Allegedly, a key assumption of Cormak-Jolly-Seber analysis is that all individuals in the population have equal likelihood of being captured? If so, what does that mean in practice? A classical method of estimating carnivore population density and size depended on placing traps (culvert traps, snares, etc.) in likely locations, then capturing and recapturing individuals are appropriate intervals to estimate survivorship.
However, the number of traps usually tended to be small relative to the number of animals, and the locations where traps were set was likely to be near roads or trails for the convenience of trappers. As a result traps were most likely to trap animals with nearby home ranges. Furthermore, some age-sex classes may have disproportionately large home range sizes or do more migration between seasonal activity centers, and thus be more likely to encounter traps. These variations among individuals and subpopulations seem to me to violate the equal-catchability assumption -- in some cases so seriously that it substantially over- or under-estimates population growth rate or size. Is there a method for estimating the magnitude of error -- or at least how much this inflates confidence bounds?
It is my understanding that the use of uniform grids in hair-snare censusing is designed specifically to avoid such bias. Is that correct?
Another source of violation of equal catchability occurs if some individuals are "trap happy", whereas others are "trap shy". If one considers just mean catchability, as the percentage of trap-happy animals increases, this tends to reduce estimates of population size (the extreme case is were one always catches the same few individuals, as though they were the only members of the population.). The reverse is true for a rising percentage of trap-shy individuals. Yet, some simulations allegedly indicate that as heterogenity of catchability rises, this necessarily tends to under-estimate population size, no matter whether heterogeneity rises because of more trap-happy or more trap-shy animals. Is that true?
Does any know of good references on this which are open-access?
hi Stephen,
Edit: I just noticed I forgot to answer the primary question: yes, one of the key assumption of CJS model is that all individuals have the same capture/detection probability. Thus, in practice, you should have a representative sample of the targeted population in your trap.
Thus, the question about this kind of violation is fundamental in Capture-Recapture studies. One way to test for such problems in capture-recapture data is the Goodness-of-fit test ( Pradel et al 2003 Biometrics http://www.creem.st-and.ac.uk/olivier/Pradeletal2003Biometrics.pdf / Pradel et al 2005 Animal Biodiversity and Conservation http://abc.museucienciesjournals.cat/files/ABC-28-2-pp-189-2041.pdf). Some other studies investigate the effect of heterogeneity of detection in adult survival and population growth rate (Marescot et al 2011 Ecological Application http://bdm.typepad.com/files/marescot2011_ea.pdf / Fletcher et al 2012 Methods in Ecology and Evolution http://onlinelibrary.wiley.com/doi/10.1111/j.2041-210X.2011.00137.x/pdf). There is also a great paper about how to deal with trap-awareness in capture-recapture model when estimating survival for example (Pradel & Sanz-Aguilar 2012 PlosOne http://www.plosone.org/article/info%3Adoi%2F10.1371%2Fjournal.pone.0032666).
When possible, to account for heterogeneity in the population directly in the model is the best way to produce unbiased estimations. Otherwise, simulations can help to estimate the error linked to heterogeneity.
Hoping these papers could help.
Cheers,
Guillaume
I think the main source of heterogeneity is the situation between the center of activity of the animal and trap. I suppose using a spatial CJS would have no problem. I'm sure you can find this paper very interesting:
http://onlinelibrary.wiley.com/doi/10.1111/2041-210X.12134/abstract
Stephen, there are two ways that I'm aware of to deal with the heterogeneity in detection among individuals due to movement of animals in and out of the sampling area. The first is to model this movement mechanistically with a spatially explicit mark-recapture model or model it with a phenomenological approach such as modeling latent heterogeneity among individuals. The link that Jose provided is to an example of the mechanistic approach that models the movement and home range sizes of individuals. And there are publication of the application of hair snare traps with this model as you mentioned.
The second way is to create a study design that approximates a 'closed' system. In other words, create a study design that minimizes the bias associated with these assumptions violation. In theory, the larger the sampling area (trap grid) is relative to the home range of the animal, the less the bias. So if you have some information of the movement patterns of the animals you can use it in a simulation to determine the size of the sampling area that reduces the bias to an acceptable level. I did this for fish sampling in the following paper.
https://www.researchgate.net/publication/256196659_Evaluating_mark-recapture_sampling_designs_for_fish_in_an_open_riverine_system?ev=prf_pub
The number of traps relative to the population size should not bias the estimates, but it will have an impact on the precision of the estimates. The more traps the higher the over-all detection probability, and thus, the greater the precision. However the trap placement can result in bias estimates. It seems reasonably possible that animals caught in traps near roads may not have detection probabilities or movement patterns that represent the population as a whole. This is the problem with non-random sampling designs. However sometimes convenience sampling is the only way.
Hope this is helpful.
Article Evaluating mark–recapture sampling designs for fish in an op...
- Badges
- Science topic
- Similar topics
- Geoscience
- Asia
- India