Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Dan Kahan <dmkahan@gmail.com> |
To | statalist@hsphsun2.harvard.edu |
Subject | Re: st: RE: analysis of mixture experiments |
Date | Thu, 23 Sep 2010 16:28:28 -0400 |
thanks! Austin--can you point me to e.g. of researchers using that strategy? It is what initially occurred to me, & is what a number econometricians I've asked seem to think is perfectly fine, but I made the mistake of thinking too much & looking for confirmation & ran into a very rich literature (in statistics but in particularly in engineering) on the appropriate way to model such data (the Cornell book deals w/ much of it) that insists that simply plugging the proportions into OLS is not appropriate--not just b/c it generates coefficients that cannot be interpreted in the conventional way but b/c the parameter estimates do not appropriately take the interrelated nature of the predictors into account. (Some propose "ridge regression," others various transformations.) Maybe I'm not seeing why what worries these researchers shouldn't worry me. --DK * * 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/