Ritz H, Frömer R, Shenhav A
Cognitive, Linguistic, and Psychological Sciences, Brown University.
Carney Institute for Brain Sciences, Brown University.
bioRxiv. 2025 Apr 14:2023.01.18.524640. doi: 10.1101/2023.01.18.524640.
Decision scientists have grown increasingly interested in how people adaptively control their decision making. Researchers have demonstrated that parameters governing the accumulation of evidence towards a choice, such as the decision threshold, are shaped by information available prior to or in parallel with one's evaluation of an option set (e.g., recent outcomes or choice conflict). A recent account has taken a bold leap forward in this approach, suggesting that adjustments in decision parameters can be motivated by the value of the options under consideration. This motivated control account predicts that when faced with difficult choices (similarly valued options) under time pressure, people will adaptively lower their decision threshold to ensure that they make a choice in time. This account was supported by drift diffusion modeling of a deadlined choice task, demonstrating that decision thresholds decrease for difficult relative to easy choices. Here, we reanalyze the data from this experiment, and show that evidence for this novel account does not hold up to further scrutiny. Using a more systematic and comprehensive modeling approach, we show that this previously observed threshold adjustment disappears (or even reverses) under a more complete model of the data. Importantly, we further show how this and other apparent evidence for motivated control arises as an artifact of model (mis)specification, where one model's putatively controlled decision process (e.g., value-driven threshold adjustments) can mimic another model's stimulus-driven decision processes (e.g., accumulator competition or collapsing bounds). Collectively, this work reveals crucial insights and constraints in the pursuit of understanding how control guides decision-making, and when it doesn't.
决策科学家们对人们如何适应性地控制自己的决策越来越感兴趣。研究人员已经证明，决定向某个选择积累证据的参数，比如决策阈值，会受到在评估选项集之前或与之同时可得的信息（例如近期结果或选择冲突）的影响。最近的一种观点在这种方法上有了大胆的飞跃，认为决策参数的调整可能受到所考虑选项价值的驱动。这种动机控制观点预测，当在时间压力下面对困难选择（价值相近的选项）时，人们会适应性地降低他们的决策阈值，以确保能及时做出选择。这一观点得到了一项有时间限制的选择任务的漂移扩散模型的支持，该模型表明，相对于简单选择，困难选择的决策阈值会降低。在此，我们重新分析了该实验的数据，并表明这一新颖观点的证据经不起进一步推敲。使用更系统、更全面的建模方法，我们发现，在更完整的数据模型下，之前观察到的阈值调整消失了（甚至出现了反转）。重要的是，我们进一步展示了这种以及其他动机控制的明显证据是如何作为模型（错误）设定的产物而出现的，即一个模型假定的受控决策过程（例如价值驱动的阈值调整）可能会模仿另一个模型的刺激驱动决策过程（例如累加器竞争或边界收缩）。总的来说，这项工作揭示了在理解控制如何引导决策以及何时不起作用方面的关键见解和限制。
