Resting Association and Social Network Data of Free-Ranging Dogs
Description
This dataset contains resting-association and social-network data from 26 groups of free-ranging dogs observed in West Bengal, India, between 2019 and 2023. The study included 131 individually identified dogs from groups containing 3–15 individuals. Data were collected across three biologically defined seasons: pre-mating, mating, and post-mating. A resting cluster was defined as two or more dogs resting within approximately one dog-body length of one another. Following the gambit-of-the-group approach, all dogs observed within the same resting cluster were treated as associated during that observation. Although observations were conducted within predefined three-hour time blocks, each calendar day was treated as one analytical sampling period. Dogs observed in the same cluster were recorded as together, whereas dogs observed on the same day in separate clusters were recorded as present during the same sampling period but not together. Dyadic association strength was quantified using the Half-Weight Index in SOCPROG 2.9. The dataset includes dyadic association values, individual identifiers and demographic attributes, group composition, season, and group-level demographic variables. It also contains individual network measures, including strength, eigenvector centrality, reach, weighted clustering coefficient, and affinity, with uncertainty estimates derived from 1,000 bootstrap replicates. Community assignments, modularity-related outputs, social differentiation, and permutation-based diagnostics of preferred or avoided associations are also included where applicable. These data support analyses of how dyadic sex composition, life-stage composition, group size, and male-to-female ratio relate to resting-association strength. They also support comparisons of individual network positions across sexes and seasons and analyses of the sex composition of the strongest dyadic associations. The dataset should be interpreted as representing resting co-occurrence rather than direct affiliative interaction. Shared use of preferred resting sites may contribute to observed association patterns.
Files
Steps to reproduce
1. Import the observation data and verify the individual identifier, group identity, sex, life stage, season, observation date, and resting-cluster identifier. Treat each calendar day as one analytical sampling period. Retain separate resting clusters observed on the same day as distinct groupings. 2. For each group and season, create a group-membership table in which dogs recorded in the same resting cluster are classified as observed together. Dogs observed on the same day in different resting clusters are classified as present in the same sampling period but not together. 3. Import each group-season membership table into SOCPROG 2.9. Calculate pairwise association strengths using the Half-Weight Index: HWI_AB = x/[x + yAB + 0.5(yA + yB)], where x is the number of daily sampling periods in which dogs A and B occurred in the same resting cluster; yAB is the number of periods in which both were observed but in different clusters; and yA and yB are the numbers of periods in which only A or only B was observed. 4. Export the HWI association matrices. In SOCPROG, calculate the correlation between true and estimated association indices, social differentiation, cophenetic correlation, modularity, and individual weighted-network measures: strength, eigenvector centrality, reach, clustering coefficient, and affinity. Estimate uncertainty using 1,000 bootstrap replicates. Run Monte Carlo permutations until the p-values for preferred or avoided associations stabilize. 5. Identify network communities using eigenvector-based modularity maximization. Evaluate community robustness using the community assortativity index with 1,000 bootstrap iterations. Network visualizations can be reproduced in Gephi 0.10.1 using HWI as the edge weight. 6. In R 4.2.0, combine the dyadic HWI data with dyad sex combination, dyad life-stage combination, group size, male-to-female ratio, group identity, and the identifiers of both dogs. Fit the GAMLSS model with HWI as the response; sex combination, life-stage combination, group size, and male-to-female ratio as fixed effects; and random intercepts for group identity, Dog A, and Dog B. Evaluate model fit using randomized quantile residuals and worm plots. 7. Compare individual network measures between males and females using Mann-Whitney U tests. Test seasonal differences using one-way ANOVA for strength, affinity, eigenvector centrality, and mean association rate, and Kruskal-Wallis tests for reach and clustering coefficient. Apply the Benjamini-Hochberg correction within each family of tests. 8. Identify the strongest dyad and the three strongest dyads within each group. Randomly permute sex labels within groups 5,000 times while preserving dyads and HWI values. For each sex combination, calculate the observed count, permutation mean, 2.5th and 97.5th percentiles, and two-sided p-value using the +1 correction.
Institutions
- Indian Institute of Science Education and Research KolkataWest Bengal, Kolkata