1

I am using panel data and I have to choose between fixed-effects and random-effects models.

I run the Hausman test, the H0 (i.e., the difference in the coefficients from the two models is not systematic) is rejected. Thus, I should use the fixed-effects.

I also run a second test, which is based on the Mundlak model. So, I run the Mundlak regression (i.e., a random-effects model where I also add the individual means of the time-varying characteristics) and I run an F-test on the extra coefficients (the coefficients of the individual means of the time-varying characteristics). I cannot reject the null hypothesis. Thus, I should use the random-effect model (although also the fixed-effect is unbiased, the random effect is more efficient).

I am puzzled. I test the same hypothesis in two different ways, obtaining two opposite results.

Andy
  • 18,070
  • 20
  • 77
  • 100
Fuca26
  • 795
  • 1
  • 9
  • 29
  • What happens when you run the Hausman test on the Mundlak model? – An old man in the sea. May 31 '15 at 10:45
  • I have not really been proceeding in that direction. You can look at an evolution of this question at: [link](http://stats.stackexchange.com/questions/100168/panel-data-within-group-estimate-individual-fixed-effects-retrieved). – Fuca26 Jun 13 '15 at 14:49
  • To answer your question, I have not used the Hausman test to choose between fixed-effects model and the Mundlak model. But, let me ask you a question: would it make sense? The estimates from both the fixed-effects and the Mundlak model would be unbiased (as far as I have understood), and I do not know which one is more efficient (I guess the Mundlak). What would one choose if the Hausman test rejected the H0? – Fuca26 Jun 13 '15 at 14:57

0 Answers0