Modeling Theories and Theorizing Models: an Attempted Replication of Miller-Cotto & Byrnes’ (2019) Comparison of Working Memory Models Using ECLS-K Data

Abstract

Working memory is an essential mechanism in the cognitive learning process. However, its definitions and mechanisms remain a topic of debate. Miller-Cotto and Byrnes (Journal of Educational Psychology, 112(5), 1074–1084, 2020) reported a comparison of three models of working memory to determine which best accounted for data obtained from a national US dataset of young children over time (Early Childhood Longitudinal Study of Kindergarten). In comparing these models, which they derived from competing theories of working memory, they found that (1) one of the tested models best fit the available data and (2) the model parameters were similar across content domains, concluding that one specific theory of working memory was best supported. The study reported here is an attempted replication of this work. We implemented a multi-phase effort at replication, first undertaking a review of the literature surrounding competing theories of working memory to aid interpretation of results, then attempting a direct replication, and lastly applying an alternative modeling technique to appropriately differentiate between between- and within-person variance. Neither effort succeeded in fully replicating the original findings. Instead, we found that parsing between- and within-person variance is an essential strategy for appropriately interpreting the relationships between working memory and domain learning. Doing so led to support for different models across domains, and interpretation of aspects of these models through the lens of relevant theory suggested alternative strategies for examining these issues in future research.