 Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# Re: st: comparing logistic regression coefficients between samples

 From Maarten Buis To statalist@hsphsun2.harvard.edu Subject Re: st: comparing logistic regression coefficients between samples Date Wed, 6 Jul 2011 09:26:03 +0200

```On Tue, Jul 5, 2011 at 11:31 PM, Sunniva Andersen wrote:
> How can I do this in stata 11.   I have estimated the same
> logistic regression model in two different samples, A and B.
> I compute AME after the logit. Can someone suggest a method by which I
> can test the hypothesis that var1 (b1) in sample A is different from
> var1(b1) in sample B.

That is an interaction effect, and those are hard when you want to do
a logistic regression and marginal effects. In essence they will be so
different from observation to observation that your conclusion will be
that for some persons var1 will have a significantly larger effect in
sample A than B, for others var1 will have a significantly smaller
effect in sample A than B, and for the remaining individuals there is
no significant difference. If this variability in interaction effect
across individuals represented something present in the data, than
there could be situations where this might be useful. However, it is
just the result of the fact that marginal effects fit a linear model
on top of your logit model in order to summarize your logit model, and
that linear model does not fit very well. Averaging is not going to
solve it as these effects are just too different from one another in
order to be meaningfully summarized with one number.

You have two options:
1) If you want to continue using logistic regression, you'll have to
interpret your results as odds ratios and the interaction term as a
ratio of odds ratios. See: M.L. Buis (2010) "Stata tip 87:
Interpretation of interactions in non-linear models", The Stata
Journal, 10(2), pp. 305-308.

2) If you want to continue interpreting your coefficients as a
difference in probabilities rather than ratios of odds than you can
use a linear probability model (i.e. just use -regress- in combination
with the -vce(robust)- option instead of -logit-).

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany

http://www.maartenbuis.nl
--------------------------

*
*   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/
```