Koichiro Okamura
>
> I have a question about xtprobit. I ran xtprobit (random effects)
> for an unbalanced panel-data and found that all independent variable
> (sometimes except one variable or two) were significant. I'm not
> familiar with such a result and afraid that something might
> be wrong.
> So I am here seeking for your suggestions/advice.
>
>
> I guess the reason for "everything is significant" could be either:
>
> R1) there were too many failures (dependent variable = 0)
> than success
> (dep = 1); the result is biased toward failures;
>
> R2) the number of sample was huge (if the sample are so huge, the
> independent variables that aren't significant tend to be found
> "significant." Does it hold for probit, too?);
> or
> R3) the independent variables are really significant
> (i.e. I don't have to worry about the results).
>
>
> What would you do to examine what causes the "everything is
> significant"
> results (and whether the results are really correct)? More
> specifically,
> my questions could be, for example, when the proportion of
> success/failure is unbalanced very much, is xtprobit an appropriate
> command? Or are there other commands, more appropriate than
> xtprobit?
> etc.
>
> Also, please tell me, if there are other potential causes/problems
> you find that I'm not aware of, please.
>
>
> The attached in the rest of this posting is an example of the output
> from Stata 8 (one independent variable is (eventually) found as
> insignificant here):
> O1) tabulate and summarize (to see the overall view of data), and
> O2) xtprobit, preceded by probit in order to make xtprobit
> converge,
> and followed by quadchk checking the stability of xtprobit's results
> together with other commands.
> # As to the usage of probit, quadchk, and other commands, the series
> # of discussions on xtprobit in Sep. 2002 here is very helpful.
>
> Any help will be appreciated. Thank you very much.
< output deleted >
In so far as your question really is about -xtprobit- I can't
answer it. But on the face of it your results, as a series of
P-values,
are entirely plausible given your sample size. This is an entirely
general consequence of what significance tests do.
Having got a set of results which many researchers would envy, you
must now
be start worrying e.g. does that coefficient of the order of 1e-6 or
less
represent scientifically or practically important effects? (Depends on
what units they are measured in, as well.)
Nick
[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/
