Dear List,
I am attempting to use negative binomial regression (NBR) with maximum likelihood estimation (estimating the overdispersion statistic) and I'm not clear on how to deal with the results I have. My variable is definitely overdispersed with many zeros, but the SPSS NBR output shows a Pearson's/df dispersion ratio of that is under 1 (.738) suggesting underdispersion. In contrast, in the parameter estimates box the dispersion statistic (Negative Binomial B value) is 11.757 so much greater than 1, suggesting extreme overdispersion. I have no idea how to deal with this. I followed something that suggested re-stimating the model setting the scale paramater method as "Deviance" rather than fixing it at 1, and when I do this the Pearson's/df dispersion ratio statistic, and overdispersion statistic do not change, but the p values of the parameter estimates are now smaller so more significant, so i'm not sure what this has gained, if anything.
Is there a way to correct for overdispersion? I thought about running a zero-inflated negative binomial regression, but can't find anything anywhere about if or how this can be done in SPSS. If anyone can offer any advice i'd really appreciated it as there doesn't appear to be much out there in terms of how to handle this in SPSS.
Kathryn
Locked
