What explanatory variables to exclude? I am conducting research on cross-sectional data of ebay auctions and want to determine the effect of reputation on price. ebay offers several measures of reputation: a users "feedback score" (calculated as the number of positive reviews received less negative reviews), the "feedback percentage" (positives over the sum of positives and negatives), the number of positives/neutrals/nagatives recieve in the  last 12 months as well as detailed seller reviews on specific categories.  I have concerns about overfitting and especially multicollinearinty between these measures (for example between the number of positives and "feedback score") as they share underlying components. How do i determine which to include? Thank you
ps. could i include both no. of positives and no. of negatives as independent variables to examine the difference in effect between them?
 A: 
ebay offers several measures of reputation: a users "feedback score" (calculated as the number of positive reviews received less negative reviews), the "feedback percentage" (positives over the sum of positives and negatives), the number of positives/neutrals/negatives received [sic] in the last 12 months as well as detailed seller reviews on specific categories.

So there are some obvious easy solutions to your problem.  First, a statistic is any function of the data.  The feedback score is a function and not data, so you can drop this immediately.  The same is true for feedback percentage.  The only thing that matters is the raw data.
That said, the very fact they provide both scores could imply that those functional forms are mathematically meaningful.  The first is linear and is easily tested.  The second is a ratio which could easily be converted to a linear form by taking the logarithm.
You can test multiple forms through model selection procedures to determine which model is closest to the data generating function.
