I'm using multiple regression analysis. I found out that in order to avoid multicollinearity one way is to do mean centering. However, I have uncertaintisles as to what type of score should I use. Should I use raw scores? Or should I get the mean of these raw scores?
Btw I'll be using spss for that. Thanks in advance!
Floren,
Typical procedure is, for all raw values of a metric IV, subtract its mean from the raw score, creating a deviation score, then use that as the IV in the regression. I judge 'avoid multicollinearity" as a bit of an overstatement. Although mean centering often reduces collinearity, it depends on what sort you have, and do note that we are not all in agreement regarding mean centering. As well, there are other reasons to center. From here the discussion could get involved, so I will, for now at least, leave it at that.
My best on your research,
John
The original paper on this topic by Marquart and Snee is attached. Dr Snee is active on researchgate and I am sure would take your questions. He is a very kind and helpful person. Best, David Booth PS I am also attaching a paper on adaptive lasso which is a current modification of ridge that you might find useful.
1. How do you know if there is? Use SPSS for that analysis, find the variables that are colinear.
2. Again, Is it structural or data multicollinearity? Moderate or severe? Use the raw score or centered data (not the average/mean of all data nor scores which tends to be conservative) it will only reduce your statistical power.
3. Multicollinearity affects the coefficients and p-values, but it does not influence the predictions, precision of the predictions, and the goodness-of-fit statistics. If your interest is just prediction , don't worry about moderate collinearity, or the affected independent variables or
5. Firstly, TRY Factor Analysis to reduce the factors/variables before fitting the regression
- Badges
- Science topic