The Automated Analysis of Constructed Responses (AACR, pronounced "acer") research group consists of researchers from several universities with backgrounds in various STEM disciplines (biology, chemistry, engineering, and statistics), assessment and educational research. We hope this research can help us gain greater insight into how students engage in scientific practices and think about “big ideas” in STEM, such as evolution, energy, and genetics.
Constructed response questions – in which students must explain phenomena in their own words – create more meaningful opportunities for instructors to identify their students’ learning obstacles than multiple choice questions. However, the realities of typical large-enrollment undergraduate classes or multiple sections of a course restrict the options faculty have for quickly analyzing student writing.
In the AACR research group, we explore the use of machine learning and computerized text analysis in science education. We use apply these techniques to evaluate students’ short written answers as part of assessments in secondary education and undergraduate STEM courses. We have created computer models that can predict expert ratings of student responses with accuracy similar to inter-rater reliability among expert raters. These techniques have potential for changing assessment practices across STEM disciplines, regardless of class size.
The AACR research group at Michigan State University is part of the CREATE for STEM Institute.