Briefly, my understanding of the c-hat parameter is that it is used to adjust for overdispersion, most commonly in count data. Because overdispersed data often violates the assumption of independence required in most modelling approaches, the variances can be biased low (i.e. precision is too high) if this lack of independence is not accounted for. C-hat functions as a “variance inflation factor” to adjust the precision of the parameter estimates and when used along with QAIC can be useful for model selection. This might be useful for occupancy modelling if you suspect spatial correlation of some kind.
I’d recommend checking out the 2002 book by Ken Burnham and David Anderson on Model selection and multi-model inference.
Thanks for this Paul, in fact that last bit is my main suspicion, that there might be some kind of spatial correlation in my data set. I did had a look at Burnham and Anderson (2002), in fact, but its always useful to get soem extras. Thanks a lot!
Seems a balancing act, if not a paradox. The basic premise of Geography is that their objects are spatiality correlated. Therefore geographic variables are often counted/sampled randomly or systematically. The crux of the matter seems now how to delineate your geographic study area and how far out do you go from your study area in counting/sampling your variable(s) for comparison.