Consistent positive performance of "bk"parameter

[leopard1]

Dear Murray and forum users

I would be indebted to you if you could provide me with a little information on the “bk” (Animal x site response - site-specific step change) and “Bk” (Animal x site response - site-specific transient response) parameters used in the fitting of my models using conditional likelihood.

To give you a little background to my data:

I performed two 40 day camera-trap surveys of African leopards in Zululand South Africa 2012. The first was a traditional survey using 30 sites with paired cameras (40 captures of 15 individual leopards – 25 re-captures). The second survey followed the exact same layout except for the fact that I used a lure attractant. In this survey I recorded 39 captures of 14 individuals, also with 25 re-captures.

From previous forum entries (eg. Lynx lady and Lourens), it seems that my data indeed qualified for analyses in secr (although they were on the cusp regarding an appropriate number of re-captures ie. 25). Nonetheless based upon your example model in the secr guide as well as some examples used by Lourens in the forum I decided to go for the following options to model leopard density using conditional likelihood:

Base model: g0 + sigma modelled as constants
Density surface model: g0~h2, sigma~1
Sex-specific models: a) g0~Sex, sigma~1 b) g0~Sex, sigma~Sex

I however also used the behavioural parameters: “b”, “bk” and “Bk” as a) leopards are likely to be “trap happy/shy” and have varying number of cameras within their home-ranges therefore varying capture probability amongst individuals (Karanth et al., 2011) and b) these were recommended in your guide, and were cited to make a potential difference in density estimates. I modelled these as a function of g0, keeping the scale parameter constant.

My results using the above parameters ranged from 2.8-3.96 individuals/100 km2 for my passive survey while the lure survey provided an estimate of 3.23-3.59 individuals/100 km2. Although I am no expert, I believe this consistency and comparability between models is positive.

I however have one question, and this is regarding the “bk” parameter, I have now analysed a total of six surveys using the above modelling approach and in the majority of cases the “bk” – animal x site response (site-specific step change) variable has the strongest support using the Aikaike’s Information Criterion approach of model assessment.

Murray, I was wondering if you could please explain to me a little more about this parameter and why it seems to be consistently an important model within the suite of candidate models I have fitted.

Thank you so much for any correspondence you may forward me.
Alex

[murray.efford]

Alex
I have a bad habit of writing long replies. Phidot knows it is best to time me out and throw these away, as has just happened.
Tersely: I don't see your problem. My hypothesis would be that you would expect a b or bk effect int he lured trial and not in the other. If that was not the case, and you get a bk effect even when the leopards did not know they were being photographed, then presumably some other unmodelled effect (site-site detector differences?) was appearing as bk. The beta coefficient is a measure of the +ve or -ve effect of lure (bear people seem to expect a negative effect as bears learn it's not worth it).

[Also: Varying number of detectors per home range does not cause heterogeneity in SECR models - that idea is a hangover from a different era. Density is not 'modelled' when you fit by maximizing the conditional likelihood, and you do not have a density surface model.]

Murray