Modeling the dynamics of belief formation: Towards a computational understanding of the timing and accuracy of probability judgments

下次你需要预报的时候，停下来问问自己，你是否可以等一等。很有可能，特别是在我们生活的这个加速发展的时代，你现在就想要这个预测。再过几分钟也不行。一小时内都不行。当然，在预测者能够收集到更多信息之后，就不会了。你现在就想要。你想要根据他们当时掌握的信息做出最好的估计。这种需求意味着，准确和及时的预测是有价值的。时间压力如何影响主观概率判断(SP)，它们如何随时间变化，以及在准确预测和及时预测之间有多大的权衡？这些问题很难回答，因为现有的SPS理论都集中在准确性上。他们中的大多数人对SP是如何演变的，因为它们是用预报员的信息构建的。本项目试图通过计算建模和行为实验来绘制SPS的时间进程，以回答这些问题。首先，该计算模型预测了人们在回答“堪萨斯大学男子篮球队赢得今年锦标赛的概率有多大”等问题时产生的SPS，并预测了人们产生判断所需的时间。其次，一系列实证研究促进了人们对人们如何产生判断以及时间压力如何影响他们的理解。在这个项目中开发的计算框架从问题数量、时间压力和激励结构方面为预测民意调查的设计提供信息。该项目还用于培训学生的计算建模和推进STEM教育，因为PI将数据科学培训纳入文科和理科课程。开发响应时间和判断模型的方法为扩展算法的预测能力创造了机会，并为社会和行为科学家提供了一个渠道，让他们在发展数据科学领域发挥积极作用。该项目旨在提供人们如何生成主观概率(SP)预测的动态描述。这项提议有三个目标。第一个目标是开发一个关于信念形成和由这一过程产生的SPS动态的计算模型(SPS目标的动态建模)。这样的模型能够预测SP，当人们构建信念时，它们在最短的时间间隔内如何变化，并预测时间压力如何影响SP。但是，对SPS的动态进行建模需要对信念如何演变的机械性理解。为此，第二个目标对当代SPS理论的一套一致性原则(一致性原则测试目标)进行了实证检验。这些原则规定，人们获得的关于一个假说的证据或支持是独立于替代假说的。在偏好领域已经做出了类似的假设，并且通过所谓的语境效应很好地确立了违规行为。这些语境效应在识别偏好构建背后的认知结构方面具有诊断性作用。数据表明，信仰的构建可能也是如此。这个项目严格测试了这些效应在不同时间的表现。使用SPS的计算模型，该项目还在预报员报告一系列待预测事件的SPS时，研究了在速度和准确性之间权衡的最佳策略。该项目共同从经验上验证了SP的计算框架，该框架有助于隔离可能阻碍人们提供准确和及时预测的机制；并开发干预措施以提高获得SP的效率。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

The next time you need a forecast, stop and ask yourself if you could wait for it. Chances are, especially in this age of accelerations we live in, you want that forecast now. Not in a few minutes. Not in an hour. Certainly not after the forecaster can collect more information. You want it now. You want the best estimate based on the information they have right then. This demand implies that accurate and timely forecasts are valued. How does time pressure impact subjective probability judgments (SPs), how do they change over time, and how much of a trade-off is there between accurate forecasts and timely ones? It is hard to answer these questions because extant theories of SPs have focused on accuracy. Most of them are silent about how SPs evolve as they are constructed with the forecaster's information. This project seeks to answer these questions using computational modeling and behavioral experiments to map the time course of SPs. First, the computational model predicts the SPs people generate in response to questions such as "What is the probability that the University of Kansas men's basketball team will win this year's tournament” and predicts the time it takes people to generate the judgment. Second, a set of empirical studies advance understanding of how people generate the judgments and how time pressure impacts them. The computational framework developed in this project informs the design of prediction polls in terms of number of questions, time pressure, and incentive structures. The project also serves to train students in computational modeling and advance STEM education as the PI integrates data science training within the liberal arts and science curriculum. Developing methods to model response times and judgments create opportunities to expand algorithms' predictive power and provide a channel for social and behavioral scientists to play an active role in developing the field of data science.This project seeks to provide a dynamic account of how people generate subjective probability (SP) forecasts. This proposal has three aims. The first aim is to develop a computational model of belief formation and the dynamics of SPs that result from this process (Modeling the Dynamics of SPs Aim). Such a model enables to predict SPs, how they change over the briefest of time intervals as people construct a belief and predict how time pressure impacts SPs. But, modeling the dynamics of SPs requires a mechanistic understanding of how belief evolves. To this end, the second aim empirically tests a set of consistency principles of contemporary theories of SPs (Tests of Consistency Principles Aim). These principles state that the evidence or support people recruit about a hypothesis is independent of the alternative hypotheses. Analogous assumptions have been made in the domain of preference, and violations are well established via so-called context effects. These context effects have been diagnostic in identifying the cognitive architecture underlying the construction of preference. Data suggest this may also be the case for the construction of belief. This project rigorously tests these effects across time. Using the computational model of SPs the project also examines the optimal policy for trading off speed and accuracy as forecasters report their SPs over a series of to-be-predicted events. Together, the project empirically-validates a computational framework of SPs that help isolate mechanisms that may inhibit people from giving accurate and timely forecasts; and develop interventions to improve the efficiency of obtaining SPs.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.