Erasmo,
Take a look at the Stata Journal paper by Norton et. al on interaction
effects in logit and probit models - see the recent thread on probit
with interaction dummies for the exact reference. I suppose you could
exhange the probabilities in their formulas with the average sample
predicted probability (using the pc1 option) as suggested by Steve
Stillman in the thread below. I'm not sure exactly how you would
interpret this though... the problem remains that marginal effects in
the usual sense of the word cannot be calculated following a fixed
effects logit model.
Arne
2008/7/25 Erasmo Giambona <[email protected]>:
> Thanks very much Arne. Hope you could give me a hint on what the b-hat
> for the interaction term to calculate the pseudo marginal effect
> should be. Is this simply the coefficient on the interaction term?
> Thanks and regards,
> Erasmo
>
> On Fri, Jul 25, 2008 at 6:21 AM, Arne Risa Hole <[email protected]> wrote:
>> Erasmo,
>>
>> You cannot calculate marginal effects following a fixed effects logit,
>> with or without interactions, since the fixed effects are not
>> estimated. See for instance the following thread:
>>
>> http://www.stata.com/statalist/archive/2005-10/msg00464.html
>>
>> Note that my -nlcom- code did not relate to the fixed effects logit
>> but to McFadden's choice model which is equivalent computationally but
>> very different conceptually.
>>
>> Arne
>>
>> On 24/07/2008, Erasmo Giambona <[email protected]> wrote:
>>> Dear Statalisters,
>>>
>>> I have a panel dataset with a (0,1) dependent variable (d), where d=1
>>> means that the firm has merged with another firm. The model looks as
>>> follows:
>>>
>>> dit = axit + byit + c(xit*yit) + fzit + gwit + ei, where xit, yit,
>>> zit, wit are continuous independent variables for firm i at time t,
>>> and (xit*yit) is the interaction of these two continous variables. I
>>> am estimating the model with clogit. I am interested in gaining some
>>> insights on the marginal effects of the interaction terms. In
>>> particular, I would like to see what is the percentage point increase
>>> in the probability of a merger as x increase by 1 standard deviation
>>> (or 1 percentage point), y is set at its 75th percentile and the
>>> remaining two regressors (e.g., z and w) are set at their sample mean.
>>>
>>> I understand that this is not an easy task. I found a thread (e.g.,
>>> "clogit marginal effects?") in Statalist where Arne Risa Hole provides
>>> an nlcom that could be used after clogit, but it does not fit what I
>>> need.
>>>
>>> Can anyone provide any help on the nlcom code needed to obtain the
>>> result that I am looking for?
>>>
>>> Thanks,
>>> Erasmo
>>> *
>>> * 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/
>>
> *
> * 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/
