Centrality and connectedness in student networks as predictors of academic achievement

The impacts of student networks (i.e., peer groups) on education outcomes (e.g., performance, dropping out college) have gained importance as a research subject these past years. This paper examines the effect of student networks’ structural components on academic achievement. Inconclusive literature invites us to compare the significance of four centrality indexes at the student level: the centrality in- and out- degrees, the Freeman’s betweenness centrality, the closeness centrality and the Bonacich Index. Furthermore, this comparison will allow distinguishing the effects of being central in the peer network (by means of the first three centrality indexes) versus being connected to central peers (by means of the Bonacich Index). On the other hand, at the network level, the roles of the size and density, which are also shown as controversial in the literature, are investigated. In this study, 574 freshmen college students were interrogated about their relations at university. Since we choose free recall and since the survey was not mandatory, students who did not participate could nevertheless be cited as ties. Social network analysis methods require the complete recording of interactions between actors belonging to the studied network. Consequently, in order to avoid losing valuable information by working with “complete-case analysis” (i.e., by deleting the nominations corresponding to students who did not respond to the survey), we use the Exponential Random Graph Models in order to impute ties for those nominated students that didn’t respond. These models are especially recommended in cases of directed graphs, with medium to large amounts of missing data, and of “no missing at random” observations. The final size of the sample was 867 students when these imputed ties were added. This paper researches a student’s own social network where many other studies use predetermined student networks or peer groups (e.g., roommates, course peers). These groups do not necessarily include the relevant peers that influence a student. In addition to asking students to nominate their relations themselves, this paper employs agglomerative hierarchical clustering algorithm in order to objectively identify the emergence of natural student sub-communities, on the basis of strong ties linking the students. The structural components of the network (i.e., centrality, size and density) will then be computed at cluster levels. Therefore, this research involves micro-units (i.e., the students) that are nested within macro-units (i.e., their own peer group or sub-community of belonging). Most researches dealing with micro- and macro- units use classic Ordinary Least Square techniques, which have not been designed for such cases. In order to predict achievement by features belonging to different analysis levels (e.g., centrality at the student level and density at the cluster level), we use hierarchical or mixed models, i.e., the most appropriate modelling technique in case of such complex structure of data. We expect that the methodological choices will improve the quality and robustness of the analyses, and contribute in a meaningful way to a thorough discussion about student networks impacts on performance. Finally, the results of this study might lead to student grouping policies that facilitate academic achievement.