Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | David Greenberg <dg4@nyu.edu> |
To | statalist@hsphsun2.harvard.edu |
Subject | Re: st: alternative to SW in Stata |
Date | Thu, 12 Jul 2012 16:01:44 -0400 |
If some of the indicators are highly correlated,you are going to lose power due to severe multicollinearity. What's more, the interpretations may not be what you really want to get at. Suppose, for example, you want to find out the impact of poverty on life expectancy. You might have several indicators of poverty: income, wealth, being on welfare, living on public housing. PUtting them all into an equation will give you the effect of each controlling for all the others. FIne if you want to know specifically the effect of being on welfare (for example), but not if you want to know the effect of poverty. This is a conceptual issue that you should think about in relation to your variables, but it has implications for your statistical strategy. David On Thu, Jul 12, 2012 at 3:54 PM, Ricardo Ovaldia <ovaldia@yahoo.com> wrote: > Thank you for the siuggestion David. However, the idea is to identify those (risk factors) variables that are most highly associated with time to failure. By creating factors scores using factor analysis, I lose the ability to identify these important variables individually. > > Ricardo > > Ricardo Ovaldia, MS > Statistician > Oklahoma City, OK > > > --- On Thu, 7/12/12, David Greenberg <dg4@nyu.edu> wrote: > >> From: David Greenberg <dg4@nyu.edu> >> Subject: Re: st: alternative to SW in Stata >> To: statalist@hsphsun2.harvard.edu, ovaldia@yahoo.com >> Date: Thursday, July 12, 2012, 2:41 PM >> If some of your predictors can be >> understood as multiple imperfect >> indicators of an underlying latent variable, you could >> create a scale >> by doing a factor analysis of your predictors, and using the >> factor >> scores in the Cox regression. David Greenberg, Sociology >> Department, >> NYU >> >> On Thu, Jul 12, 2012 at 2:55 PM, Ricardo Ovaldia <ovaldia@yahoo.com> >> wrote: >> > Hello, >> > >> > We recently submitted a manuscript where we use -stcox- >> and the stepwise procedure to reduce the number of potential >> risk factors from 30 to 5. One reviewer commented that using >> stepwise was inappropriate but did not provide an >> alternative other than to say that we should retain all 30 >> factors. Giving our sample size (n=1000 and 105 failures) >> retaining all the factors would result in an over-fitted >> model. I know that there are limitations to using stepwise, >> many already discussed on statalist, bt we felt that the >> results we obtained were reasonable based on the science and >> current literature. However, what would be an alternative to >> SW for –stcox-? Is there a command in Stata or can someone >> suggest an acceptable method to generate a more parsimonious >> model? >> > >> > Thank you in advance, >> > Ricardo >> > >> > >> > Ricardo Ovaldia, MS >> > Statistician >> > Oklahoma City, OK >> > >> > * >> > * 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/ >> >> * >> * 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/ >> > > * > * 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/ * * 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/