POPAN query - over-dispersion, under-dispersion and groups
Posted: Tue Jun 02, 2020 12:44 amDear RMark users,
I am analysing a CMR dataset on frogs (n = 10 repeat surveys each season, for 3 seasons) and I am interesting in comparing the super-population size N between each season to assess if there is a population increase or decrease per season. Therefore I am producing a POPAN model for each season and comparing N between the best models of each season. I am using sex as a group factor.
However, after using release.gof I found that one season is under-dispersed (~0.6). In the manual, underdispersion can be treated by either ignoring or using a c-hat adjustment as in the case of over-dispersion. I originally thought that not addressing under-dispersion would be crude so I tried adjusting N for underdispersion and I got really strange and small N estimates which are lower than Ni estimates (and also the amount of captures made).
This does not appear to be correct so perhaps ignoring it is the better approach. However this is unfortunate as QAIC adjusted for under-dispersion appeared to differentiate close models better than AIC. Any ideas what the best approach is?
Additionally, I am using popan.derived to calculate the combination of N of the groups. I noticed that the number change depending on if chat is adjusted in the model or not, which is convenient. However in the output, the combination of N of the groups does not come with 95% CI's whereas the combination of the groups per survey period does. Perhaps I am missing it in the output or maybe this obtainable with the output using the Delta-method?
Any help or advice would be greatly appreciated. Many thanks in advanced.
Kind regards,
Chad
I am analysing a CMR dataset on frogs (n = 10 repeat surveys each season, for 3 seasons) and I am interesting in comparing the super-population size N between each season to assess if there is a population increase or decrease per season. Therefore I am producing a POPAN model for each season and comparing N between the best models of each season. I am using sex as a group factor.
However, after using release.gof I found that one season is under-dispersed (~0.6). In the manual, underdispersion can be treated by either ignoring or using a c-hat adjustment as in the case of over-dispersion. I originally thought that not addressing under-dispersion would be crude so I tried adjusting N for underdispersion and I got really strange and small N estimates which are lower than Ni estimates (and also the amount of captures made).
- Code: Select all
##Estimate of N without chat adjustment
Real Parameter N
1
Group:SexF [b]17.64930[/b]
Group:SexM [b]35.81409[/b]
##Estimate of N with chat under-dispersion adjustment (chat=0.6)
get.real(ms3,"N",se=T)
all.diff.index par.index estimate se lcl ucl
N gF a0 t1 57 12 [b]4.649306[/b] 2.353316 1.723978 12.53847
N gM a0 t1 58 13 [b]9.814086[/b] 3.669627 4.715932 20.42359
This does not appear to be correct so perhaps ignoring it is the better approach. However this is unfortunate as QAIC adjusted for under-dispersion appeared to differentiate close models better than AIC. Any ideas what the best approach is?
Additionally, I am using popan.derived to calculate the combination of N of the groups. I noticed that the number change depending on if chat is adjusted in the model or not, which is convenient. However in the output, the combination of N of the groups does not come with 95% CI's whereas the combination of the groups per survey period does. Perhaps I am missing it in the output or maybe this obtainable with the output using the Delta-method?
Any help or advice would be greatly appreciated. Many thanks in advanced.
Kind regards,
Chad