View Single Post
11-19-2012, 02:29 PM
Czech Your Math
Registered User
Czech Your Math's Avatar
Join Date: Jan 2006
Location: bohemia
Country: Czech_ Republic
Posts: 4,845
vCash: 500
Originally Posted by barneyg View Post
The main reason it may be important is that if the correlation between 2 regressors is very high, the coefficient on those 2 variables will be very sensitive to the use of the other variables included in the regression. In English: when 2 variables are closely related, the associations suggested by OLS could be strange or at least misleading.
I can see why that could be important. However, many of the high cross-correlations seem to have been in large part due to contemporaneous changes in the trends of the variables which don't have much logical basis. IOW, in many cases I would have expected much less correlation or even correlation in the opposite direction.

Originally Posted by barneyg View Post
I don't have a good sense of which one is better (raw differences vs % differences). I think you want a Y that can easily be interpreted, do you want "change in adjusted scoring for the top players" or "% change in adjusted scoring for the top players". Seems to me like raw differences might be better here.
I think either one (raw or %) could be interpreted properly. In general, I think the raw difference model may be a bit easier to work with and may be more logical when looking at non-consecutive seasons.

Czech Your Math is offline   Reply With Quote