Data from: Social ties drive post-fission group choice in blue monkeys
Data files
Aug 19, 2025 version files 206.54 KB
-
GroupChoice_2025_Dataset1_InconsistentTies.csv
21.62 KB
-
GroupChoice_2025_Dataset1_Nodes.csv
1.25 KB
-
GroupChoice_2025_Dataset1_PostFissionNetwork.csv
21.62 KB
-
GroupChoice_2025_Dataset1_PreFissionNetwork.csv
21.62 KB
-
GroupChoice_2025_Dataset1_RankDistance_St.csv
37.64 KB
-
GroupChoice_2025_Dataset1_RankDistance.csv
22.23 KB
-
GroupChoice_2025_Dataset1_Relatedness.csv
22.26 KB
-
GroupChoice_2025_Dataset1_StrongConsistentTies.csv
21.62 KB
-
GroupChoice_2025_Dataset1_WeakConsistentTies.csv
21.62 KB
-
GroupChoice_2025_Dataset2.csv
6.62 KB
-
README.md
8.45 KB
Aug 19, 2025 version files 206.56 KB
-
GroupChoice_2025_Dataset1_InconsistentTies.csv
21.62 KB
-
GroupChoice_2025_Dataset1_Nodes.csv
1.25 KB
-
GroupChoice_2025_Dataset1_PostFissionNetwork.csv
21.62 KB
-
GroupChoice_2025_Dataset1_PreFissionNetwork.csv
21.62 KB
-
GroupChoice_2025_Dataset1_RankDistance_St.csv
37.64 KB
-
GroupChoice_2025_Dataset1_RankDistance.csv
22.23 KB
-
GroupChoice_2025_Dataset1_Relatedness.csv
22.26 KB
-
GroupChoice_2025_Dataset1_StrongConsistentTies.csv
21.62 KB
-
GroupChoice_2025_Dataset1_WeakConsistentTies.csv
21.62 KB
-
GroupChoice_2025_Dataset2.csv
6.62 KB
-
README.md
8.46 KB
Abstract
Permanent group fissions present rare opportunities for individuals in philopatric groups to select their groupmates. Studying post-fission group choice allows insights into how sociality influences animal decision-making and which social ties are important to group-living individuals. Our first analysis investigated which social ties influenced post-fission group choice in adult female blue monkeys by considering the strength and consistency of their ties with their original group’s members, as well as their dominance relations, relatedness to other female group members, and risk of infanticide. We used these dyadic and nodal characteristics in a separable temporal exponential random graph model to model edge persistence across two timesteps, before vs. after fission. Our second analysis used a conditional logit model to investigate the role of the original group’s resident male in a female’s post-fission group choice, assessing the strength of her tie to him and her vulnerability to infanticide. We present two datasets, one for each analysis. We also present the R code for the analyses reported in the associated manuscript.
https://doi.org/10.5061/dryad.xd2547drj
Permanent group fissions present rare opportunities for individuals in philopatric groups to select their groupmates, and so by studying post-fission group choice, we can gain insight into how sociality influences decision-making. Our study investigated which social ties and individual attributes influence post-fission group choice in blue monkeys by considering the strength and consistency of female’s ties to social affiliates, her relatedness to female peers, her relative rank, her vulnerability to infanticide, and her tie with the original group’s resident male. We found that females maintained different kinds of relationships during fission (those with social affiliates and the original group’s resident male). Here we present two datasets (and associated R code) used in this analysis, one (Dataset 1) including attributes of dyadic ties and nodal covariates that could influence tie persistence before vs. after fission, and the other (Dataset 2) related to an analysis of how the original group’s resident male influenced female post-fission group choice.
Description of the data and file structure
DATASET 1
Dataset 1 comprises eight matrices with dyadic covariates and one dataset with nodal traits for the individuals in the analysis, each in their own .csv file. In all seven matrix files, the columns and rows are identical in length (86 x 86) and composition, with cell values expressing an edge attribute for the dyad consisting of the row and column individuals. The first column and row indicate individual ID codes. ID codes are identical among all the matrices and in the nodal dataset. Some codes have a number “2” added at the end, indicating that a given female was a part of a second fission event, to differentiate between her dyads and attributes in one fission vs. another. Diagonal cells (same row and column individual) are undefined (NA).
