# Zelig reports $R^2$ of a negative binomial regression - nonsense?

Using Zelig in R I fitted a negative binomial model to my data. The psychology APA standard demands to report the overall R squared, F-Value and p-value for the whole model.

I looked at the formula for R square and it is using sum of squares. Since the negative binomial model is not using least square method assuming a linear model, does it make sense here to report any R squared?

If not, could I argue in my paper that reporting R square is non-sense or not possible?

If yes, how can I calculate an equivalent R Squared, and how can I perform the F-Test to report the demanded numbers?

thanks

-
I think you 'rage' is justified, but you can't do much. If the only 'hammer' you have is a Gaussian distribution, then you have to fall to F-test, R-Sq, etc. It is definitely not non-sense to report $R^2$ if the CLT is satisfied, which may not be true in your case. It is good that you question such issues, instead of just using them. –  suncoolsu Sep 5 '11 at 0:03
add comment

## 2 Answers

Incidentally, F values also assume normal errors. I don't think these requirements were made with count data in mind. I'm not sure what to tell you. If apa requirements weren't an issue, I'd report something like proportion of explained deviance instead of R2, along with my regression coefficients and overdispersion parameter, and improvement (in deviance units or AIC) attributable to including different model terms.

Good luck!

-
Hi, yes I found this: ats.ucla.edu/stat/stata/output/stata_nbreg_output.htm It shows an example how to calculate McFadden's pseudo R-square using your mentioned deviance approach. –  user670186 Jun 6 '11 at 22:47
add comment

That requirement is...nonsensical. You should probably considering emailing the editor for guidance, or note your issue with reporting something like R-squared

-
add comment