Application of cluster analysis to identify different reader groups through their engagement with a digital reading supplement

Ma, Yawen and Ushakova, Anastasia and Cain, Kate (2024) Application of cluster analysis to identify different reader groups through their engagement with a digital reading supplement. Computers and Education. ISSN 0360-1315 (In Press)

Text (MaCainUshakovaC&E2024) - Accepted Version
Available under License Creative Commons Attribution Non-commercial No Derivatives.
Download (0B)

Text (MaCainUshakovaC&E2024) - Accepted Version
Available under License Creative Commons Attribution Non-commercial No Derivatives.
Download (0B)

Text (MaCainUshakovaC&E2024) - Accepted Version
Available under License Creative Commons Attribution Non-commercial No Derivatives.
Download (0B)

Text (MaCainUshakovaC_E2024) - Accepted Version
Restricted to Repository staff only until 1 January 2040.
Available under License Creative Commons Attribution.
Download (0B)

Text (MaCainUshakovaC_E2024) - Accepted Version
Restricted to Repository staff only until 1 January 2040.
Available under License Creative Commons Attribution.
Download (0B)

Text (MaCainUshakovaC_E2024)
MaCainUshakovaC_E2024.pdf - Accepted Version
Restricted to Repository staff only until 1 January 2040.
Available under License Creative Commons Attribution.
Download (1MB)

Abstract

The focus of this study is the identification of reader profiles that differ in performance and progression in an educational literacy app. A total of 19,830 students in Grade 2 from 347 Elementary schools located in 30 different districts in the United States played the app from 2020 to 2021. Our aim was to identify unique groups of readers using an unsupervised statistical learning technique - cluster analysis. Six indicators generated from the students’ log files were included to provide insights into engagement and learning across four different reading-related skills: phonological awareness, early decoding, vocabulary, and comprehension processes. A key aim was to evaluate the implementation and performance of Gaussian mixture models, k-means, k-medoids, clustering large applications and hierarchical clustering, alongside provision of detailed guidance that can benefit researchers in the field. K-means algorithm performed the best and identified nine groups of readers. Children with low initial reading ability showed greater engagement with code-related games (phonological awareness, early decoding) and took longer to master these games, whereas children with higher initial ability showed more engagement with meaning-related games (vocabulary, comprehension processes). Our findings can inform further research that aims to understand individual differences in learning behaviour within digital environments both over time and across various cohorts of children.