Hybrid pairwise likelihood analysis of animal behavior experiments

Area of Science:

• Ethology and behavioral ecology
• Statistical modeling in biology

Background:

• Tournament experiments are crucial for studying animal behavior and aggression.
• Paired comparison models are standard for analyzing these experiments.
• Dependence between fight outcomes complicates traditional statistical analysis.

Purpose of the Study:

• To develop statistical models accounting for between-fights dependence in animal contests.
• To introduce a robust method for analyzing tournament experiment data with correlated outcomes.

Main Methods:

• Discusses two model specifications for between-fights dependence.
• Employs a hybrid pairwise likelihood method for model fitting.
• Iterates between estimating equations and pairwise likelihood inference.

Main Results:

• The hybrid method requires only means and covariances, simplifying complex analyses.
• Successfully applied to real-world data on Cape Dwarf Chameleons.
• Simulation studies validate the proposed methodology.

Conclusions:

• The developed statistical approach effectively handles dependence in animal fight data.
• Offers a flexible method applicable even when joint distributions are intractable.
• Enhances the statistical rigor of analyzing animal behavior in tournament settings.