Recommendations for Orchestration of Formative Assessment Sequences: a Data-driven Approach

International audienceFormative assessment aims to improve teaching and learning by providing teachers and students with feedback designed to help them to adapt their behavior. To face the increasing number of students in higher education and support this kind of activity, technology-enhanced formative assessment tools emerged. These tools generate data that can serve as a basis for improving the processes and services they provide. Based on literature and using a dataset gathered from the use of a formative assessment tool in higher education whose process, inspired by Mazur's Peer Instruction, consists in asking learners to answer a question before and after a confrontation with peers, we use learning analytics to provide evidence-based knowledge about formative assessment practices. Our results suggest that: (1) Benefits of formative assessment sequences increase when the proportion of correct answers is close to 50% during the first vote; (2) Benefits of formative assessment sequences increase when correct learners' rationales are better rated than incorrect learners' ones; (3) Peer ratings are consistent when correct learners are more confident than incorrect ones; (4) Self-rating is inconsistent in peer rating context; (5) The amount of peer ratings makes no significant difference in terms of sequences benefits. Based on these results, recommendations in formative assessment are discussed and a data-informed formative assessment process is inferred