Re: st: Transforming regressors prior to estimation in count models

[Richard Williams <Richard.A.Williams.5@nd.edu>]

At 03:43 PM 6/7/2005 -0400, Rkaminski@aol.com wrote:
> Dear All,
>
> I'm about to embark on an  analysis using count models to replicate a study
> in which the authors log  transformed or used some other transformation on
> their regressors prior to  estimating their models. I was under the 
> impression
> that typically one begins  with untransformed variables in count models, 
> assess
> the fit, and then -- if  necessary -- transforms regressors.
>
> Any thoughts on these  approaches?
First off, I'm not aware of anything "special" about count models when it 
comes to transformations of the regressors, e.g. I don't know why your 
strategy would inherently be different for count models than, say, for 
OLS.  But, there are lots of things I am not aware of, so maybe somebody 
can offer more insights on this.

In general, it seems to me that variable transformations are driven by 
theoretical and/or empirical concerns.  Theory might argue that you should 
take logs, squares or whatever; and if theory is ambivalent or silent then 
observed empirical relationships may guide the choice.

If you are in the theory is silent category - then I would probably proceed 
as you say.  There are all sorts of ways to transform variables, so I don't 
think I would try them all out just to see what happens.  But, if you are 
replicating someone else's work, they may have had good reasons for doing 
what they did, so theory may not be silent here.


-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
FAX:    (574)288-4373
HOME:   (574)289-5227
EMAIL:  Richard.A.Williams.5@ND.Edu
WWW (personal):    http://www.nd.edu/~rwilliam
WWW (department):    http://www.nd.edu/~soc

*
*   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/