Network clustering of students based on answers to a questionnaire on refraction

Students’ answers to conceptual questionnaires are mostly analysed at the cohort level. However, most studies have not considered clusters of students within the cohort. This study uses newly published data [0] to analyse students' (N=1368, across five continents) answering patterns to a refraction questionnaire. Other student data include age, courseID, and type of programme. This study uses a network approach to find clusters: (A) we calculate the similarity [1] of each pair of student responses. This results in a weighted adjancency matrix, forming a 'similarity network'. (B) A backbone network is extracted [see: 2] to remove spurious links. (C) Clusters are found using community detection [see: 2]. The presentation will show results of analyses on how students clusters can be characterised in terms common answers and other student data. This addresses research aims:

1. How can student similarity clusters be characterised in relation to the context in which they study?

2. How can student similarity clusters be characterised in terms of particular answering patterns?

The results of this study could further inform instructional strategies developed in PER.

[0] Linder, C., et al. (2024). Relationship between semiotic representations and student performance in the context of refraction. PR-PER.<br type="_moz" />
[1] Lin, D. (1998, July). An information-theoretic definition of similarity. In Icml (Vol. 98, No. 1998, pp. 296-304).

[2] Brewe, E., et al (2016). https://doi.org/10.1103/PhysRevPhysEducRes.12.020131