Huggins closed captures or Otis closed captures model?

[grazielabiavati]

Hello,

I’m analyzing capture-mark-recapture data for 3 anuran species in Amazônia using robust design. At this moment, I’m trying to understand differences in population estimates (N) using Huggins closed captures and Otis closed captures model. I’ve used both model and my results ended up very different.

My question is about which model is the best or if there are important conceptual differences between them.

Thanks, Graziela

[Bill Kendall]

Two issues come to mind. First, if you are using the Pledger mixture models in your analysis Gary made changes recently which could cause differences between the two approaches in some cases. Second, if you are not using the mixture models, but your sample size is small, I believe the conditional (Huggins) approach tends to be more stable, and is therefore preferable.

[grazielabiavati]

Bill,

my sample is small but I'm a little confuse about these two options. What do you mean with stable? You said "conditional" and I think that it is my question. When I first read the MARK mannual I didn't understand this word. When I use Otis Closed Captures and I ask for Robust Design inside MARK, and I run my models openning PIM chart I have the oportunity to change N and it is not possible when I use Huggins. Is "conditional" related to that?

[Bill Kendall]

Conditional just means that the model conditions on being captured, and therefore does not include N in the likelihood. This is the Huggins approach. The likelihood including only p's and c's (and pi if you use the heterogeneity model) is maximized, and then N is estimated by dividing the probability of being captured at least once into the total number of animals captured at least once within the primary period. This estimate can be found by clicking on OUTPUT and making your way to the DERIVED PARAMETER option.

The Otis model includes N (or really the number of animals not captured during the primary period) in the likelihood as a parameter. That is why it is included in the PIM's. My comment is that it is less stable referred to my impression that for smaller sample sizes it is more difficult for MARK to converge to the "right" answer. I suggest you go with the Huggins option.

[gferraz]

I have a slightly different question related to Graziela's Amazon frog models. When she compared her results from the Huggins closed capture with those from a CJS model (colapsing secondary visits), she obtained qualitatively similar results for survival between primary periods: 2 species with very low survival and one with fairly high survival probability. Temporal variation in survival OR in capture probability does not seem to be particularly important (in terms of AIC). What is really important is the temporal survival in capture probability between secondary occasions in the Huggins model. Now, despite the qualitative similarities, if we look at confidence intervals of the survival estimates, the difference between high and low survival species comes out significant in the CJS but not significant in the Huggins' model.

My question is: could there be any obvious reason to trust one approach (CJS or Huggins) more than the other. When it comes to the survival comparison, we are not particularly concerned in the estimates of population size.

Gonçalo