School Entry Detection of Struggling Readers From Gameplay Data

• In Event: Bedtime Stories in the 21st Century: What We Know—and Need to Know—About Digital Reading for Young Children

Abstract

Significance
Current methods for detecting the risk of reading difficulty at school entry remain error-prone and reliant on relatively few behavioural assessment datapoints. Today, less than half of future struggling readers are identified at school entry when using traditional assessments and background data. In this study we adopt a new approach, relying on fine-grained dynamic data from a new literacy app called GraphoGame.

Methods
GraphoGame was played in class daily for ten minutes by 1,676 Norwegian students entering their first year of school. Children played the app over a five-week period during fall of their first year. The mean age during gameplay was 5.7 years.
We used machine learning algorithms to identify aspects of reading that children may need additional support (e.g., phonological awareness). Machine learning algorithms were trained on the process data from the app. To determine the validity of our results, we examined students’ scores on a national screening test conducted two years after gameplay. The screening test was conducted at the start of children’s third year in school and is a psychometrically validated indicator of reading performance.

Results and Implications
In this presentation, we share how gameplay features were extracted from the rich process log data to serve as input to the machine learning algorithms. The features include progression rates related to response times, number of syllables and words mastered during gameplay, distractor stimuli most likely to cause mistakes (from multiple choice games), and frequency/time spent in extracurricular game elements (e.g., avatar play).
Our results improve upon the current methods used at school entry to detect future reading difficulties. Our findings suggest that process data from literacy apps in combination with machine learning may supply teachers with powerful new tools for early identification of struggling readers.

Authors

• Njål Foldnes, University of Stavanger

• Jenny Thomson, University of Sheffield