3

My dependent variable is the decision of individuals for a value between 1 and 8 in an economic experiment. Every individual repeated that decision 20 times. I want to regress individual characteristics on that choice. I expect that the decisions of one individual are correlated with each other. So I need some kind of panel model.

Should I go for random effects or cluster-robust estimation? RE seems to be prevalent in the literature, but cluster-robust was a suggestion by my supervisor, but it was simply her intuition. What are the reasons for either? What should I pick?

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
asdir
  • 145
  • 2
  • 13
  • 3
    See [this answer by @Stask](http://stats.stackexchange.com/a/38430/1036) addressing this question. – Andy W Jun 28 '13 at 13:51
  • Thank you, Stask's answer is indeed helpful. What I don't understand, however, is the use of a Hausman specification test, as he suggests. According to Stata at least, a Hausman specification test is not appropriate for cluster-robust (i.e. vce(cluster)) regressions. (BTW: my exact specification is xttobit, if that makes a difference) – asdir Jun 28 '13 at 15:05

0 Answers0