I am struggling to understand overdispersion and when to check for overdispersion. Can you tell me in which case I should or should not check for overdispersion? And maybe if there is a good book chapter explaining that as well...

Can you correct/confirm this:

For binary GLMs-> no

For binomial but not binary GLMs-> yes

For Poisson GLMs -> yes

For linear models-> yes


Overdispersion means more (higher) dispersion than assumed by the model so is a concept that is relative, it depends on the model used. Many (most) models do not assume anything about the dispersion (variance), for instance linear models, which have one (or sometimes more) parameters modeling the variance directly. In such cases (linear models and generalizations anova, linear mixed models, multilevel models ...) there is no possibility of overdispersion, because the variance is modelled directly and independently from the mean.

Overdispersion typically is a possibility when the model do not have a parameter which is directly modeling the variance. The most common cases is binary models like logistic regression where the variance is a function of the mean. The same applies for models based on the multinomial distribution. The other main case is count data models like poisson regression, since the poisson distribution only have one parameter, and the variance equals the mean. But overdispersion is also possible for negative binomial models.

But the main thing to watch out for is models which do not model the variance directly by its own parameter(s).

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.