A guide to null models for animal social network analysis.

Area of Science:

• Ecology
• Behavioral Ecology
• Network Science

Background:

• Null models are essential for hypothesis testing in social network analysis but are underutilized.
• Existing null model construction methods vary in effectiveness and test different hypotheses.
• Understanding and implementing appropriate null models is critical for accurate interpretation of animal social networks.

Purpose of the Study:

• To underscore the importance of null models for robust hypothesis testing in animal social network studies.
• To evaluate the performance of different statistical tests and null models when observation bias is present but unknown.
• To provide practical guidance and code for implementing effective null model approaches.

Main Methods:

• Simulated data with known observation bias was used to test statistical tests and null models.
• Permutation tests of raw observational (pre-network) data were analyzed for their ability to account for network structure.
• R code was developed to implement various null model routines across different data types.

Main Results:

• Permutations of pre-network data effectively controlled for underlying social network structure, reducing both Type I and Type II errors.
• Despite their effectiveness, pre-network data permutations are infrequently used due to implementation challenges with certain data types (e.g., focal follows, GPS tracking).
• The study provides accessible routines and R code to facilitate the application of these methods.

Conclusions:

• Pre-network data permutation is a powerful and recommended method for robust hypothesis testing in animal social network analysis.
• Overcoming implementation challenges can significantly enhance the reliability of findings in behavioral ecology.
• Widespread adoption of these permutation methods will advance the field by enabling more rigorous scientific inquiry.