The contents of the eight matrices are detailed below:
GroupChoice_2025_Dataset1_PreFissionNetwork.csv: This matrix represents the first time-step for the STERGM model and indicates which pairs of individuals co-occurred in a pre-fission group, with 1 representing such co-occurrence. NA was assigned when dyad members were not present in the same group, or when row and column individuals were the same.
GroupChoice_2025_Dataset1_PostFissionNetwork.csv: This matrix represents the post-fission network and the second time-step for the STERGM model. 1 indicates that a dyad from the same pre-fission group stayed together after fission, and 0 indicates that the dyad members joined different groups after fission. NA was assigned when dyad members were in different pre-fission groups, or when row and column individuals were the same.
GroupChoice_2025_Dataset1_StrongConsistentTies.csv: 1/0, with 1 indicating that the dyad had a DSI >1 for the two annual periods before the onset of subgrouping and 0 otherwise. This matrix thus expresses whether a dyad had a consistently strong tie prior to fission. NA was assigned when dyad members were in different pre-fission groups, or when row and column individuals were the same.
GroupChoice_2025_Dataset1_WeakConsistentTies.csv: 1/0, with 1 indicating that the dyad had a DSI <1 for the two annual periods before the onset of subgrouping and 0 otherwise. This matrix thus expresses whether a dyad had a consistently weak tie prior to fission. NA was assigned when dyad members were in different pre-fission groups, or when row and column individuals were the same.
GroupChoice_2025_Dataset1_InconsistentTies.csv: 1/0, with 1 meaning that the dyad had a DSI >1 for one annual period before the onset of subgrouping and a DSI < 1 for the other, and 0 otherwise. This matrix thus expresses whether a dyad had an inconsistent tie prior to fission. NA was assigned when dyad members were in different pre-fission groups, or when row and column individuals were the same.
GroupChoice_2025_Dataset1_Relatedness.csv: cell values in this matrix indicate the degree of relatedness between dyad members. Mother-offspring dyads had a relatedness value of 0.5, sister and grandparent-grandchild dyads had values of 0.25, aunt-niece dyads 0.125, cousin dyads 0.0625, and more distant relatives had a value of 0.03125. Unrelated individuals and individuals with unknown maternal relatedness had a value of 0. NA was assigned when dyad members were in different pre-fission groups, or when row and column individuals were the same.
GroupChoice_2025_Dataset1_RankDistance.csv: cell values in this matrix are integers indicating how many steps in the ordinal dominance hierarchy of the pre-fission group separated the individuals in the dyad. A small value indicates the two females in the dyad were similarly ranked, while a larger number indicates more discrepant ranks. We used z-score standardized rank values in the analysis, but these values are unstandardized. NA was assigned when dyad members were in different pre-fission groups, or when row and column individuals were the same.
GroupChoice_2025_Dataset1_RankDistance_St.csv: cell values in this matrix represent the z-scored rank distances provided in the matrix listed above.
GroupChoice_2025_Dataset1_Nodes.csv: this file contains the two nodal covariates used in the analysis, and the columns are as follows:
- ID: a unique identifier for females included in the models. Each ID here matches the ID codes in the seven matrices.
- atrisk: “atrisk” indicates a female was pregnant or had an infant <1 year old at the time of fission, “no” indicates she was not pregnant and did not have an infant.
- group: the pre-fission group ID for a given ID.
DATASET 2 (GroupChoice_2025_Dataset2.csv)
Dataset 2 is set up to evaluate how the presence of the original group’s resident male influenced a female’s post-fission group choice, but also includes attributes of the female herself. Each female in each fission event has two rows of data, one for each post-fission group she could have joined. Each row includes the following variables, listed from left to right:
ID: unique ID code for each adult female subject in the study.
choiceID: unique ID code linking a female's two rows of data per fission event. For a given fission event, the two rows with the same choiceID each represent one of the groups she could have chosen.
fission: unique ID code for each of the five group fission events.
postfissiongroup: unique ID code for the post-fission group represented in this row of data.
postfissiongroupchosen: unique ID code for the post-fission group the subject joined. This value is the same for both rows with the same choiceID.
