Peer social networks in Czech lower-secondary classrooms

Published: 22-03-2021| Version 1 | DOI: 10.17632/5vzy6rykm7.1
Tomáš Lintner,


This dataset is based on a study aiming to identify aspects influencing formation of peer social networks with the use of exponential random graph modelling (ERGM). The results from the individual classrooms were pooled in meta-analyses using maximum likelihood estimation, which yielded estimates of the overall effects across the classrooms. The data come from a non-probability sample containing data from 435 ninth grade (ISCED 2A) 14 to 15-year-old students in 21 classrooms in 14 lower-secondary schools in Moravia Region of the Czech Republic, with data collected in November and December of 2017 as a part of a larger project - GA17-03643S. Standardized sociometric questionnaires designed for assessment of likeability between students in classrooms were used in primary data collection. Personal questionnaires were used to assign individual students the variables of SES (socioeconomic status) and gender. SES is represented by parents‘ highest occupational status using a three-class version of the ESeC - European Socio-economic Classification. This dataset contains: - adjacency matrices representing directed cross-sectional unweighted likeability and antipathy social networks (CLSxx - Likeability network - Matrix.csv, CLSxx - Antipathy network - Matrix.csv, and CLSxx - Attributes.csv) - visualizations of the social networks made with Gephi (CLSxx - Likeability/Antipathy network - Visualization.png) - convergence check plots from statnet (CLSxx - Likeability/Antipathy model A/B - diagnostics.pdf) - goodness of fit plots from statnet (CLSxx - Likeability/Antipathy model A/B - goodness of fit.pdf) - results from the meta-analyses (Results.xlsx) Likeability model A could be fit to all 21 likeability networks. It includes terms for SES and gender popularity (in-degree), SES and gender homophily (nodematch), reciprocity (mutual), transitivity (gwesp), SES and gender expansiveness (out-degree), and overall connectedness (edges). Likeability model B could be fit to 14 likeability networks only, as 6 of the networks did not converge. It includes all the terms from Likeability model A plus effects of preferential attachment (gwidegree), out-degree distribution (gwodegree), dyad-wise shared partners (gwdsp), and connectedness across two edges (twopath). Antipathy model A is the most complete specification, which could be to the highest number of antipathy networks - 18. It includes the same terms as Likeability model A. Antipathy model B could be fit only to 9 networks as 12 of the networks did not converge. Compared to Antipathy model A, it does not include the transitivity term, however, it includes preferential attachment and out-degree distribution terms. Antipathy model B is more complete and has slightly better goodness of fit compared to Antipathy model A; however, it is based on a smaller number of classrooms. ACKNOWLEDGEMENTS This dataset is based on primary data collected by: Anna Drexlerová Jakub Kychler Jana Navrátilová Martin Majcík