Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Maarten buis <maartenbuis@yahoo.co.uk> |
To | statalist@hsphsun2.harvard.edu |
Subject | Re: st: Multicollinearity problem in Logistic survival analysis |
Date | Sat, 27 Mar 2010 09:25:25 +0000 (GMT) |
--- On Sat, 27/3/10, Lu, Zhenyan wrote: > 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). <snip> So I am really > concerned about the potential problem in the model. By adding multiple explanatory variables you want to be able to distinguish between them. If two variables are *perfectly* correlated, how would you be able to distinguis between the two? This is why Stata will drop variables when there is perfect correlation. If two variables are strongly but not perfectly correlated, then that means that it will be more difficult for Stata (or any other statistical software package) to distinguish the effects of the two variables. This leads to higher standard errors, which is exactly as it should be: It is more difficult to distinguish the variables, so we are more uncertain about the results, so the standard errors should be larger. In other words there is no 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. Adding square terms is a very limited way of checking for curvilinearity. I like the linear spline (see: - help mkspline-) as good compromise between a flexible non-linear curve and parameters with an easy interpretation. Others like more smooth non-linear curves like restricted cubic splines or fractional polynomials. If you want to interpret the results of those curves you'll have to make graphs. For restricted cubic splines see: http://www.stata.com/meeting/sweden09/se09_orsini.pdf and http://ideas.repec.org/p/boc/dsug09/04.html For fractional polynomials see: -help fracpoly-. Hope this helps, Maarten -------------------------- Maarten L. Buis Institut fuer Soziologie Universitaet Tuebingen Wilhelmstrasse 36 72074 Tuebingen Germany http://www.maartenbuis.nl -------------------------- * * 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/