Huge difference in AIC values in different models

[anacjesus]

Hello everyone,

I am investigating survival rates, movements, site fidelities and recapture probabilities in a population of grey seals in Scotland identified using 3 different mark types. I have run 16 models in order to cover all the data separatelly by each mark type. I had no problem doing this but now I am wondering around interpretation of results.

In all 16 times, the best model had a delta QAIC of zero, which is great! But between the best models of each case I have a huge difference in the value of QAIC. After reading about it in MARK's book I attributed this difference to the amount of parameters and animals used for each run. However, some of them are very similar to each other and still have a huge difference (for example QAIC=1484 for 117 branded animals using multi-state model and QAIC=788 for 101 tagged animals using multi-state model).

I am thinking we could say that tagged animals data in this case is more fitted than branded animals, but can I make this assumption knowing that both of them have a delta AIC = 0?

Thank you!!

[Bryan Hamilton]

I'm not an expert at this so I could be wrong.

I don't think you can compare AIC between models using different data. AIC is really for model selection given the same data.

I think what you want to do is include all the data with marking method as a factor, then look at AIC and the beta estimates to see if incorporating marking method is supported.

[cooch]

I'm not an expert at this so I could be wrong.

I don't think you can compare AIC between models using different data. AIC is really for model selection given the same data.

I think what you want to do is include all the data with marking method as a factor, then look at AIC and the beta estimates to see if incorporating marking method is supported.

Bryan is entirely correct.

[anacjesus]

Thank you so much! I thought the exact same thing.