Serial dependence carries the action that produced the previous report

Technical Summary: Response Production and Serial Dependence

Problem and Motivation
Serial dependence is the phenomenon where current perceptual judgments are biased toward recent past experiences. While traditionally attributed to the persistence of the previous stimulus or the previous decision, the specific source of this carryover remains debated. A critical gap in the literature is the distinction between the content of a report (the represented magnitude) and the action used to produce that report. Previous studies often confounded these factors or utilized different response modalities for different dimensions (e.g., time vs. number), making it difficult to isolate whether cross-dimensional attraction arises from shared magnitude codes or from the retrieval of a specific response episode.

This paper investigates whether serial dependence carries the trace of the action through which a previous judgment was expressed. The authors propose that a "binding-and-retrieval" mechanism (based on the Binding and Retrieval in Action Control, or BRAC, framework) may bind the stimulus, task, and response features into an event file. Retrieval of this file on subsequent trials could bias judgments, even across different magnitude dimensions, provided the response production features recur.

Methodology
The authors conducted three magnitude-reproduction experiments using a novel "accumulate-and-stop" paradigm.

• Stimuli: Participants viewed static dot arrays where duration and numerosity varied independently.
• Task: Participants were cued to reproduce either the duration (Time) or the numerosity (Dots) of the array.
• Response Mechanism: Both tasks utilized the same physical action: holding a key while a dot display accumulated (or changed) and releasing it when the elapsed time or visible count matched the memory. This ensured that both magnitudes were produced through the same time-extended motor action.
• Experimental Manipulations:
  • Experiment 1: Used a single response key for both tasks to test for cross-dimensional carryover and the effect of task switching.
  • Experiment 2: Assigned separate response keys to each task to determine if carryover required the repetition of the specific physical key or if it generalized to the task context.
  • Experiment 3: Randomized the mapping between the reported number and the hold duration on a trial-by-trial basis (ascending vs. descending accumulation). This dissociated the reported numerical value from the motor duration required to produce it.
• Analysis: The authors employed Structural Equation Modeling (SEM) to partition serial dependence into two paths: a "stimulus-history" path (attraction to the previous stimulus) and a "response-history" path (attraction to the previous response feature, controlling for stimuli). They analyzed data across three transition contexts: pure repetitions, mixed repetitions, and mixed task switches.

Key Results

1. Cross-Dimensional Response Carryover: In Experiments 1 and 2, the previous response significantly predicted current bias even when the task dimension changed (e.g., a previous numerosity report biased the next duration judgment, and vice versa). This occurred even when the tasks used different physical keys (Experiment 2).
2. Task Switch Attenuation: While cross-dimensional carryover persisted after task switches, it was significantly weaker than after task repetitions. This suggests that task context modulates the strength of retrieval but does not act as an absolute gate.
3. Dissociation of Stimulus and Response History: When controlling for the previous response, the conventional stimulus-history path (attraction to the previous stimulus) was often reduced or became negative (repulsion), particularly for duration judgments. This supports a "push-pull" model where judgments are repelled from the previous stimulus but attracted to the previous response.
4. Dependence on Response Production (Experiment 3): The critical finding emerged in Experiment 3. When the mapping between the reported number and the hold duration was reversed (ascending vs. descending), the direction of the carryover effect reversed. A larger reported number biased the next duration judgment toward longer durations only when that number was produced by a long hold (ascending mapping). When the same number was produced by a short hold (descending mapping), this bias disappeared or reversed. This demonstrates that the carryover is not driven solely by the abstract magnitude value but by the specific production features (e.g., hold duration) of the previous report.
5. No Cross-Task Stimulus History: Conventional stimulus-based measures showed no reliable cross-dimensional serial dependence, indicating that the stimulus history alone does not explain the observed cross-task biases.

Significance and Claims
The paper claims that serial dependence is not merely the persistence of a stimulus representation or a decision variable but can carry the "trace of the action" through which a judgment was expressed.

• Beyond Dimension-Specific Codes: The results challenge dimension-specific accounts, showing that response history transfers between duration and numerosity when the response structure is shared.
• Beyond Shared Magnitude Codes: The results challenge pure shared-magnitude accounts (which predict carryover based on abstract "more/less" values). Experiment 3 proves that the manner of production (the motor trace) is part of the retrieved episode.
• Binding-and-Retrieval Mechanism: The findings support an extension of the BRAC framework to continuous magnitude reports. The "event file" formed during a trial binds the reported value and the production action. Subsequent trials retrieve this file based on feature overlap (e.g., the response mode), biasing the current judgment.
• Methodological Implication: The authors conclude that studies measuring serial dependence solely via the previous stimulus may miss a significant carrier of history. Identifying the source of serial dependence requires analyzing the response production history, not just the stimulus history.

The paper maintains a modest tone regarding generalization, noting that the findings are specific to the accumulate-and-stop response format and static encoding arrays, and that the exact nature of the "production feature" (motor vs. temporal vs. visual) requires further dissection. However, the core claim remains robust: the history of a previous report episode, including its production action, is a distinct and potent source of serial dependence.