Panel data

[tboy ani]

Hello everyone!
My name is Tobi and i want to ask you to help me.
In my panel data a RE model was chosen as the best model than a FE model and Pooled regression model.
I've conducted tests on autocorrelation, heteroskedasticity and serial correlation.
and results showed I had these problems.

I decided to use robust standard errors but then my control variables became insignificant !
could someone please advise as to why that is happening ?

[Carlo Lazzaro]

Toby:
you do not say anything about the N and T dimensions of your dataset.
That said, if you actually detected those problems, you were right in invoking non-default standard errors.
As highlighted pretty often on this forum, statistical significance is not the aim of inference: both non-significant and significant results are equally informative.
Your effort should focus on giving a fair and true representation of the data generating process.

[tboy ani]

Hi Carlo, for my dissertation I am looking at how ESG scores leads to increases in Corporate financial performance.My data is composed of 50 firms in 8 years

1.) i Set my data to panel data

2) I then ran fixed effects and random effects followed by the hausman test and got the following results

The results suggest that random effects model is suitable , am i Correct ?


3.) my fixed effects results as follows

4) Then to account for heteroskedasticity , i use clustered robust standard errors

But now my ESG score variable has decreased significantly , I am not sure what implication this has on my hypothesis that ESG scores affect Tobinsq measure ?

are you able to advise thank you very much

[Carlo Lazzaro]

Toby:
it is not clear wether you performed -hausman- and then invoked non-default standard errors or not.
Please note that -hausman- does not allow non-default standard errors, whereas the community-contributed command -xtoverid- does (type -search xtoverid- to spot and install it).
As previously replied, if non-default standard errors made some of your coefficients (no more) statistical significant, they were only pretending to be so when you used default standard errors.

[tboy ani]

Thank you for the reply, so random effects model is corrected ? also what exactly does xtoverid do ? thanks

[Carlo Lazzaro]

Toby:
the -hausman- test does not show evidence of rejecting the null: hence (even considering the warning message about VCE matrix), I would go -re-.
As far as -xtoverid- is concered, please read the help file that comes with it.