最近也遇到这个问题，逛了下论坛，说是因为算出来的hausman test的统计量为负，而这个统计量应该服从卡方分布，所以软件把Hausman statistic set to zero。具体为什么会出现这种情况还弄不清楚，也不知道真么解决，求指教
供参考：http://www.google.com.hk/search?sclient=psy&hl=en&safe=off&source=hp&q=Cross-section+test+variance+is+invalid.+site%3Aforums.eviews.com&aq=f&aqi=&aql=&oq=&pbx=1&cad=b&cad=cbv

Warnings & Interpretation Hausman Test
by Emily on Fri Apr 02, 2010 3:02 am

Hi,

I want to use the Hausman Test in order to check whether my panel data allows to use random effects models instead of fixed effects models. I know how to perform the test, I just have difficulties with the interpretation of the output and the warnings that I get. I already searched the Eviews User Guide and this forum, but I am still confused and I hope that you can help me. My Hausman output looks like this:


Test Summary Chi-Sq. Statistic Chi-Sq. d.f. Prob.

Cross-section random 0.000000 16 1.0000

* Cross-section test variance is invalid. Hausman statistic set to zero.
** WARNING: robust standard errors may not be consistent with
assumptions of Hausman test variance calculation.

My questions are:
1) Does the warning mean the test is invalid?
2) If "yes" to question 1), what can I do?
3) May I use random effects panel models?

One maybe important information: I use White (diagonal) robust standard errors. I use Eviews version 6.

Thank you very much and enjoy your easter holidays!Emily
  
Posts: 1
Joined: Fri Apr 02, 2010 2:29 am Top
--------------------------------------------------------------------------------
Re: Warnings & Interpretation Hausman Test
by EViews Gareth on Fri Apr 02, 2010 8:08 am

Essentially what this is saying is that the estimate of the cross-section effects indicates that there are, in fact, no cross-section effects (i.e. they have no variance). This means that testing for the difference between random effects and fixed effects may not be a valid thing to do (if your best guess is that there are no effects, how can you test what sort of effects they are?).

There isn't really a "fix" for this, nor should you really think of it as a problem.

我也跪这上面了。。。求指导。。。。。。

没说明白

楼上的英文书信已经说得很明白了，就是说模型根本就不是变截距的，所以也就不用考虑是固定效应还是随机效应了。