Poster #122 - Predicting Friendship Nominations Using Character Traits: A Social Network Analysis

• In Event: 1-005 - Poster Session 01
  In Poster Session: PS 01 Section - Methods, History, Theory

Integrative Statement

Previous research has demonstrated that character attributes can predict how central students are in their classroom networks (Kinderman & Gest, 2018; Van Zalk, Van Zalk, Kerr, & Stattin, 2011). The present study aims to confirm these findings using a different statistical model that has not been applied in the developmental science literature; this model is called a hierarchical latent space model (HLSM; Sweet et al., 2013) and can be thought of as a combination of hierarchical linear models and social selection models.

This study introduces HLSMs and illustrates that HLSMs offer a valid alternative to other methods. HLSMs allow researchers to estimate the impact of covariates on network data, and they can determine to what extent character traits impact friendship ties. Further, these models accommodate multiple networks. We claim that HLSMs fill a void given that they accommodate cross-sectional data.

Latent space models (Hoff et al., 2002) assume that individuals have positions in a low-dimensional latent space; closer individuals in this space are more likely to have a tie between them, controlling for parameters in the model. These models are easy to interpret since ties are conditionally independent given these latent positions. See Figures 1 and 2 for an example of a HLSM for binary network data.

HLSMs are more appropriate for network data than HLMs or SEM approaches because they properly account for dependence among ties. In addition, HLSMs use cross-sectional data to fit models with only one round of data collection. The most common network model in developmental science research is SIENA (Snijders et al., 2008), which requires at least two waves of longitudinal data. HLSMs are comparatively easy to interpret.

To illustrate these models, we fit HLSMs to friendship network data collected in 38 fifth-grade classrooms in a racially and ethnically diverse suburban school district. Students nominated up to five classmates as friends in the fall and spring. These nominations created asymmetric friendship ties that we used as the outcome variable. Our data include ratings of popularity, leadership, rejection, aggression, and shyness, also collected across two semesters. Thus, our models include the sender’s (student nominating) ratings on these traits, the receiver’s (student being nominated) ratings on these traits, and a dyad-level variable indicating whether the students are the same gender; thus, we predict the impact of these covariates on friendship nominations.

Our findings are similar for the two time points: the receiver’s character traits predicted whether a friendship nomination was received (p<.05); students were more likely to receive friendship nominations from peers if they were perceived as being high in popularity or leadership, after controlling for the other variables. Conversely, students were less likely to receive friendship nominations if they were perceived as highly shy/withdrawn, aggressive, or rejecting/excluded. Students were more likely to nominate same-gender friends.

Our approach focuses on predicting ties without longitudinal data. These models could be a first step to SIENA models, allowing developmental psychologists to address preliminary research questions without collecting data across multiple time points.

Authors

• Casey Jonathan Archer, University of Maryland College Park
  Presenting Author

• Aditya Ravindra Bhat, University of Maryland College Park
  Non-Presenting Author

• Tracy M. Sweet, University of Maryland College Park
  Non-Presenting Author

• Kenneth H. Rubin, Department of Human Development & Quantitative Methodology, University of Maryland
  Non-Presenting Author