Speed-Accuracy Tradeoffs in Sample-Based Judgment and Choice

目前的研究项目集中在基于样本的决策的速度和精度的权衡。随着对两个选项的信息采样量的增加，期望精度增加，但决策速度将不可避免地下降。当用于提高准确性的时间可以用来增加在可用时间内完成的正确决策的数量时，就会出现权衡。以前的研究在很大程度上忽略了这些权衡，然而，这些权衡在日常生活的许多领域都很常见。在我们的初步工作中，开发了一个实验任务，允许参与者在做出选择之前确定采样的信息量。在每一次连续任务的试验中，参与者根据两个投资基金在随机选择的n天的成功率的二项式随机样本在两个投资基金之间做出选择。参与者每做出一个正确的选择（即样本中成功率较高的选项），就能赢得1个回报单位，每做出一个错误的选择，就损失1个回报单位。因为总收益是平均收益（准确性）和完成的选择数量（速度）的乘积，所以参与者必须既准确又快速。然而，至关重要的是，因为随着n的增加，完成选择的数量减少的速度比准确性增加的速度更快，理论上可以预测的结果是一个巨大的准确性偏差。绝大多数参与者收集了太大的样本，从而错过了完成选择的最佳数量。心理学和经济学学生以及抑郁症患者作为参与者的三个平行实验证实了这一基本结果。尽管设计了几种操作（反馈;样本限制;选择难度;回报计划）让参与者体验快速策略的优越性，但过采样仍然存在，反映了明显的元认知缺陷。即使当参与者只能赢，永远不会输的时候，他们也不愿意利用快速策略。所描述的项目的目标是理解和阐述这一具有挑战性的现象。首先，我们将在一个新的领域复制所有的初步发现，这个领域独立于货币投资的剩余意义。其次，计算机模拟将确定不同的停止规则和采样模式的最佳策略。第三，在功能层面的方法来界定未能解决的权衡，我们研究的影响，迫使参与者体验快速战略的优势。本计画的最后一部分，是以谢国忠所提倡的可评估性概念，对机械论的解释进行实证检验。因此，准确性偏差可以被消除，并且当呈现格式使得增加样本的信息成本相对于针对准确选择获得的回报的可评估性更可评估时，速度的主导影响将被理解。

The present research project focuses on speed-accuracy tradeoffs in sample-based decisions. With increasing information sampled about two choice options, the expected accuracy increases while decision speed will decrease inevitably. A tradeoff arises as the time expended to increase accuracy can be used to increase the number of correct decisions completed within the available time. Previous research has largely neglected these tradeoffs, which are however quite common in many areas of everyday life. In our preliminary work, an experimental task was developed that allows participants to determine the amount of information sampled before they make a choice. On every trial of a sequential task, participants make a choice between two investment funds, based on binomial random samples of the two investment funds’ success on n randomly selected days. Participants win 1 payoff unit for each correct choice (of the option with the higher success rate in the population from which the sample is drawn) and they lose 1 unit for each incorrect choice. Because the total payoff is the product of the average payoff (accuracy) and the number of choices completed (speed), participants have to be both accurate and fast. However, crucially, because the number of completed choices decreases faster with increasing n than accuracy increases, a theoretically predictable finding is a dramatic accuracy bias. The vast majority of participants gather far too large samples, thereby missing the optimal number of completed choices by magnitudes. Three parallel experiments with students of psychology and economics and with depressed patients as participants corroborate this basic result. Oversampling persisted in spite of several manipulations (feedback; sample limit; choice difficulty; payoff scheme) devised to let participants experience the superiority of speedy strategies, reflecting a conspicuous metacognitive deficit. Even when participants could only win and never loose, they were reluctant to exploit speedy strategies. The goal of the depicted project is to understand and elaborate on this challenging phenomenon. First, we will replicate all preliminary findings in a new domain, independent of the surplus meaning of money investment. Second, computer simulations will determine optimal strategies for different stopping rules and sampling modes. Third, in a functional-level approach to delimit the failure to solve the tradeoff, we examine the impact of various interventions that force participants to experience the advantage of speedy strategies. The final part of the project is devoted to empirical tests of mechanistic account in terms of the evaluability notion propagated by Christopher Hsee. Accordingly, the accuracy bias can be eliminated and the dominant impact of speed will be appreciated when the presentation format renders the information cost of increasing samples more evaluable, relative to the evaluability of the payoff gained for accurate choices.