Hi,
I have been trying to find an answer to this question without success. Maybe, some one can help me.
I have been fitting dynamic occupancy models. I tested the most complex model with a Goodness-of-fit test as recommended by (MacKenzie & Bailey 2004; MacKenzie et al. 2006) and obtained no evidence of lack of fit (p>0.05) and c-hat= 1.17. I know c-hat should be 1 and if it is >1 data is probably over dispersed. However, how much higher than 1 is evidence of over dispersion? I have not seen a single model (GLMs) with a c-hat value of just 1, and I am wondering whether or not I should conduct model selection with QAIC, as my c-hat value is pretty close to 1.
I will appreciate very much a response.
Ramiro
Trust the p value! But QAIC is useful too because you can compare goodness of fit directly and compare different models, and it incorporates c-hat. High c-hats have low QAIC and therefore fit your model more closely :-)
Thanks Andrew for your response. I guess I will try re-running the models using QAIC and see if that changes too much the model rankings using the 1.17 value of c-hat.
You're welcome. But I'd say there isn't over-dispersion :-)
Thank you Nareff for your suggestion. I really appreciate your time. Yeah, I am following the group and is being really helpful. However, I did not find an answer to how much is considered overdispersion. That is why I wanted to ask here. I guess, I feel more confortable in this forum :). Thank you again!
I agree that there is no overdispersion. Some softwares allow to insert the c-hat to adjust for the overdispersion, I do not know if yours does that? Even if you adjusted for overdispersion with C-hat=1.2, you would probably not change the results to any great effect.
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology