RE: st: hierarchical logistic regression command

[Nick Cox <n.j.cox@durham.ac.uk>]

-hireg- (SSC) is indeed configured only to work with -regress-. That could be a little clearer in its help, but it is at least implicit. It does not work with categorical outcomes because it was never designed to do that. It's a dog that was never designed to be a cat too. 

I'd add to the pot -nestreg- which seems to me a much more obvious candidate for what you want to do than -stepwise-. If tempted to use -stepwise- Google "Harrell stepwise" and read around a bit first. 

I'd echo Maarten to the extent of adding a gnomic comment that being able to do this doesn't make it easy to see how well you did it. 

Nick 
n.j.cox@durham.ac.uk 

Maria Niarchou

thank you for your e-mail. I am sorry I didn't mention this, the hireg program was written by Paul H. Bern (Syracuse University phbern@syr.edu) so I think we are talking about the same one. 

I am currently using this command:
stepwise, pe(.2) hierarchical: logistic 

which seems to work. Do you think it is a correct way of doing hierarchical logistic regression? 
I only have 2 independend variables that I add to the model.

From: maartenlbuis@gmail.com

> 2012/2/16 Maria Niarchou :
> >> I would like to perform a hierarchical logistic regression analysis in which
> >> independent variables are entered in blocks. 'Hireg' doesn't seem to work with categorical outcomes.
> >> Could you please let me know if there is an alternative command to do this?
>
> -hireg- is a user written program, so per the Statalist FAQ you must
> tell us where you got it from. The purpose of that rule is not to make
> your life hard, but to make sure that all of us are talking about the
> same program. There are often different versions of user written
> programs floating around in cyber space, if you do not tell us which
> version you are referring to than it can easily happen that we are
> talking about different versions and you will get advise that does not
> help you.
>
> Anyhow, it is good news that -hireg- (I assume you got it from SSC)
> does not work with logistic regression, because that is not a good
> idea with non-linear models like -logit-. A lot of the nice properties
> of these comparisons of models with different sets of independent
> variables critically depend on the linearity (in parameters) assumed
> in -regress-, and these nice properties do not generalize to
> non-linear models. For explanations see:
>
> M.L. Buis (2010) "Direct and indirect effects in a logit model", The
> Stata Journal, 10(1), pp. 11-29.
>
> Williams, R. (2009) "Using heterogenous choice models to compare logit
> and probit coefficients across groups", Sociological Methods &
> Research 37: 531-559.
>

*
*   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/