With reference to the above result, your data is Heteroskedastic. This is because the Chi- value is high and it is significant at 1%. You need to run for Hausman test as a procedure for the selection of either a fixed or a random model.

Dear Anurag,

in addition to Hussaini answer, typically the outcome of the Breusch-Pagan test is considered to decide between POLS and RE in panel regression.

Best regards,

Florian

Mr Bala,

Does hausman test run only when the chi 2 is high? Pls when it is otherwise what will one do?

Lasisi Olalekan Isiaka : The two test assess different things: Breusch-Pagan test assesses heteroscedasticity and Hausman test assesses endogeneity.

Best regards,

Florian

@Lasisi Olalekan Isiaka. Check, I dont think Hausma test examines endogeneity. Hausman test is just a procedure for selection between random and fixed models. Thus, when heteroscedasticity test Chi value is low and p-values is high, it indicates that there is no presence of heteroscedasticity. One can also go father to conduct xttest3, a test for a modified Wald statistic for groupwise in the disturbances of a fixed effect regression model. It is for employed after xtreg, fe or xtgls (with the default panels option). Best Regard.

The Hausman test in the context of panel analysis compares the RE and FE parameter estimates. Under H0 (the independent variables are uncorrelated of the fixed effects) both estimator produce consistent estimates but RE estimates are efficient. However, under H1 RE estimates are biased and FE estimates not. Typically, if H0 is rejected, it is concluded that at least one of your independent variables suffer from endogeneity.

Best regards,

Florian

Yes in the contest of instrumental variables. Durbin–Wu–Hausman (DWH) test or the augmented regression can test for endogeneity.

Put the word robust after the regression

Hussaini Bala I have run Hausman test for the selection of either a fixed or a random model, and the result shows that it is Random effect.

One more clarification, I have red somewhere that in Panel data, there will be Heteroskedasticity problem. Is is true?

I have 190 firms over 17 years (unbalanced panel)

I have run VIF test to check Multicollinearity, and for all the variable the VIF value is between 1 to 4. (Mean VIF 1.86). So i guess there is no Multicollinearity problem in the model.

In addition, do I really need to go for Heteroskedasticity test? M Imran Khan

Please Suggest.

Thanks in advance for your answer...@ Mohammed W. A. Saleh Hussaini Bala Florian Schuberth

1/ You can just first run your regression using robust standard error xtreg Y X1 X2, fe r or re r. If the result is good, that means the coefficients are significant, you don't need to worry about heteroskedasticity and serial correlation. In case, the result does not fit your expectation, you can use breush pagan or white test to test whether theire is heteroskedasticity or not. In case p-value=0, that means you have to use wLS or FGLS (weigh Least square or feasible Least square) to solve the problem.

2/ The Hausman test is run to see whether whether to use random effect or fixed effect. It does not deal with heteroskedasticity problem.

I hope I have answered to your question.

Souleymane Yameogo Thank you so much...

@Anurag Anand. In that case you can run variance clustered estimator robust standard error as it tends to be more trustworthy in the presence of heteroskedasticity . You may go through the following articles they could be helpful.

Hussaini Bala Thank you Sir...

I see no point in testing for these things as it is much more informative to model heteroscedasticity as an interesting phenomena not as a technical failure . Moreover differences revealed by robust standard errors suggest that the model is wrong not that the answers are now correct.

see

http://gking.harvard.edu/publications/how-robust-standard-errors-expose-methodological-problems-they-do-not-fix

on robust SEs and these on modelling heteroscdesasticty

Article Explaining Fixed Effects: Random Effects Modeling of Time-Se...

Article Fixed and Random effects models: making an informed choice

Also, it may sometimes be a sign that the model or data or both have a substantial problem if you do not have heteroscedasticity, a natural phenomenon.

Hi,

is it also possible to conduct the Beusch-Pagan Test also for three level random coefficient models?

Thanks :)

use xttest0 in stata