A simple way of assessing dispersion in models for count data
It has been some time since I have updated my blog, but I have plenty of good
excuses for justifying my lack of activity (or so I like to tell myself). In my post on simple models for abundance, I
introduced the Poisson-log GLM as the basic approach for
modelling count data. The Poisson-log GLM has an important limitation when it
comes to modelling ecological data: it assumes that the variance and the mean
of the response (dependent) variable are the same, so it is rather inflexible
for modelling response data with a variability exceeding the variation in the
mean. When this condition is not met, we obtain over- or under-dispersion in
the model. It is
important to note that the terms over- and under-dispersion do not refer to the
raw data values, but to the mean and variance expected with respect to a Poisson-log
GLM. It is a frequent mistake to think that dispersion refers to the raw
values of the response variable.
