Validity of methods for model selection, weighting for model uncertainty, and small sample adjustment in...

Area of Science:

• Ecology
• Population Dynamics
• Statistical Modeling

Background:

• Log-linear capture-recapture models are crucial for estimating population size.
• Model selection and sparse data present significant challenges in these estimations.
• Existing methods may yield biased population estimates due to uncertainty and data sparsity.

Purpose of the Study:

• To evaluate various model selection approaches for log-linear capture-recapture models.
• To assess a proposed small sample correction for multi-source capture-recapture data.
• To compare the performance of different information criteria and model types.

Main Methods:

• Compared Akaike's Information Criterion (AIC) and two Bayesian Information Criterion (BIC) formulations (Draper's and Schwarz's).
• Evaluated model weighting methods, the independent model, and the saturated model.
• Assessed a small sample correction for sparse cells in multi-source populations.
• Utilized 20 datasets from five different investigator groups.

Main Results:

• Akaike's Information Criterion (AIC) generally provided preferable estimates over BIC formulations.
• Draper's BIC formulation was slightly better than Schwarz's.
• Adjustment for model uncertainty offered marginal improvements.
• The small sample correction reduced bias only in the presence of sparse cells.

Conclusions:

• AIC is often preferable for model selection in log-linear capture-recapture analyses.
• The saturated model is optimal when confidence intervals are acceptable and interactions are absent.
• Most alternative methods resulted in population estimates that were too low.