[Imen Fredj]

I run an ordered probit regression and I have small sample. I want to know what is the bootstrap commend in stata when applying an ordered probit model??

[Carlo Lazzaro]

Imen:
I assume you refer to bootstrap standard errors (SEs).
If that were the case, you can tweak the following code according to your needs:

oprobit <depvar> <indepvars>, , vce(bootstrap, reps(200))

You can change the number of bootstrap replicates; usually, 50-200 are considered enough for bootstrap SEs (see http://matthieuchemin-research.mcgil...aper/efron.pdf).

[Imen Fredj]

Dear,
My dependent variable is recommendation and takes a score from 1 to 4. I have two sub-samples and I want to ascertain whether the difference between the two sub-samples exists. Shall I use wilcoxon rank test??
Thank y

[Carlo Lazzaro]

Imen:
you may add an -i.subsample- preedictor in your probit regression model.

[Imen Fredj]

I did that when regressing but as descriptive statistics what shall I Do? or I can run the Wilcoxon rank test?
Thank you

[Carlo Lazzaro]

Imen:
you may want to try something along the following lines:

forval i=1(1)4 {
ttest <depvar> if <depvar>==`i', by(subsamples) unequal
 }

Please note:
- check whether you data are suitable for a -ttest- (otherwise, see -help bootstrap- and related entry in Stata .pdf manual for a bootstrap ttest;
- being a -ttest- a parametric test of hypotheses, it is an inferential procedure; for descriptive statistics, see:

bysort <depvar> subsamples: tabstat <depvar>, stat(count mean sd p50 min max)

[Imen Fredj]

Hello,
I run my model firstly for the full sample and then for the financial and non-financial sub-samples. I found that some variables are significant for the sub-samples and not for the full sample.
Is it because of the heterogeneity of the full sample ?

Thank you

[Carlo Lazzaro]

Imen:
I would say that your results are by no means surpising.
As usual, posting what you typed and what Stata gave you back will increase your chances of getting helpful replies.

[Imen Fredj]

Thank you for your response:
I use an ordered probit regression to examine the dependencies between analysts' recommendations and corporate governance mechanisms. The wilcoxon rank test documents that recommendations differs within financial and non financial samples. So, I decided to rerun my model for these two sectors.
I found for example that Full sample Financial Non-financial sample

lLnBsize

-1.664

-5.181**

-5.149*

This variable is not significant for the full sample but significant in the two sub-samples and I'm asking for reasons. I said probably because of heterogeneity.
Have you another reasoning?

Thank you

[Carlo Lazzaro]

Imen:
your query was clearly stated.
However, my reply elaborated on the scant details you shared with the list.
Perhaps, the difference in statistical significance you complained about was simply due to a different number of observations in the two subsamples (which, in turn, can be driven by the amount of miissing values in any variable).
Unless you share what you typed and what Stata gave you back (as per FAQ), I'm afraid you cannot receive more helpful advice.

[Imen Fredj]

Thank you very much for your reply.

[Imen Fredj]

Hi,
I have one question, Does 3SLS correct for omitted variable bias?

[Carlo Lazzaro]

Imen:
please start a new thread as this one is far off the original tiltle. Thanks.