This study describes an exploratory practice of a reflection activity which uses a learning management system. The reflection activity, part of English classes in Hiroshima University that span a period of two academic quarters, aims to observe students’ engagement in learning behaviors in week-long clusters. The university administers a large-scale English education program which blends various learning modes such as offline and online learning. Due to the complexity of the program, we found difficulty in supporting students' engagement in their autonomous and habitual learning. In order to gather time series data, the authors asked students to report their engagement in learning behaviors and week-long lifelogs using a mobile compatible learning management system. The reflection form consists of 11 items using binary rating. To evaluate this exploratory practice, the present study constructed a model which combines a two-factor two-parameter logistic model and a latent curve model to fit the observations. We found that students' engagement levels are sequentially influenced by their health levels as represented by the lifelogs over time, and the health levels vary stochastically. Based on the findings, we also discuss how to further improve the program.
