Measurement of species associations in mixed-species bird flocks across environmental and human disturbance gradients

In a changing world, it is important to understand how species interactions, such as those among species in mixed-species animal groups, are impacted by human activity. New techniques for measuring associations, in particular social networks and null simulations, have been applied to mixed-species group analyses. Unanswered questions include: (1) How do these methods compare to each other and to the traditional frequency table approach? (2) How similar are associations for the same pair of species in different habitats, such those with different disturbance histories or at different altitudes? (3) What traits of species influence these association strengths? Using data from an intensive study of mixed-species bird flocks in Sri Lanka, we used Mantel tests to compare associations calculated through three methods: the phi coefficient, social networks, and null simulations. Results from the three methods varied (0.53 < r < 0.75), but converged when out-of-flock data were added to the phi coefficient and social network methods (0.75 < r < 0.94). Correlations between the associations in different habitats were uniformly low (all <0.50) across all methods. These results, when combined with the fact that only a small percentage of the associations were statistically significant (9–14%, depending on the method used), suggest that on average, birds may not have strong preferences for other species in flocks. The trait analysis, conducted using the Multiple Regression Quadratic Assignment Procedure found that species that are more similar to each other in their body mass and diet had higher association strengths. This result was similar to what has been found before in global meta-analyses; however, contrary to earlier reports, in this study association strength increased with increasing phylogenetic distance. The strength of these results decreased when non-statistically significant species associations were included in these analyses or out-of-flock data were excluded, again emphasizing that the method of calculating species associations can influence conclusions. Nevertheless, it is clear that species associations in avian mixed-species groups are not greatly consistent across natural or anthropogenic gradients, although they are influenced to a certain extent by similarities in species traits, especially for species that show statistically significant associations.