Predicting Reading Comprehension from Constructed Responses: Explanatory Retrievals as Stealth Assessment

Open-ended constructed responses promote deeper processing of course materials. Further, evaluation of these explanations can yield important information about students’ cognition. This study examined how students’ constructed responses, generated at different points during learning, relate to their later comprehension outcomes. College students (N = 75) produced self-explanations during reading and explanatory retrievals after reading. The Constructed Response Assessment Tool (CRAT) was used to analyze these responses across multiple dimensions of language and relate these textual features to comprehension performance. Results indicate that the linguistic features of post-reading explanatory retrievals were more predictive of comprehension outcomes than self-explanations. Further, these models relied on different indices to predict performance.