Differences in Evidence Representations Give Rise to the Illusion of Collapsing Choice Boundaries

Collapsing choice boundaries, where the criterion a person applies to make a decision gets less strict over time, are a key mechanism thought to underpin decisions under time pressure. However, previous work on collapsing boundaries has taken for granted that decisions are driven by a balance of evidence between options, as in diffusion decision models (DDMs). In this paper, we demonstrate that the key empirical pattern used to justify collapsing boundaries -- where slower decisions are made based on a smaller difference in evidence between choice options -- can arise from accumulator models with fixed thresholds. Through mathematical analysis, simulations, and reanalyses of three behavioral datasets, we show that fitting DDMs to data generated by accumulator models systematically favors collapsing boundary models, even when the true boundary is static -— a phenomenon we refer to as the collapsing boundary illusion. We further reveal that model mismatch not only distorts inferences about time-dependent thresholds, but also confounds estimates of between-condition differences of psychologically distinct parameters such as drift rates and decision boundaries. Furthermore, our results suggest that models with and without collapsing boundaries may be difficult to distinguish under relatively common models and experimental designs. Taken together, our results call for caution when interpreting inferences about decision boundaries, and indicate that researchers should consider alternative assumptions about evidence representation rather than relying on a single framework.