August 2009
Intermediate to advanced
648 pages
17h 35m
English
Section 8.2.7 mentions overdispersion in the case of logistic regression. Overdispersion occurs when the observed variance is larger than the nominal variance for a particular distribution. It occurs with some regularity in the analysis of proportions and discrete counts. This is not surprising for the assumed distributions (binomial and Poisson, respectively) because their respective variances are fixed by a single parameter, the mean. Overdispersion can have a major impact on inference so it needs to be taken into account. Underdispersion also occurs. See McCullagh and Nelder (1989) and Dean (1998) for more detail on overdispersion.
One way to manage overdispersion is to assume ...
Read now
Unlock full access