Modelling human and non-human animal network data in R using STRAND

Abstract

There have been recent calls for wider application of generative modelling approaches in applied social network analysis. These calls have been motivated by the limitations of contemporary empirical frameworks, which have generally relied on post hoc permutation methods that do not actively account for interdependence in network data. At present, however, it remains difficult for typical end-users—e.g., field researchers—to apply generative network models, as there is a dearth of openly available software packages that make application of such methods as simple as other, permutation-based methods.Here, we outline the STRAND R package, which provides a suite of generative models for Bayesian analysis of human and non-human animal social network data that can be implemented using simple, base R syntax.To facilitate ease-of-use, we provide a tutorial demonstrating how STRAND can be used to model binary, count, or proportion data using stochastic blockmodels, social relations models, or a combination of the two modelling frameworks.

BaYaka forager food sharing networks in the Congo Basin: The roles of gender homophily and kin sharing

“…We implemented a Bayesian latent network model that jointly assesses the factors that guide food sharing and the possible biases associated with self‐reported social networks (Redhead, McElreath, & Ross, 2022) using the STRAND package (Redhead, McElreath, & Ross, 2022; Ross et al, 2022) in R (v.4.0.4. ; R Core Team, 2017).…”

Friendship and partner choice in rural Colombia

“…To model how various individual-, dyadic-, and blocklevel covariates are related to the probability of network tie formations, we use a generalization of the Social Relations Model (Kenny and La Voie, 1984;Snijders and Kenny, 1999), which integrates block-level random effects (see Redhead, McElreath and Ross, 2022;Ross, McElreath and Redhead, 2022, for technical outlines and tutorials). Specifically, we estimate the probability of a directed tie between two individuals in the friendship network, 𝐹 , as a function of: sex, 𝑆, ethnic group, 𝐸, religious group, 𝑅, age, 𝐴, physical attractiveness, 𝑃 , BMI, 𝐵, years of education, 𝑈 , grip strength, 𝐺, reproductive success, 𝑄, log wealth, 𝑊 , RICH giving propensity, 𝑍, RICH leaving propensity, 𝐿, dyadic spatial distance, 𝐷, dyadic age distance, Ā, dyadic attractiveness distance, P , dyadic BMI distance, B, dyadic education distance, Ū , dyadic political opinion difference, 𝑂, dyadic log wealth distance, W , relatedness, 𝐾, food/money sharing ties, 𝑀, dyadic RICH giving, Z, and dyadic RICH leaving, L.…”

Multi-group analysis of grooming network position in a highly social primate

“…As the study of animal social networks is a rapidly developing field, especially regarding advances in statistical methodology, future studies on primate grooming networks could aim to include more state-of-the-art analytical approaches to investigate variation in grooming network measures. For example, besides using parametric models or permutation methods, generative models can be used to study social network data as well, as they account for the non-independence of social network data [ 89 ]. In addition, as sampling effort varies across groups, but also within groups (e.g., certain dyads might be observed more often than others and hence their grooming relationship might be more ‘certain’), uncertainty in network edges could be taken into account in downstream analyses.…”