Robust Inference from Conditional Logistic Regression Applied to Movement and Habitat Selection Analysis.

Area of Science:

• Ecology
• Quantitative Biology
• Statistical Modeling

Background:

• Conditional logistic regression (CLR) is standard for analyzing animal habitat selection and movement, but temporal autocorrelation in observations biases variance estimation.
• Generalized estimating equations (GEE) offer a method to correct these biases by clustering data, yet optimal clustering rules remain unclear.

Purpose of the Study:

• To establish the relationship between generalized estimating equations (GEE) clustering rules and their effectiveness in removing statistical biases in conditional logistic regression (CLR) variance estimation.
• To provide guidelines for robust statistical inference in animal movement and habitat selection studies.

Main Methods:

• Simulated longitudinal animal movement and habitat selection data with varying autocorrelation and individual observation counts.
• Evaluated the impact of different numbers of clusters on GEE variance estimator effectiveness.
• Assessed the utility of destructive sampling for increasing cluster numbers.

Main Results:

• Approximately 30 clusters were sufficient for unbiased and precise variance estimates in CLR parameters.
• Destructive sampling effectively removed bias when autocorrelation was present and individual heterogeneity was low.
• GEE demonstrated robustness across unbalanced datasets.

Conclusions:

• For studies with at least 30 individuals, assigning each individual to a cluster is recommended for GEE analysis.
• For studies with fewer individuals, destructive sampling is advised under specific conditions (temporal autocorrelation, weak heterogeneity).
• These findings support the development of reliable habitat selection and movement models with robust statistical inference.