Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: RE: proportion of explained variance with log-transformed outcome


From   Buzz Burhans <wsb2@cornell.edu>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: RE: proportion of explained variance with log-transformed outcome
Date   Tue, 18 Nov 2003 15:12:45 -0500

Many thanks to both Rich Goldstein and Nick Cox for their responses to my question under the above subject thread. Both had useful and informative ideas, for which I am grateful. Unfortunately, I remain unsure of how to describe the extent to which a significant independent variable contributes to explaining the variance in the original metric in a model of a log transformed dependent variable. Part of my difficulty arises from my use of gllamm with a two level model, because I am interested in associated random subject effects as well (else Rich's suggestion of -brsq- would work) . In fact, I am interested in the magnitude of the fixed treatment effect relative to the magnitude of the variance at the second level of my 2 level model. The issue is how to best form a basis for interpreting / describing the magnitude of the fixed effect relative to the random subject variance. I am satisfied with how I have done this within the transformed metric, but I need to interpret / describe it relative to the original metric as well. Basically, I have tested the treatment effect, and I now want to describe it relative to the random subject variance and the overall variance in terms of the original values as opposed to the transformed values, which I have already done.

I am considering examining the magnitude of the backtransformed fixed effect at given values of the covariates relative to the IQR for the random subject effects as an indication or description of how much greater the dispersion is in the random effects relative to the smaller difference for treatment, however, I would be interested in any further thoughts on how to get at this given the model is in a transformed metric and it would be advantageous to describe this information in the original metric.

I again thank both Rich Goldstein and Nick Cox for helpful responses.


Buzz Burhans
wsb2@cornell.edu


*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/




© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index