To split behaviour into bouts, log-transform the intervals

Area of Science:

• Animal behavior analysis
• Ethology
• Quantitative biology

Background:

• Accurate estimation of bout criteria is crucial for analyzing bouted behaviors.
• Current methods using log-transformed cumulative frequency distributions lack biological meaning for feeding behavior analysis.
• A previous study proposed modeling the frequency distribution of log-transformed interval lengths.

Purpose of the Study:

• To test a proposed method for estimating bout criteria in animal feeding behavior.
• To evaluate the biological meaningfulness and flexibility of log-normal models for interval data.
• To investigate the potential inclusion of drinking intervals in feeding bout analysis.

Main Methods:

• Modeling the frequency distribution of log-transformed interval lengths between feeding events.
• Utilizing a large dataset of 35,171 feeding intervals from 38 cows across three dietary groups.
• Fitting double and triple log-normal models to interval length distributions.

Main Results:

• Initial application of a two-Gaussian model showed limitations for some individuals.
• Incorporating a third log-normal component significantly improved model fit, suggesting drinking intervals.
• Fitting double or triple log-normal models yielded meaningful meal criteria for all individuals.
• Log-normal models demonstrated superior biological meaningfulness and flexibility compared to log (cumulative) frequency models.

Conclusions:

• Log-normal models provide a more biologically relevant and flexible approach to defining feeding bout criteria.
• The inclusion of a third log-normal distribution effectively accounts for intervals that may include drinking behavior.
• This refined methodology enhances the quantitative analysis of feeding patterns in animals.