5

Possible Duplicate:
Poisson regression vs. log-count least-squares regression?

This post got me to thinking about when it makes sense to use poisson regression instead of log transforming a continuous non-negative outcome variable. I wondered if and when it might make more sense to log transform in some cases. (The Stata blog post has more detailed resources linked at the bottom).

Community
  • 1
Michael Bishop
  • 2,191
  • This is a duplicate of a prior question, see http://stats.stackexchange.com/q/8505/1036. If you look at the related questions on the right, there are a host of very similar ones as well (although unfortunately for this specific question there does not appear to be a real satisfactory answer). – Andy W Aug 25 '11 at 18:56

1 Answers1

2

The key distinction is the assumption about the relationship between the mean and the residual variance. With Poisson regression, the variance is equal to the mean. With ordinary regression, the variance is constant, independent of the mean.

Karl
  • 6,197