way to deal with outliers? Should I just eliminate them or there is any better way?
There is a lot written about general advice what to do with "outliers" in RG. Just search, and you will find.
If you plan stepwise regression, I think it does not matter if you just throw away any part of your data (e.g. outliers), as the result will be drastically biased by random patterns in your observed data set.
Check whether outliers are meaningful for your analysis. The best way to deal with outliers is not having any! However, you can fit a robust regression MM-estimator model to deal with outliers. You can use the lmrob function from the robustbase package in R to fit the robust regression model. There are some other ways too. One such way is the quantile regression.
There is a lot written about general advice what to do with "outliers" in RG. Just search, and you will find.
If you plan stepwise regression, I think it does not matter if you just throw away any part of your data (e.g. outliers), as the result will be drastically biased by random patterns in your observed data set.
Nidhi Rana, as suggested above by Jochen W., there is a lot of general advice around concerning how to deal with outliers. For what it's worth, I have found that winsorising them might be valid. Employing that method still permits the outliers to be retained and to have some "meaning", but they will not distort the analyses undesirably.
If people disagree with me, I'd be happy for them to come back and do so. I am aware, however, of more sophisticated methods.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Regression