st: Multicollinearity problem in Logistic survival analysis

["Lu, Zhenyan" <lu_285@fisher.osu.edu>]

Dear all,

I am using the logistic command to do a discrete-time survival analysis. I know logistic mode is fixing a curve for the data rather than a linear line, but I am not sure whether,  and to what extent, the collinearity problem among variables in the model will impact the results. If yes, how to deal with it.
In my research I have 6 variables that are highly correlated, correlation value up to .71 to .83 based on large samples (n>140,000). These 6 variables come from two subpopulations and each of three variables in each subpopulation is calculated on three overlapped levels, such like region, product, product and region. I think the way these variables calculated is the main reason of the high multicollinearity problem. But as they are coming from the theory and are key predictors in my model I cannot delete any one without omitted variable problem. And even more complicated is that I have to include square terms for each of these 6 variables in the model at the same time to test the curvilinear relationship. So I am really concerned about the potential problem in the model. I did some robust tests by adding  or deleting some control variable, use smaller sample and others, but there is no sig changes.
If anyone has any experience or idea in how to deal with this problem, it will greatly help me out and I really appreciate all your valuable suggestions.
Zhenyan (Catherine) Lu
Fulbright Visiting Scholar
Management and Human Resource
Fisher College of Business
Ohio State University
700 Fisher Hall
Columbus, OH 43210

IPh.D Candidate
Strategic Management
Guanghua School of Management
Peiking University
Beijing, China 100871

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/