large DeltaAIC values

questions concerning analysis/theory using program MARK

large DeltaAIC values

Postby LisenkadeVries » Tue Jun 02, 2020 2:54 am

Posted for Maja

Hi

I would appreciate your opinion/advice on the following:

I have a very large dataset (21 occasions, 19158 individuals) and am using Burnham’s model for joint live-dead analysis. My results table shows very large AICc values and very large differences in AICc-values (these are the 10 best models):
Code: Select all
 
         model                                 npar  AICc   DeltaAICc  weight  Deviance
78   S(~time)p(~ageclass)r(~time)F(~sex)        47  118320.2   0         1     34072.4
76   S(~time)p(~ageclass)r(~time)F(~ageclass)   47  118666.6   346       0     34418.9
75   S(~time)p(~ageclass)r(~1)F(~sex)           28  118697.6   377       0     34487.9
73   S(~time)p(~ageclass)r(~1)F(~ageclass)      28  119032.4   712       0     34822.7
3    S(~ageclass)p(~ageclass)r(~1)F(~sex)       10  119182.4   862       0     35008.7
6    S(~ageclass)p(~ageclass)r(~time)F(~sex)    29  119347.5   1027      0     35135.8
77   S(~time)p(~ageclass)r(~time)F(~1)          45  119362.7   1043      0     35118.9
49   S(~sex)p(~ageclass)r(~1)F(~ageclass)       10  119540.4   1220      0     35366.7
54   S(~sex)p(~ageclass)r(~time)F(~sex)         29  119567.0   1247      0     35355.3
52   S(~sex)p(~ageclass)r(~time)F(~ageclass)    29  119618.6   1298      0     35406.9


Should I worry about these large values?
And what could be causing them?
I have already tried using new initial values for the second best model, but the results are the same, so lack of convergence does not seem to be the problem (moreover nearly all differences between AICc values are large).
I haven’t corrected the AICc values for the median c-hat yet, but the first rough calculations of the median c-hat showed only little overdispersion (c-hat ~ 1.1-1.2).

Many thanks,

Maja
LisenkadeVries
 
Posts: 4
Joined: Tue May 29, 2018 4:39 am

Re: large DeltaAIC values

Postby simone77 » Tue Jun 02, 2020 10:21 am

I don't see any problem with the AICc, their absolute values are meaningless (they may even be negative). Based on the information you have provided and on the assumption that there are no errors in the design of the models, the large difference in AICc between the models indicates (i) a large difference between male and female fidelity and (ii) a (less so) large temporal variation in the probability of recovery (cf. DeltaAICc between the 1st and 7th models and between the 1st and 3rd models). If you correct the AICc for the c-hat, these differences should decrease a bit because the most parameterized model (1st) will pay a higher penalty (because of the larger number of parameters) than the ones you compare with (e.g. 3rd and 7th).

A side observation: I think a question like this would go better to the "analysis & design questions" section.
simone77
 
Posts: 197
Joined: Mon Aug 10, 2009 2:52 pm


Return to analysis help

Who is online

Users browsing this forum: No registered users and 21 guests