You should make LM test with eViews programs..Anyway of course you must check some articles at the literature about your subject...

please visit:

https://www.stata.com/statalist/archive/2007-07/msg00278.html

Dear Ibrahim Cutcu,

Thanks for your reply.

I am using logistic regression in my model and in the relevant papers they did not mention the auto-correlation. They simply checked fixed effect/random effect and run the model. Therefore, I confused whether I should check auto-correlation or not?

Thanks

Thanks Ehsan Khalili

Hi Everyone,

I just run my logit model and get the following result, where my Presdo R2 is just 1%, while result is model is significant. Could you please give me comment whether model presents the goodness of fit or not? I am examine whether non-audit fees(independent variable) impairs audit opinion(Opinion qualified) or not

logit Opinion_Qualified LnAF LnNAF NAFGROWTH

Iteration 0: log likelihood = -1971.1174

Iteration 1: log likelihood = -1950.9304

Iteration 2: log likelihood = -1950.8297

Iteration 3: log likelihood = -1950.8263

Iteration 4: log likelihood = -1950.8263

Logistic regression Number of obs = 3,397

LR chi2(3) = 40.58

Prob > chi2 = 0.0000

Log likelihood = -1950.8263 Pseudo R2 = 0.0103

-----------------------------------------------------------------------------------

Opinion_Qualified | Coef. Std. Err. z P>|z| [95% Conf. Interval]

------------------+----------------------------------------------------------------

LnAF | .0842173 .0397921 2.12 0.034 .0062261 .162208

LnNAF | -.051061 .0080633 -6.33 0.000 -.0668646 -.0352573

NAFGROWTH | -.0001973 .0004767 -0.41 0.679 -.0011316 .000737

_cons | -1.64969 .4301426 -3.84 0.000 -2.492754 -.8066258

-----------------------------------------------------------------------------------

Thanks

first of all, R2 in logit models don't have the straightforward interpretation as OLS.

significance of model is one of the criteria.

please check normality (and outliers) of your data (important).

xtsktest in stata...

please check frequency of zeros and ones of your data (important).

Auto-correlation is not really an issue with panel data. It is more an issue in time series data. For your logit model, you will have to declare your data as panel in Stata and then run the logit model using xtlogit to fit random effects, conditional fixed effects as well as pop-averaged logit models. As Ehsan says, due to the nature of the dependent variable and the model techniques, you get the Pseudo Rsquared in binary dependent regression models and not the Rsquared you obtain in OLS. Refer to https://www.stata.com/manuals13/xt.pdf for more details.

Dear Ehsan Khalili.

I got it. Really appreciated.

Take care

Hi John Ng'ombe,

Thank you very much for your very clear explanation. Its cleared my doubts.

Actually I found the reason of my low Pesudo R2 as I was running model without controlled variables. Therefore, after adding control variables my Pseudo R2 raised and reached to 33% with 0.000 sig level.

Take care

Hi Everyone,

I have one question regarding fixed effect and random effect. In my research my moderate variable is economic condition, by this reason I divided my full sample into three different sub samples: (1) before global financial crisis(GFC) 2005 to 2007, (2) Mid GFC 2008-2009, and (3) after GFC 2010-2014. It means, I have total four samples three samples for economic conditions and fourth sample is Full sample (2005 to 2014). So my question is, do I have to check fixed/ramdom effect for 3 sub samples or just for full sample. When I check fe/re for full sample it shows me fe while when I check for subsample it shows re except for after GFC.

Looking forward to hear.

Thanks for your time

Dear Ehsan Khalili,

As you mentioned in your suggestion to check the frequency of Zero values. Actually, in my data there are lots of zero in non-audit fees which means the company did not invite same auditor(the auditor who gives audit services) to get non-audit services. For this reason there are many zeros in my non-audit fees sample. I hope it won't affect my result quality.

kindly give me your opinion on it.

Thanks

Please visit: https://www.statalist.org/forums/forum/general-stata-discussion/general/1297526-multicollinearity-panel-data/

Thanks Elma,

Its really helpful to clear my some other doubts as well.

Appreciated.

I have one more question please.

I already run my model (Probit) and my result is significant as I expected. Now I am working on some post estimation tests to find the goodness of fit of my probit model. Here is the result of my model (Prob > chi2)

Wald test (test): 0.000

Hosmer Lemeshow test (gof): 0.080

Likelihood Ratio test (lrtest): 0.000

could you please tell me, above result shown enough of model goodness or I need to run some other (important) tests to confirm my model's goodness?

Thanks for you time.