Ordered probit with omitted variable bias

[Imen Fredj]

I need to regress ordered probit regression, but I have an omitted variable bias,

Could You recommend me which regression method should I use?

Thanks

[Carlo Lazzaro]

Imen:
you should provide more details to get helpful replies.
That said, what't the reasonable cause of the omitted variable bias that you experience? Endogeneity? Else?

[Imen Fredj]

I'm examining two relations:

The first link: My dependent variable takes score from 1 to 4, in such case I should use an ordered probit regression. However, my independent variables are corporate governance mechanisms and are subject to endogeneity bias and specifically omitted variable bias (some relevant variables could not be included because their unavailability). I'm asking, please, what regression model should I use (ordered probit regression can not solve the problem).

The second link: I have a panel and I face the same problem. The instrumental variable should be used but as I did not respect the assumptions related to the instrumental variable (not correlated with the error term ) it is rejected. And I'm using the fixed effect: Am I right?

Thank you??

[Carlo Lazzaro]

Imen:
first link: the problem lies in the fact that you have no instruments available fror an -ivprobit-. Skin through the literature in your reserach fiels and see what Others did in the past when presented with the same research topic and change your meodel specification accordingly.
second link: from your description I cannot get what's the problem there. I'm not clear the meaning of

...I did not respect the assumptions related to the instrumental variable...

: you have an endogeneity issue or not.

[Imen Fredj]

The past research used the OLS regression but I'm not convinced because past corporate finance literature argues the presence of endogeneity bias. I want to use the instrumental variable but I found that it is correlated with the explanatory variable and with the error term and hence I did not satisfy the exclusion restriction.

Thanks

[Carlo Lazzaro]

Imen:
can't you find different instrument(s) to make -ivreg- feasible?

[Imen Fredj]

How can test whether the instruments are valid or not in Stata?
Thank you

[Carlo Lazzaro]

Imen:
the whole stuff is covered in https://www.stata.com/bookstore/micr...metrics-stata/, chapter 6.

[Imen Fredj]

Thank you !

[Imen Fredj]

Unfortunately, I can't read it for free. Could you recommend me another site?

[Carlo Lazzaro]

Imen:
take a look at http://www.econometricsbysimulation.com/

[Imen Fredj]

Thank you! So kind

Another question to ask: when the variation of the regressors is absent over time (my variables are roughly unchangeable over time) I use OLS rathar then panel fixed effect?
Thanks

[Carlo Lazzaro]

Imen:
if your dependent variable is continuous, pooled OLS might be the way to go in that instance, but I would consider -xtrreg, re- first.

[Imen Fredj]

I run xtreg., fe the F statistic is 2.75 and Pvalue <5% : I use panel rather than OLS. After that I did hausman test and I found Pvalue> 5% so the use of random effect.
Stata gives this output.
The problem that the estimates of the OLS are more relevant than random or fixed panel. What shall I do?

[Carlo Lazzaro]

Imen:
if -hausman- does not reject the null, you should go -xtreg,re-.