Modeling behavioral effect as time trend

[jlaufenb]

Everything I have studied regarding behavioral closed population models (Mb) state that only the initial captures are used to estimate p and N. Recaptures are only used to estimate c, which is considered a nuisance parameter. White et al (1982) discuss this model and the effect that a simulated trap happy response has on the average capture probability over the course of 7 sampling periods (pg 57-58 ). Basically, the total number of animals captured per period and average capture probability would increase over time as the proportion of the population with a higher recapture probability would approach 1. Over a sufficient number of trapping periods, these relationships should be asymptotic. Considering that the behavioral model essentially "throws away" data (i.e., recaptures), could the average capture probability be modeled using a time trend covariate? The linear trend may not perform well because of the asymptotic nature of the process. Instead, would fitting a log or inverse squared transformation to the linear trend values be valid? Is this potentially a better way to model a behavioral effect because it would utilize all of the data? I'm concerned that there may be some structural issues that I'm not seeing, which would invalidate this approach. Any insight from those of you with more experience would be greatly appreciated.