Sequential Effects & Dynamics

From typing to cooking to playing an instrument, many tasks involve making a series of decisions about which actions to perform and in what order. Cognitive theory has focused on the representations of sequential context that guide the selection of each action. What is lacking is an account of (a) the speed and accuracy of performance in serial production; (b) how people recover from errors; and (c) the extent to which context representations are position-based or item-based. We present an extension of the Context Retrieval and Updating model (Logan, 2018) to address each of these points. The model treats action selection as the outcome of a race between diffusion processes, with one process for each possible item. The item that is produced is the one associated with the process that first reaches its upper threshold; selection failure occurs if all processes terminate at their lower thresholds. The drift rate for an item depends on the degree to which it is associated with the context of the current position in the sequence, and this context is updated based on the result of each race in the sequence. By applying the model to data from different serial production tasks (previously reported by Logan, 2021), we show that the model (a) accounts for the joint distributions of accuracy and speed in each task by allowing the next race to begin as soon as the previous one has ended, even if the previously selected item has not yet been produced; (b) recovers from errors by, in the event of a selection failure, updating context for the next race with a distributed vector of item activations rather than a single selected item (Caplan, 2022); and (c) accounts for differences in tasks by allowing context to depend on different mixtures of item- and position-based representations. By coupling theories of context representations to immediate performance, this model enables new insights to be gained from the rich set of data available in serial production tasks.

This is an in-person presentation on July 20, 2026 (09:00 ~ 09:20 EDT).

Even simple perceptual choice tasks exhibit rich trial-by-trial dependencies in choices and response times. Previous accounts attribute some of these sequential effects to adaptive learning of base-rate and repetition-rate statistics in the stimulus sequence. However, these accounts largely focus on mean response times, overlooking full response time distributions, overt choice patterns, and the specific cognitive processes shaped by learning. Here, I propose a dynamic evidence-accumulation framework that jointly models error-driven learning and its trial-wise effects on decision parameters. Across multiple open-access datasets, model comparisons reveal that base-rate and repetition-rate statistics both induce response biases, but rely on distinct learning signals: base-rate learning depends on one’s own responses, whereas repetition-rate learning depends on presented stimuli. These results demonstrate how dynamic evidence-accumulation models can uncover latent cognitive processes underlying sequential effects and provide a unified account of trial-by-trial variability in decision-making.

This is an in-person presentation on July 20, 2026 (09:20 ~ 09:40 EDT).

Human decision-making in uncertain environments depends not only on how individuals learn from experience but also on how contextual factors modulate this learning. The present studyvinvestigates how environmental valence (win vs. loss domain) and feedback availability (full vs. partial) jointly shape value learning and evidence accumulation. N = 180 participants completed an adapted two-stage decision-making task in which transitions between states follow a Markovian structure, while the above manipulations induce systematic shifts in motivational and informational context. We compared model-free and model-based learning algorithms embedded within a reinforcement learning-diffusion decision model framework that maps subjective value differences onto drift rate and overall value context onto decision thresholds. This approach allowed us to quantify how learning strategies translate into temporal decision dynamics. We further tested the condition effects of environmental valence and feedback availability on cognitive parameters. Hierarchical Bayesian model comparison results reveal the best fit for model-free learning algorithms. Both environmental valence and feedback availability interact with learning rates, with higher learning rates in the win domain and the partial feedback condition. Feedback availability influences non-decision time and accuracy. The study offers a way to characterize how motivational and informational factors modulate the interplay between model-free and model-based learning and the dynamics of evidence accumulation. We discuss how the findings advance a mechanistic understanding of adaptive decision-making.

This is an in-person presentation on July 20, 2026 (09:40 ~ 10:00 EDT).

Recent decision making models have explained behaviour using mental sampling mechanisms, but there is still little agreement on the specific sampling process, such as whether sampling rates match true probabilities. Here, we trace the sampling process using generation tasks: in two experiments using general online samples (Ns = 52, 51), participants repeatedly produced potential outcomes from pairs of monetary gambles before choosing between them. Results found over-generation of rarer outcomes and under-generation of common outcomes overall but not in initial responses, as well as avoidance of direct repetitions. Participants also tended to select options with higher average utility across their responses, implying generations guided choice. These findings suggest systematic biases in mental sampling that may filter through to choices, constraining models of this behaviour. We thus suggest explicit generation is a valuable method to access underlying choice processes, offering new assessments of existing theories of decision making.

This is an in-person presentation on July 20, 2026 (10:00 ~ 10:20 EDT).