Re: st: sign of coefficients in a xtprobit

[David Hoaglin <dchoaglin@gmail.com>]

Hi, Luke.

Is your supervisor's expectation for the sign of the coefficient of
speaker_fellow based on the likely relation between the outcome
variable and speaker_fellow when speaker_fellow is the only predictor?
 If so, bivariate relations often change when the model adjusts for
other variables, as your model does.

You should definitely check for problems in the data.  I am not sure
what diagnostics are available with xtprobit, but you should be able
to learn quite a lot by fitting a logistic regression model.  If the
data have no problems (e.g., with erroneous or highly influential
observations), the problem may be in your supervisor's expectation for
the sign of that coefficient.  Many people, including supervisors, do
not have an adequate understanding of regression models of various
types.

David Hoaglin

On Sat, Feb 9, 2013 at 10:40 AM, Luca Marcolin <gonzalo07@hotmail.it> wrote:
> Hi,
>
> the command is xi: xtprobit voto national_attitude party_attitude speaker_fellow speaker_nation i.nation i.party, re
>
> The problem is with speaker_fellow (defining if the speaker is from the same nation of the person who votes). I would expect that the relation was positive, but it is not (and the variable is significant).
>
> Yes, probably should not be so surprising, but my supervisor doesn't think so...
*
*   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/