Network measures for dyadic interactions: stability and reliability

Abstract

Social network analysis (SNA) is a general heading for a collection of statistical tools that aim to describe social interactions and social structure by representing individuals and their interactions as graph objects. It was originally developed for the social sciences, but more recently it was also adopted by behavioral ecologists. However, although SNA offers a full range of exciting possibilities for the study of animal societies, some authors have raised concerns about the correct application and interpretation of network measures. In this article, we investigate how reliable and how stable network measures are (i.e. how much variation they show under re-sampling and how much they are influenced by erroneous observations). For this purpose, we took a data set of 44 nonhuman primate grooming networks and studied the effects of re-sampling at lower re-sampling rates than the originally observed ones and the inclusion of two types of errors, "mis-identification" and "mis-classification," on six different network metrics, i.e. density, degree variance, vertex strength variance, edge weight disparity, clustering coefficient, and closeness centrality. Although some measures were tolerant toward reduced sample sizes, others were sensitive and even slightly reduced samples could yield drastically different results. How strongly a metric is affected seems to depend on both the sample size and the structure of the specific network. The same general effects were found for the inclusion of sampling errors. We, therefore, emphasize the importance of calculating valid confidence intervals for network measures and, finally, we suggest a rough research plan for network studies.

Establishing the reliability of rhesus macaque social network assessment from video observations

“…These differences may reflect variations in the quantity and frequency of behaviours and their respective distributions across dyads. Our findings complement prior literature investigating the effects of error, thresholding and small sample size on the reproducibility of dyadic associations (Croft et al, 2011; Handcock & Gile, 2010; James et al, 2009; Sundaresan et al, 2009; Whitehead, 2008; Whitehead et al, 2005), metrics representing global and local network properties (Haddadi et al, 2011; David Lusseau et al, 2008; Perreault, 2010; Voelkl et al, 2011) and social matrices. Our approach extends these methods by measuring error across an entire social structure and by predicting the amount of data necessary to derive reliable social networks.…”

“…Prior studies have validated multiple strategies for measuring the amount of error in the data acquired or for predicting the amount of error one may expect. Some studies have examined randomization of individual dyadic associations (Croft, Madden, Franks, & James, 2011; James, Croft, & Krause, 2009; Sundaresan, Fischhoff, & Dushoff, 2009; Voelkl, Kasper, & Schwab, 2011; Whitehead, Bejder, & Andrea Ottensmeyer, 2005) or compared the observed network to modelled networks (Handcock & Gile, 2010) to measure error at the level of individual associations. Others have measured the error of network properties, by using bootstrapping to establish confidence intervals for observed network metrics (Lusseau, Whitehead, & Gero, 2008), to determine the sampling rate required to capture variation in communities (Haddadi et al, 2011) or to examine the effect of sampling rate on network metrics from artificial networks (Perreault, 2010).…”

A multievent approach to estimating pair fidelity and heterogeneity in state transitions

“…The estimation of association rates in other types of dyadic interactions (such as associations between individuals in social network, Croft et al 2008), where the nature or strength of association is determined from field-based observations of marked individuals could also suffer from these same problems (e.g., Voelkl et al 2011). Therefore, if we want to robustly quantify rates of maintenance of pair bonds and other types of dyadic interactions, a modeling framework that incorporates and accounts for both imperfect and heterogeneous recapture rates, and problems connected with state assignment is needed.…”

Social Risk Dissociates Social Network Structure across Lateralized Behaviors in Spider Monkeys

“…Within Cytoscape, we used a variant of the "Kamada-Kawai Algorithm," a spring-embedded algorithm that forces connected nodes together while also forcing disconnected nodes away from the center [19]. We constructed weighted networks because this method is best suited for graphically representing the variation in social bonds [20,21]. All edges were weighted based on frequency of interaction with thicker edges denoting more interactions and thinner edges denoting fewer interactions.…”