Dear Nimble users,
My question is quite specific, so a lot of thanks in advance for people who will take time to give an answer.
I am working on the Bayesian Integrated Population Model developed in Gamelon, Nater et al. (2021) (Ecography, https://doi.org/10.1111/ecog.05738). This IPM has been built for wild boar females and is structured in body mass classes: small, medium and large. My aim is to explore the role of males on wild boar population dynamics. I have thus included males in the IPM code (please, find it in the attached file).
I first used the same datasets (e.g. CMR data, hunting data) for both males and females. I expected to obtain the same estimates of mortality probabilities and population sizes for the two sexes. We obtained estimates for females (population size, stage-specific mortalities) perfectly comparable to the ones published in the paper (Gamelon et al. 2021) last year. But for males, estimates are different from females. The temporal variation in all parameters seems the same for both sexes, but the mean estimates are not correct for males. For example, population size is underestimated for males (especially for medium and even more for large individuals), moreover, the probability to grow from medium to large (gPm[2]) and the probability to grow from small to large (gSLm) were lower in males than in females. In the same way, the probability of hunting mortality (sHm) was higher for medium and large males than for medium and large females. The probability of natural mortality (sNm) was higher for small and medium males than for females. However, population size for offspring and small are similar for males and females.
We tested a lot of things to understand why we obtained such a discrepancy between males and females:
- Change seed: we used 100 instead of 0;
- One matrix 6*6 for the two sexes: instead of having two different growth matrices (dim = 3*3), one for males and one for females, we wrote one single matrix of dimensions 6*6;
- A lognormal distribution for population size, instead of the truncated normal distribution;
- A sex-specific postnatal survival, so we added a parameter for postnatal survival of males, instead of having one common parameter for the two sexes;
- We tested the IPM with males only, everything worked perfectly; we tested the IPM for females only; everything worked. But when both sexes were included, males’ estimates became wrong;
After talking with Chloé Nater (many thanks for your help), we tested additional changes:
The bias is still present when running multiple chains and it is not a matter of convergence to alternative solutions;
Running the CMRR model separately gives parameters that are identical between males and females, the issue is not in the CMRR data/likelihood;
Unlinking the sexes (i.e. treating M and F as two independent female populations) also does not eliminate the discrepancies; it is not the link of the two populations through reproduction that causes the problem
We do not have any ideas of the issue; however, we suppose we are doing something wrong with NIMBLE. If you have a bit of time, I would be more than grateful if you could have a quick look at our code (in attachment). There is probably an obvious mistake that we cannot see.
However, I can not publish the data, so if you need to run the code, ask me and I’ll give you a part of this dataset and the code for the initial values.
Many thanks in advance for your valuable help.
Cheers,
Jessica Cachelou
--
You received this message because you are subscribed to the Google Groups "nimble-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to nimble-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/nimble-users/17114f7f-e8a2-439d-8302-1033d3a70814n%40googlegroups.com.
You received this message because you are subscribed to a topic in the Google Groups "nimble-users" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/nimble-users/olonWSd1Ayc/unsubscribe.
To unsubscribe from this group and all its topics, send an email to nimble-users...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/nimble-users/CABeTmqgG1MA2w372xE89mk2qdC33TZNN8ZMZvk1rE7cvHedKkA%40mail.gmail.com.