Right skewed distribution of dep var: What method to use?

[Stef Bruck]

Dear all,

I have a dependent variable measuring the expectations for job change on a range from 0% to 100%. Most individuals indicate an expectation of 0%, while there exist a critical mass at values of 50% and 100%. The distribution of the dependent variable is atypical (see figure below) and I was wondering what estimation method I should use to address the distribution properly. I was thinking about using Negative Binomial or Poisson regression. Since I am not sure if this is the right method, I would be grateful for any help.

Thank you all in advance!

[Nick Cox]

As it’s bounded at both ends Poisson and negative binomial don’t seem very appropriate. I would check out -fracreg- and -betareg-.

[Clyde Schechter]

Just a pedantic point: the skew of the distribution strikes me as less a problem than the big spike at zero. With even a moderate size sample, the level of skew here is small enough that the central limit theorem would leave you OK to even use OLS linear regression here. But you'd need a very large sample to overcome the spike enough to get away with that. I think Nick Cox 's suggestions in #2 are right on target. Another possibility would be two-stage modeling: a binomial model for any expectation of job change vs none, and then a simple linear model of the magnitude of expectation of job change among those with any.

[Carlo Lazzaro]

Stef:
I'm probably late to the party, but I propose some (hopefully useful) references notwithstanding:
A comparison of zero-inflated and hurdle models for modeling zero-inflated count data | Journal of Statistical Distributions and Applications | Full Text
and John Mullahy 's Specification and testing of some modified count data models - ScienceDirect