Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | "Clyde Schechter" <clyde.schechter@einstein.yu.edu> |
To | statalist@hsphsun2.harvard.edu |
Subject | RE: Re: st: Interaction terms |
Date | Wed, 4 May 2011 11:24:31 -0700 |
Maarten Buis writes: "We cannot change the correlation among explanatory variables, so the only thing we can do is to gather more data..." in response to a query about dealing with multicollinearity. This is true if we restrict ourselves to simple random sampling. But if it is essential to disentangle the effect of one of a group of correlated variables (at the expense of losing any information about the others' effects) matched-pair (or matched n-tuple) sampling may be useful. It requires a different approach to analysis. And it comes at a price: if the correlations are truly high you will have to discard many potential observations because a proper match cannot be found for all. A less stringent approach, stratified random sampling, entails less loss of data and information and achieves less disentanglement but might still be satisfactory for some situations. And, of course, these approaches have to be chosen at the design phase. By the time the data are in hand, it is too late for these. Clyde Schechter Associate Professor of Family & Social Medicine Albert Einstein College of Medicine Bronx, NY, USA Please note new e-mail address: clyde.schechter@einstein.yu.edu * * 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/