RE: st: Re: margins after xtlogit,fe

["Santos Silva, J.M.C." <jmcss@essex.ac.uk>]

Dear Andre,

I am not an expert on this, but I do not think it makes sense to compute marginal
effects after FE logit. The reason is that the marginal effects depend on the value
of the FE, which are not estimated. Therefore, no meaningful marginal effects can
be computed after logit FE. Having said that, I suggest that you have a look at the
following (rather obscure) paper to see a type of marginal effect that can be 
estimated:

Kitazawa, Yoshitsugu (2012). "Hyperbolic Transformation and Average Elasticity in
the Framework of the Fixed Effects Logit Model" Theoretical Economics Letters, 
Vol.2, No.2, 192-199.

All best wishes,

Joao

> Dear all,
> 
> Apparently I missed some important information in my previous question:
> 
> -margins- should only be applied in the context of interaction terms when
> proper factor language is set.
> 
> Thus,
> 
> xkc_f1 = xkc lnden c.xkc#c.lnden
> margins eydx(*) will produce marginal effects only for xkc and lnden,
> which absorb also the effects produced by the interaction term. -mfx- old
> command apparently reported an effect that is not clear to be right.
> 
> Further, [XT] manual (p. 234 for stata 12) refers that -pu1- cannot be
> correctly handled by margins after -xtlogit, fe-. Same thing for -pc1- (p.
> 285 in [R]).
> 
> Is this correct or there is any other way for obtaining marginal effects
> of interactions?
> 
> Any clue is appreciated.
> 
> André
> 
> 
> On Tue, 25 Dec 2012 03:55:28 +0100, André Ferreira Coelho
> <andre.f.coelho2011@novasbe.pt> wrote:
>> Dear Statalist,
>> 
>> I've have been trying to compute marginal effects after xtlobit, fe with
>> an interaction term.
>> 
>> Essentially my model is xkc_f1 = xkc lnden c.xkc1#lnden, where xkc_f1 is
> a
>> leading dummy variable and lnden is continuous.
>> 
>> And the -margins, dydx- work fine after -logit-.
>> 
>> Tough, i am not sure about how to handle with xtlogit, fe. It seems that
>> -predict- pu0 (which is not adequate for FE), xb and pc1 are possible
>> solutions.
>> 
>> However, writing
>> 
>> xtlogit f1.xkc xkc lndens c.xkc#c.lndens, fe
>> margins, dydx(*) predict(xb)
>> 
>> produces the following output:
>> 
>> 
>> Average marginal effects                          Number of obs   =
>> 63355
>> Model VCE    : OIM
>> 
>> Expression   : Linear prediction, predict(xb)
>> dy/dx w.r.t. : xkc lndens lnpinteract 2.time 3.time 4.time 5.time 6.time
>> 7.time
>> 
>> 
> ------------------------------------------------------------------------------
>> |            Delta-method
>> |      dy/dx   Std. Err.      z    P>|z|     [95% Conf.
>> Interval]
>> 
> -------------+----------------------------------------------------------------
>> xkc |    .406874   .0246301    16.52   0.000     .3585998
>> .4551481
>> lndens |   .1191858   .0125079     9.53   0.000     .0946708
>> .1437008
>> lnpinteract |     .03134   .0077541     4.04   0.000     .0161422
>> .0465377
>> ...
>> 
>> While, -margins, dydx(*) predict(pc1)- generates the error message:
>> "predict option pc1 not appropriate with margins"
>> 
>> I was wondering if margins is correctly applied and if there is any
>> different way for using -pc1- option.
>> 
>> I know that odds-ratio are generally preferable but following Marteen
> Buis
>> (2010) article I was also thinking in present both EM and OR.
>> 
>> Thank you in advance for any answers.
>> 
>> André


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/