DDL in Context: A meta-analysis of regional variations and predictors of effectiveness in data-driven learning

Abstract

Data-driven learning (DDL) provides corpora access to learners as a tool for language learning (Johns, 1991). This meta-analysis attempts to assess the volume of DDL publication and investigates whether regional variations exists. Also, this meta-analysis updates effect size findings, provides a regional analysis, and situates DDL application in the classroom context. This is done through assessing the importance of predictors of effectiveness related to learner-related factors (population variables) and intervention-related factors (treatment variables). The meta-analysis of 98 unique samples shows that DDL is a flourishing field frequently responding to research calls, and that it is an effective intervention with a large effect (g = 1.01). Publication volume is not distributed equally across regions and is mostly centred in Asia and the Middle East, with medium and large effect sizes, respectively. Secondly, population variables and treatment variables within classrooms are evaluated using the multi-model inference approach to determine the importance of each factor in predicting effect size. The results suggest that the most important population variables are region, institution, and sample size, respectively. Experiment duration, corpus type, and DDL interaction type were the most important predictors among the treatment variables. The findings also indicate that the language proficiency level of students and the linguistic target of intervention are not critical predictors of the effectiveness of DDL. These results provide evidence for the versatility of DDL. Despite DDL application being largely exclusive to research and universities (Timmis and Templeton, 2023, p.420), the findings prove that it is also promising in high schools as well as other language teaching institutions and academies. The meta-analysis suggests that medium- and long-term DDL interventions in smaller samples with direct access by learners to public corpora are the most effective.