Common datastream permutations of animal social network data are not appropriate for hypothesis testing using...

Area of Science:

• Behavioral Ecology
• Network Science
• Statistical Modeling

Background:

• Social network analysis is crucial for understanding animal social structures.
• Specialized statistical methods are needed due to data non-independence and confounds.
• Datastream permutations are commonly used to test null hypotheses in animal social networks, particularly with regression models.

Purpose of the Study:

• To evaluate the appropriateness of datastream permutations for testing hypotheses in animal social network analysis.
• To identify potential pitfalls associated with using datastream permutations in regression modeling.
• To propose alternative statistical approaches.

Main Methods:

• Simulations were conducted to assess Type I error rates of datastream permutations.
• Node-label permutations were used as a comparative method.
• The study analyzed the suitability of permutation methods for regression models in social network analysis.

Main Results:

• Datastream permutations exhibited extremely high Type I error rates (approaching 50%) when testing relationships between network structure and covariates.
• Node-label permutations yielded appropriate Type I error rates (~5%) in the same simulations.
• Datastream permutations do not accurately represent the null hypothesis of interest in these analyses.

Conclusions:

• Datastream permutations are inappropriate for testing hypotheses in regression models applied to animal social networks.
• Node-label permutations are a more suitable method for such analyses.
• Addressing data non-independence and observation reliability separately, followed by node-label permutations, is recommended.