Employing changing status over time Closed population

[iwvining]

I am using MARK to estimate number of species in a stream. Field crews went out 5 times, with a different gear type each time to maximize species caught (some fishes will be more susceptible to one gear than another). For each gear, field personnel identified (without looking at catch) whether a specific gear was "likely to catch" a specific species or "unlikely" (1,0 for each gear for each species). I am interested in a single N (total number of species), and not an estimate of species richness for each gear type. For this reason, I wanted to geat susceptibility to each gear for the likely and unlikely groups, but a not a different species richness. Can these 1's and 0's be used as covariates, which change during each time? If I had no pre-knowledge of species suscptibility to different groups, I figured this would be a hetrogeneity model that varyed over time. This model is a bit more complex than I've run in the past, any ideas? Am I missing something?

[abreton]

Field crews went out 5 times, with a different gear type each time

Given that gear type was changed on each occasion (five gear types?), time and gear type are confounded. I suspect, for this reason, you're thinking about heterogeneity models, and specifically a 'mixture' model (developed by Pledger), to attempt to tease out variation in capture rates that are a function of time versus a function of gear type. In the MARK workshops at Colorado State University, it is often suggested that a minimum of 10 occasions are needed to get reliable/good results from mixture models. Five falls far short of that threshold. Even if these streams are exceptionally rich in fish species, I would think that only a two-part mixture would be possible but even this simple model will likely either cause MARK to crash or provide non-sense results.

"likely to catch" a specific species or "unlikely" (1,0 for each gear for each species)

To accommodate this scenario as an individual covariate --

00010 0 1 0 1 0;

In this encounter history, the species was caught/encountered only on pass 4 (00010) and it was "likely to [be captured]" on passes two and four -- 0 1 0 1 0 is a time-varying individual covariate and 0 and 1 are as you defined them for gear and susceptibility.

I wanted to [get] susceptibility to each gear for the likely and unlikely groups

If you used five different gear types then time and gear type (as I noted above) are confounded -- teasing the effects of the two apart may not be possible even with a two-part mixture model unless time-variation in capture always produced one of two capture probabilities, e.g., some passes with high capture probability near 0.8 and others with relatively low near 0.3. If time-variation caused capture to vary in a more complex way then variation due to gear type and time could not be teased apart with a two-part mixture. This complicates your desire to know "susceptibility to each gear for the likely and unlikely groups."

If you used just two gear types then time and gear are no longer confounded...and I could make other suggestions...as more gear types are deployed the issue of confounding increases...at five gear types the confounding is complete and we're back to my thoughts above.

andre