Confused with Haumsna...!!!

[Guest]

~~Hi everyone.
I am doing a panel data research on 53 Asian countries. I am a little confused about hausman test.
After saving the results of fixed and random effects, I run hausman using the command 'hausman fe re'. STATA shows me the message 'chi2<0 ==> model fitted on these data fails to meet the asymptotic assumptions of the Hausman test; see suest for a generalized test'.
Conversely, if I run 'hausman re fe' , STATA generates results and shows that I should go with the random effects. I just want to know if there is a significant difference between 'hausman fe re' and 'hausman re fe'???
Can anyone help? Thanks in anticipation.

[Guest]

Hi everyone.
I am doing a panel data research on 53 Asian countries. I am a little confused about hausman test.
After saving the results of fixed and random effects, I run hausman using the command 'hausman fe re'. STATA shows me the message 'chi2<0 ==> model fitted on these data fails to meet the asymptotic assumptions of the Hausman test; see suest for a generalized test'.
Conversely, if I run 'hausman re fe' , STATA generates results and shows that I should go with the random effects. I just want to know if there is a significant difference between 'hausman fe re' and 'hausman re fe'???
Can anyone help? Thanks in anticipation.

[Carlo Lazzaro]

Maham:
the problem you've posted is quite often reported on this forum.
It usually depends on the clash between the finite sample and the asymptotic properties of the hausman test.
You can easily search for similar posts (and related answers) by plugging in the key-word hausman in the top-right Search window.
The sequence

hausman fe re

is the way to go.
You can also try to repeat the hausman test with the -sigmamore- option (please, see -help hausman-) and check if the outcome is still foggy..
Eventually, you may want to google with the following string: -augmented regression AND vince wiggins- for a possible solution to your problem.
As an aside, I would also investigate if your problems are related to the fact that the default standard errors (SEs) for -xtreg- differ substantially from clustered-robust SEs, as this is a possible drawback of standard Hausman's test. A guidance on how to deal with this issue is reported in http://www.stata.com/bookstore/micro...ata/index.html (pages 267-268).

[Carlo Lazzaro]

Duplicate post. Please, follow this thread at http://www.statalist.org/forums/foru...-about-hausman

[William Lisowski]

-help hausman- has the information you need. Specifically see the sections titled Description and Remark: An alternative to hausman. In short, the order of the two results is crucial: one order is incorrect, and the correct order may well yield the diagnostic you show.

[Guest]

Lots of Thanks Carlo. That was very helpful.

[Guest]

Many thanks for your help. That was very helpful.

[Guest]

[CODE]
. hausman fe re,sigmamore
Note: the rank of the differenced variance matrix (4) does not equal the number of coefficients being tested (5); be sure this is what you
expect, or there may be problems computing the test. Examine the output of your estimators for anything unexpected and possibly
consider scaling your variables so that the coefficients are on a similar scale.
---- Coefficients ----
| (b) (B) (b-B) sqrt(diag(V_b-V_B))
| fe re Difference S.E.
-------------+----------------------------------------------------------------
primaryedu | -7.646653 -6.829918 -.8167354 .3944492
eduindex | 175.081 -126.8597 301.9408 259.0725
inequality | -5.163755 -5.06627 -.0974851 .0944555
inflation | .6494059 .6660607 -.0166548 .0603733
gdppc | -.0046073 -.0039503 -.0006571 .000766
------------------------------------------------------------------------------
b = consistent under Ho and Ha; obtained from xtreg
B = inconsistent under Ha, efficient under Ho; obtained from xtreg
Test: Ho: difference in coefficients not systematic
chi2(4) = (b-B)'[(V_b-V_B)^(-1)](b-B)
= 4.89
Prob&gt;chi2 = 0.2987


Hi Carlo!
I ran Hausman with sigmamore option ,and this time it advised me to go with random effects just like 'hausman re fe' command did. is it okay now? should I go with random effects now without worrying about any other issue?

[Marcos Almeida]

Hi, Maham,

We have to keep in mind that "order matters" whenever performing Hausman test.

In other words, we MUST insert fe before re.

There is no way out of this. Therefore, I fear the last lines of your above-written message are sort of confusing.

When using the sigmamore option, I gather we "force" (or "assert that") the difference between variances is positive definite.

By the way, homoskedasticity and positive definite differences of variance are among the main assumptions of the Hausman test.

On second thoughts, if you are still in doubt over the best model (why not?), I suggest you perform the Mundlak test.

Among the reasons, the Mundlak test copes with heteroscedasticity, serial correlation and robust estimations. Big deal, therefore.

On top of that, you will surely be able to compare the results of both tests and have "food for thought" so as to "explain" the reasons for choosing the final model.

Best,

Marcos

[Carlo Lazzaro]

Maham:
I would go -re-.
However, as Marcos echoed, the code sequence for the hausman test should be:

hausman fe re

.
Hence, postponing -fe- to -re- is not correct and any result is wrong as well.
By the way, you may want to consult any decent econometric textbook to see why things should be that way.
If you' re going to perform the Mundlak test in Stata, you can find an ad hoc user-written programme via -search mundlak-.

[Guest]

Lots of thanks for your suggestion. It was really helpful.

[S Mehta]

Hi everyone!

On running a hausman test with the panel data I have, stata shows the same message: 'chi2<0 ==> model fitted on these data fails to meet the asymptotic assumptions of the Hausman test; see suest for a generalized test'. I tried using all options given in 'help hausman' but it shows me the same message with every alternative. Moreover, I also tried 'suest' for which Stata says that xtreg is not supported by suest. Is there any other alternative to make hausman test work?

Thankyou

[Carlo Lazzaro]

S (as per FAQ, real names and surnames are preferred on this list. Thanks):
have you taken a look at: http://stats.stackexchange.com/questions/65650/hausman-test-for-panel-data-fe-and-re-error-in-the-estimation-what-to-do-sta^

[Steve Samuels]

"S. Mehta". Please re-register with your full name, as Carlo suggests. Use the CONTACT US button at the bottom right. You have asked a new question as a reply or answer to an old question. As a result most people will pass it by as they will see the original question, which has been answered, not our question.. I suggest you start a completely new topic.

[Sana Ayayda]

Hello Everybody,
I have the same problem, I received this message when I applied Husman test in stata, How can I solve this problem??
chi2<0 ==> model fitted on these
data fails to meet the asymptotic
assumptions of the Hausman test;
see suest for a generalized test
Any help would be appreciated!
Thanks in advance.