groupchosen: coded 0/1, 0 if postfissiongroup does not match postfissiongroupchosen, or 1 if the entries do match. As such, this variable encodes information on whether the subject joined the post-fission group which this row of data describes.
resident: coded 0/1, 0 if the original group's resident male did not join the post-fission group recorded in postfissiongroup or 1 if he did join that group.
infanticiderisk: yes/no, with "yes" meaning the subject was at risk of infanticide at the time of fission, whereas "no" means she was not. A subject was considered to be at risk of infanticide if she had an infant <1 year old or was pregnant at the time of fission. This value is the same for both rows with the same choiceID.
resident_bs: the subject's tie strength with the original group's resident male. This value is the same for both rows with the same choiceID. Tie strength is expressed as a dyadic sociality index, as described in the Methods section of the accompanying publication.
Code/Software
R CODE (GroupChoice_2025_Code.R)
This file contains all the code used to generate the models and results in our manuscript. The code is written using the dataset and variable names described and provided here. In-line comments provide justification for each step of analysis. We ran this code in R version 4.4.1 and used the following packages: dplyr (v 1.1.4), tidyverse (v 2.0.0), mclogit (v 0.9.6), sna (v 2.8), network (v 1.19.0), networkDynamic (v 0.11.5), ergm (v 4.7.5), and tergm (4.2.1).
Observational data were collected from blue monkeys (Cercopithecus mitis stuhlmanni) in the Kakamega Forest, western Kenya surrounding five instances of group fission that occurred between 2008 and 2019. During the study period, trained observers monitored the different groups on a near daily basis, conducting focal animal samples on all adult females (classified as adults from the day they give birth to their first offspring). Focal animal samples were designed to last 30 min, and were retained in the dataset if they were at least 20 min long. On each day, females were chosen as subjects so that focal samples accumulated evenly across individuals and among different periods of the day (morning, midday, afternoon). Instantaneous recording of the focal subject’s behavior occurred at 1-minute intervals and included the identities of any social partners and individuals in proximity (within 1m). Agonistic interactions (with one individual showing submission) were recorded during focal samples and ad libitum.
We included females as subjects in the analysis if they were adults (so were the subject of focal sampling) for at least 6 months prior to the fission and survived to join a post-fission group. Tie strength, the basis for several variables in this dataset, was calculated using a dyadic sociality index (DSI) that combined the amount of time the individuals in a dyad spent grooming and in proximity (<1m), adjusted by the amount of time each individual in the dyad was observed during focal animal sampling (see manuscript for full explanation of this calculation). We aggregated all agonism data in the calendar year leading up to fission to calculate the original (pre-fission) group's dominance hierarchy using the I&SI method as implemented in Domicalc (Schmid and de Vries 2013). We considered a female to be at risk of infanticide if she was pregnant or had an infant younger than a year old at the time of fission. We identified a female as pregnant by back-counting one mean gestation length (176 days) from when the female gave birth.
We quantified ties as strong if a dyad’s DSI was > 1, and consistent if a dyad’s DSI was either > 1 or < 1 for both of the two annual periods prior to the onset of fission. Ties were categorized as “consistently strong” if DSI >1 for both annual periods prior to the onset of fission, and “consistently weak” if DSI < 1 for both periods. Ties were categorized as “inconsistent” if DSI > 1 in one annual period and <1 in the other.
We conducted two analyses. First, we ran a separable temporal exponential random graph model to evaluate post-fission group choice, including the strength and consistency of female’s ties to affiliates (3-way classification), relatedness to female peers, her vulnerability to infanticide, and her relative position in the dominance hierarchy as predictors. We also included interaction terms: (1) female infanticide risk x relatedness, (2) binary variable representing whether a dyad had a consistently strong tie x relatedness, and (3) a binary variable representing if a dyad had a consistently weak bond x relatedness. The second analysis used conditional logit models to evaluate how post-fission decision-making was influenced by a female’s tie to the original group’s resident male. We created two datasets, one comprising females at risk of infanticide at the time of fission and other comprising females not at risk. We ran a model on each dataset that included a binary predictor, expressing whether the original group’s resident male was present in each of the daughter groups. We also included an interaction term between this variable and a covariate that measured the strength of a female’s tie (DSI) to the original group’s resident male. All analyses were conducted in R version 4.4.1 (see code).