Model selection and model robustness for population size estimation in 2-sample capture-recapture studies.

Area of Science:

• Ecology
• Epidemiology
• Statistics

Background:

• Two-sample capture-recapture studies are prevalent in ecological and epidemiological research.
• The Lincoln-Petersen estimator is the most common analysis method, particularly in epidemiology.
• Limitations exist in current methods when dealing with complex capture probability factors.

Purpose of the Study:

• To evaluate the performance of the Lincoln-Petersen estimator against Huggins' and Pledger's closed-population methods.
• To assess the effectiveness of Akaike information criterion (AIC) for model selection under various conditions.
• To investigate the impact of model misspecification on population size estimates.

Main Methods:

• Comparative analysis of Lincoln-Petersen, Huggins' conditional likelihood, and Pledger's likelihood methods.
• Simulation study to assess model performance with time, behavioral effects, and heterogeneity in capture probabilities.
• Evaluation of AIC's model selection capability with limited capture occasions.

Main Results:

• The examined closed-population models demonstrated a lack of robustness to model misspecification.
• AIC was unable to accurately select the correct model when only two capture occasions were available.
• Model misspecification significantly affected the accuracy of population size estimates.

Conclusions:

• Standard capture-recapture models, including the Lincoln-Petersen estimator, are sensitive to misspecification.
• AIC is not a reliable tool for model selection in capture-recapture studies with only two occasions.
• Careful consideration of potential misspecification is crucial for accurate population estimation in ecological and epidemiological studies.