Misclassification models in Presence and R

[hf.hwa]

Hi - I've been running analyzes to examine false positive errors, specifically how these vary by observer experience level. I've been using a combination of Royle & Link's (2006) R code, Presence and the Chapter 12 Excel spreadsheet from the (excellent) tutorial exercises.

I've modified Royle & Link's R code to give me survey-specific detection and misclassification probabilities. My R code produces values for the probabilities and psi that match those I get from the Excel sheet & Presence. However, for some of the models, the AIC values are not the same between R, presence, and Excel.

To get to the bottom of this, I also tried using Royle & Link's data in both R and Presence & again, I get different AIC values. Using the R code and Royle & Link's BLJA dataset, I get the AIC value of 168.08 for the constrained model - same as that reported in the paper. When I run these data in Presence, I get an AIC of 443.21.

The R code models psi_p(.)_alpha=0, whereas Presence models psi_p(.). Shouldn't these be equivalent?

[darryl]

Is the difference in AIC and -2log-likelihood values between models the same though? It may be that there's an additional constant of proportionality being calculated in the likelihood for one or the other implementations even though they're the same biological model.
Darryl

[hf.hwa]

No, they are not. All approaches give the same values for the p's and psi's. Presence and Excel give the same answers for AIC, R gives a different answer for the psi_p(.)_alpha=0 model. Here's a summary of the results

Presence / Excel -2Log.L AIC Delta.AIC
psi_p(.)_alpha=0 368.59 376.59 0
psi_p(.)_alpha=0 386.94 390.94 14.35

R
psi_p(.)_alpha=0 368.59 376.59 89.32
psi_p(.)_alpha=0 279.27 283.27 0