Emerging patterns of understanding: tracing patterns in students’ conceptions of evolution using machine learning

Background Evolution is at the core of modern biology education, yet students continue to struggle to form explanations of evolutionary phenomena. Formative assessment has the potential to support students in their learning, but comprehensive assessment of the concepts and preconceptions students hold presents teachers with difficulties. In this study, we analyze patterns of concepts and preconceptions in upper-secondary level students' explanations of evolutionary phenomena. We investigate how the patterns develop, which interdependencies between concepts and preconceptions we can observe and in which way the findings diverge from the intended sequence of taught concepts. Method We analyzed the explanations students gave to the ACORNS instrument when implemented at five time points during a unit on five factors of evolution using the EvoGrader tool. We constructed binary patterns indicating presence and absence of six key concepts and three preconceptions. Using these patterns, we traced the development of students’ explanations over the course of the unit. Results In total, we found 95 of 512 possible different explanatory patterns in 1662 student answers. We analyzed the patterns for frequency and qualitatively analyzed those patterns that were (1) particularly frequent in at least one time point, (2) showed a directed change of frequency over the course of the unit, (3) differed in frequency between animal/gain and plant/loss contexts, or (4) contained a particularly high number of key concepts or preconceptions. We compared the patterns against established learning progressions to determine students’ conceptual progress in relation to the progression underlying the unit design. Conclusions Our findings highlight not only how a nuanced analysis of student responses can provide instructionally actionable information, but also how crucial elements of student ideas are missed when student explanations are graded as sum scores. We discuss the implications of these findings for formative assessment practices in evolution education.