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

# Re: st: Binary Choice Model and fixed effects - interpreting the interaction effects?

 From Maarten Buis To statalist@hsphsun2.harvard.edu Subject Re: st: Binary Choice Model and fixed effects - interpreting the interaction effects? Date Mon, 2 Apr 2012 16:00:55 +0200

```That is due to the effect of that the constant is not calculated in
these models. So you'll have to interpret the odds ratios without the
baseline odds. This is not ideal but can be done, and I guess that is
the price you'll have to pay for estimating a fixed effects model...

-- Maarten

On Mon, Apr 2, 2012 at 3:39 PM, Benjamin Niug wrote:
> Maarten - thanks a lot for clarification.
>
> In case of a clogit xtlogit, the baseline-trick you applied, namely
> generating a variable
>
> gen baseline = 1
>
> and then running the logit-regression also on baseline such that the
> odds ratios of the interaction effect can be compared to the baseline
> odds ratio does not work due to multicollinearity of the baseline
> variable.
>
> How do I solve this problem? I guess a constant could serve the same
> purpose as your baseline variable - however it is not reported
> neither. How do I still come up with a meaningful interpretation?
>
>
>
> Am 2. April 2012 15:17 schrieb Maarten Buis <maartenlbuis@gmail.com>:
>> On Mon, Apr 2, 2012 at 2:37 PM, Benjamin Niug  wrote:
>>> @Maarten. Thanks. I tried to calculated the marginal effects as
>>> indicated in the paper you mentioned (M.L. Buis (2010) "Stata tip 87:
>>> Interpretation of interactions in non-linear models", The Stata
>>> Journal, 10(2), pp. 305-308)
>>>
>>> However, some interactions are not estimated / "estimable" by Stata
>>> using the -margins- command.
>>
>> The point of that article is that you should _not_ estimate marginal
>> effects. In that article I tried to be nice towards Edward Norton and
>> colleagues and tried to find some situation where marginal effects
>> might make some sense. I did find such a special situation in the case
>> of a fully saturated model(*), but in practice you should just forget
>> about that and go for odds ratios. In retrospect that inclusion of
>> marginal effects in the article was a mistake as this confuses more
>> than it helps.
>>
>> So the bottom line is: There is only one solution and that is to
>> interpret the results in terms of odds ratios.
>>
>> Hope this helps,
>> Maarten
>>
>> (*) A fixed effects model with covariates cannot be a fully saturated
>> model, so this is not an "escape route" open to you. You really really
>> really have no other option than to learn how to use and report odds,
>> odds ratios and ratios of odds ratios.
>>
>>
>>
>> --------------------------
>> 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/
>
> *
> *   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/

--
--------------------------
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/
```