Murasko, Jason Elliot <[email protected]> noticed that -mfx- will report
different results when used after a survey estimation command than when used
with the standard estimation command (using the same sampling weights):
> I am trying to derive marginal effects after a probit estimation. I have
> survey data, so I use the "svyprobit" command after specifying weight, psu,
> and strata variables. I then use the "mfx" command and obtain the usual
> results.
> However, when I use the "mfx" command after a simple probit estimation where
> I account for weights but ignore the survey design, I obtain very different
> marginal effects.
> In Stata language, I am comparing:
>
> svyprobit y x, subpop(sample)
> mfx
>
> probit y x if sample==1 [pweight=weight]
> mfx
>
> Is there a reason I should be obtaining different results from these
> commands? If so, which is correct? Any help will be appreciated.
The difference is in the value of 'x' used to compute the marginal effect.
Jason should notice a different value in the "X" column after -mfx- between
the two analyses. The difference here is cause by the use of the -subpop()-
with -svyprobit- and -if sample==1- with -probit-. By default -mfx- uses
the weight average of the predictors from the estimation sample. For
-svyprobit- the estimation sample is more that just those values for which
-sample==1-, but Jason specifically restricts the estimation sample in the
call to -probit-.
Jason can use the -at()- of -mfx- to chose at which value of -x- to compute
the marginal effect.
--Jeff
[email protected]
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
