Mundlak interpretation

[Krissy Philips]

As per: http://blog.stata.com/2015/10/29/fix...dlak-approach/

I have run the following:

quietly xtreg netreturn sin religiositymean sin*religiositymean beta lmarketcap pb bev lgdp spread inflationrate open law year1 year2 year3 year4 year5 year6 year7 year8 year9 year10 mean_lmarketcap mean_pb mean_lgdp mean_spread mean_inflationrate mean_law mean_open mean_religiositymean i.country_c, re

estimates store mundlak
test mean_lmarketcap mean_pb mean_lgdp mean_spread mean_inflationrate mean_open mean_law mean_religiositymean

giving me:

1) What does this mean, how do I interpret the chi2 stat?

2) Also I do not get what benefit the user written"mundlak command gives" - results attached - how to interpret this table?

[Marcos Almeida]

Hello Krissy,

You may wish to take a look at this article in the Stata Blog.

Additionally:

Mundlak test has a similar interpretation as Hausman's.

It can be used under "robust" estimations, preferrably with e(sample).

Also to notice, the Hausman test assumes homoscedasticy and shall not accept serial correlation.

[Krissy Philips]

Hello Krissy,

You may wish to take a look at this article in the Stata Blog.

Additionally:

Mundlak test has a similar interpretation as Hausman's.

It can be used under "robust" estimations, preferrably with e(sample).

Also to notice, the Hausman test assumes homoscedasticy and shall not accept serial correlation.

Hi Marcos, thanks for your reply.

I did look at that article, but I was confused as I was encountering conflicting conclusions on the internet.


A) In this paper here, http://www2.warwick.ac.uk/fac/cross_...e8issue1/brown, she finds that P>F is 0.0000, means are strongly significant, so concludes to use random effects.


However when looking at stata's help one how to compute the test: http://blog.stata.com/2015/10/29/fix...dlak-approach/ they find
Prob > chi2 = 0.0114 So they "reject the null hypothesis. This suggests that time-invariant unobservables are related to our regressors and that the fixed-effects model is appropriate. "


1) I don't understand how megan got a F stat for the test when it produces a chi^2
2) Even so she would reject the null on the basis that its 0.000 prob, hence should use FE, contrary to what she concludes?

[Marcos Almeida]

Hello Krissy,

As I underlined in the previous, the interpretations of both the Stata Blog and the Stata Manual are surely on the same track.

These are the ones I learned.

[Krissy Philips]

Hi Marcos, thanks for the reply. So you would agree that the one in the appendix of the attached specified paper is not correct conclusion/interpretation of the test?

[Sebastian Kripfganz]

Yes, the conclusion drawn by Brown is simply wrong. Her test result does not support a RE specification.

[Carlo Lazzaro]

Krissy:
Sebastian is right.
I suspect that in writing the paper Megan Brown mixed up the description of the Breusch-Pagan results with the ones of the Mundlak test.
The paper echoes the usual lesson: having too many tests around increases the chances of making mistakes in text and/or tables.

[Marcos Almeida]

Absolutely agree with Sebastian and Carlo.

As said before, differences apart, already remarked as well, there is no polemic on this matter: both tests have a similar interpretation in terms of p-values.

Contrary to the plethora of statistical "tests", and just as a side note: the pie graph risks raising eyebrows, well, for being potentially "not so much informative".

[Krissy Philips]

Thank you all for the useful help! Much appreciated.