Birds in mixed-species flocks benefit from greater foraging efficiency and reduced predation but also face costs related to competition and activity matching. Because this cost-benefit trade-off is context-dependent (e.g., abiotic conditions, habitat quality), the structure of flocks is expected to vary along elevational, latitudinal, and disturbance gradients. Specifically, we predicted that the connectivity and cohesion of flocking networks would (1) decline towards tropical latitudes and lower elevations, where competition and activity matching costs are higher, and (2) increase with lower forest cover and greater human disturbance. We analysed the structure of 84 flock networks across the Andes and assessed the effect of elevation, latitude, forest cover and human disturbance on network characteristics. We found that Andean flocks are overall open-membership systems (unstructured), though the extent of network structure varied across gradients. Elevation was the main predictor of structure, with more connected and less modular flocks upslope. As expected, flocks in areas with higher forest cover were less cohesive, with better-defined flock subtypes. Flocks also varied across latitude and disturbance gradients as predicted, but effect sizes were small. Our findings indicate that the unstructured nature of Andean flocks might arise as a strategy to cope with harsh environmental conditions.

The file "Andeanflocks.csv" includes a comprehensive dataset of Andean mixed-species flocks surveyed in six countries, across latitudes (10°N to 41°S) and elevations (400 to 4000 m a.s.l.). It presents the species composition of 3676 independent flocks surveyed between 1976 and 2019 and organised into 84 independent sites across the Andes and adjacent lowlands (> 400 m), extracted from published and unpublished sources. 

In each study, a flock was defined as an aggregation of individuals of at least two species that move together while foraging in the original study. Similarly, all flock data across studies were collected using the ‘gambit of the group’ method, where all species observed within a single flock are assumed to be associating and are assigned reciprocal ties in the network. Thus, in our database, we do not include aggregations of frugivorous species which form in response to clumped resources and are not mobile associations.

We updated and standardised the taxonomy for all the studies following the most recent taxonomy from BirdLife International [2020]. 

For our analyses, we included only studies with ≥10 surveyed flocks per site to construct social networks, a minimum required to adequately describe a complete network when using gambit of the group sampling to measure fluid social interactions such as those among flocking species. If a study compared flocks at more than one site and/or across season, each site and season combination was included as an independent dataset following each author’s criteria (i.e. if the original author considered these as two independent datasets in the original manuscript, because of a complete turnover of species, for example). If coordinates and elevation information were provided for each flock, we used the averaged values to characterize the geographical location and elevation for each site. We present these average values in our dataset.

All analyses were conducted in R. 

The file "CodeMSF_final.R" includes all code required to replicate our analyses.