Hausman test: Differenced variance matrix's rank not equal the number of coefficients

[Samy Gallienne]

Dear Listers,

I'm currently working with an unbalanced panel of 9000 individuals over a 10 years period.

Whenever I run my hausman test, I get the following error message:

Note: the rank of the differenced variance matrix (11) does not equal the number of coefficients being tested (12); 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.

I've tried to look for a clear explanation of what it means, so I can address it. Sadly, most of what comes up highlights error in code (random before fixed) rather than what is probably, in my case, a misunderstanding of econometrics. Then again, maybe my google-fu is too weak.

Here's my code:

// Regression 

xtset id date

* Fixed effects

xtreg ideology male age education revenue rural religion extroversion conscientiousness stability openness compassion politeness, fe 
estimates store fixed

* Random effects

xtreg ideology male age education revenue rural religion extroversion conscientiousness stability openness compassion politeness, re 
estimates store aleatoire


// Hausman Test

hausman fixed random

Description of my variables:

• ideology : 0 to 10
• male : 0 or 1
• age (years)
• education (completed)
• revenue (Euros)
• rural : 1 to 5
• religion : 0 or 1
• The rest are Likert scales :
  • extroversion
  • conscientiousness
  • stability
  • openness
  • compassion
  • politeness

Hopefully, you can help me understand my mistake here.

NB: I'm using Stata 13

[Samy Gallienne]

I've modified the revenue variable to ln(revenu+1) to control for aberrant values (it introduces a slight bias but it's negligible). Somehow, that fixed the error message I was receiving. While I'm happy to have solved my issue, I'd love to understand why it solved it.

[Phil Bromiley]

You didn't get a quick answer. You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output and sample data using dataex.

It is not necessarily a mistake per se. The test requires full rank of a matrix that may not be of full rank. You might look at user written xtoverid. Note that including a future value in a regression seems extremely questionable.