We have a data-set with one categorical and several other continuous variables. After entering all these predictors in main effect model of Poisson regression , it was found that the deviance was zero.
Can deviance be calculated by including the categorical variable in negative binomial regression?
Hi,
Go through this link. I hope it helps.
Thanks
https://stats.idre.ucla.edu/stata/dae/negative-binomial-regression/
Any model is use, including Poisson and Negative Binomial, I use indicator variables. I assume you are referring to the calculation of an over-dispersion parameter. This is calculated regardless of the type of variables uses as your predictors. What software are you using for this analysis?
The Poisson distribution is the only probability distribution for which the mean and the variance are equal. For the negative binomial the variance is larger than the mean. Therefore it is naturally associated with over-dispersion. This is why Jason is referring to the the calculation of an over-dispersion parameter.
A deviance of zero implies goodness-of-fit or closeness to the saturated model of the fitted Poisson model. The negative binomial model may therefore not be as well fitted to the data. Calculation of the deviance on its basis may not be necessary.
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Software
- R