Testing the interleaving effect without response bias: A forced-choice reevaluation of Kornell and Bjork (2008).

In a highly influential study across a wide breadth of literature, Kornell and Bjork (Psychological Science, 19[6], 585-592, 2008) showed that learning is enhanced by presenting exemplars (paintings) of to-be-learned categories (artists) in an interleaved sequence (e.g., A1, B1, C1 … A2, B2,, C2 …) rather than a blocked sequence (e.g., A1, A2, A3 … B1, B2, B3 …). However, this study, and nearly all direct replications, used an identification procedure that confounds memory abilities with response biases (i.e., one's criteria for using certain response choices); any interleaving effect assessed through an identification task may be an overestimate or, indeed, an underestimate of the effect (Hautus et al., Detection Theory, 2021). To address this, we conducted a direct replication: online, N = 288; t(287) = 8.08, p < 0.001, dz = 0.48, 96% CI [0.36, 0.59]-for the first time to our knowledge-accompanied by measures of differential learning and response biases across learned categories (which were substantial, with participants using some response categories at a rate several times higher than others). We then conducted a critical conceptual replication, changing the task from identification ("Which artist painted this?") to n-alternative forced-choice ("Which of these was painted by [e.g.] Seurat?"), the gold standard of memory tests (Brady et al., Psychonomic Bulletin & Review, 30, 2023). Reassuringly, the 2AFC experiment showed an interleaving effect comparable to our direct replication: online, N = 276; t(275) = 7.19, p < .001, dz = 0.43, 95% CI [0.32, 0.55]. Put together, this study showcases the challenges of interpretation facing any identification paradigm, illustrates a straightforward method to address them, and puts the interleaving effect on firmer footing.