Attention Probes May Inflate Real Effects and Create Pseudoeffects: A Rerun and Reassessment of Hemed et al. (2020)

Abstract

This study documents the potential influence of attention probes in experimental paradigms by addressing their unintended effects on response time measurements. Attention probes, which are commonly used to assess participant engagement, may introduce task-switch costs that can confound experimental results. We show how probe-induced biases inflated findings related to reinforcement from sensorimotor predictability and generate a fictitious behavioral response to sensory prediction error. We show that by excluding task trials immediately after attention probes, these biases can be corrected. We validate this approach in two new experiments devoid of probes. The results confirm that response reinforcement from predictable action effects only accumulates as long as predictions hold. These findings challenge traditional (reward-based) reinforcement models by suggesting a distinct mechanism for reinforcement from sensorimotor predictability. The validated corrective method provides a practical tool for mitigating similar confounds in past and future studies.