Hierarchical modeling of sequential behavioral data: an empirical Bayesian approach.

Area of Science:

• Behavioral science
• Statistical modeling
• Psychology

Background:

• Traditional methods for measuring behavioral contingency, such as the binomial z score and adjusted cell residual, have limitations.
• Accurate measurement of behavioral sequence strength is crucial for understanding interaction dynamics.

Purpose of the Study:

• To introduce a new, more robust approach for measuring the strength of contingency between behaviors in sequences.
• To address the limitations of existing statistical methods in behavioral sequence analysis.
• To present hierarchical models for analyzing behavioral sequence properties.

Main Methods:

• Review of common contingency measures (binomial z score, adjusted cell residual) and their limitations.
• Development of a novel approach utilizing log odds ratios and empirical Bayes estimation within hierarchical models.
• Application of hierarchical models to test for stationarity, homogeneity, and covariate effects in behavioral sequences.

Main Results:

• The proposed log odds ratio and hierarchical modeling approach overcomes limitations of traditional methods.
• Hierarchical models effectively test for stationarity, homogeneity, and covariate influences on behavioral sequences.
• The method was successfully applied to observational data from couple interactions.

Conclusions:

• Log odds ratios and hierarchical modeling provide a powerful and flexible framework for analyzing behavioral sequences.
• This new approach offers enhanced capabilities for understanding complex behavioral interactions.
• The methodology is applicable to various fields studying sequential behavior patterns.