Imprecise Inference from Sequentially Presented Evidence award

Understanding how people make decisions is crucial to advancing our understanding of economic mechanisms. Rational-choice theory, despite successes in accounting for some aspects of human decision making under uncertainty, fails to capture certain recurrent patterns observed in behavior, such as biases and apparent randomness in choices. Some of these deviations from optimal behavior suggest that information is processed in the brain in a way that introduces imprecision, similar to the imprecision in sensory perception. A growing literature on perceptual judgments argues that in sensory contexts, these imprecisions often actually represent efficient adaptations, given constraints on the available cognitive resources. This raises a question as to whether patterns of imprecision in the case of higher-level cognitive processing (comparisons of numerical magnitudes, estimates of the average values of fluctuating series, and inference about unknown variables) cannot be understood in a similar way. We investigate this hypothesis through an array of experiments in which human subjects are asked to make decisions on the basis of multiple pieces of information, presented sequentially. Decisions made on the basis of sequential evidence are not only important in actual economic life, but also have the advantage as an object of study that they allow us to investigate how each piece of information is separately taken into account in the decision. This reveals the relative allocation of cognitive resources to the processing of each piece of evidence, which in turn sheds light on the underlying constraints faced by the brain. Overall, the main goal of our research is to provide an account of human judgments and economic decisions that is founded on normative principles, and that can be generalized to many problems of human decision-making.

In more technical terms, we investigate, in the context of economic decisions, the neurocognitive hypothesis that the brain faces a problem of constrained optimization: that of allocating its limited information-processing resources to achieve the best possible decisions. From this general principle, we derive testable quantitative models of cognitive processing and decision making that make specific predictions regarding the bias and randomness in human choices. In particular, these suboptimal patterns are predicted to depend on prior beliefs about the probability distribution from which the presented evidence is drawn, and on the rewards associated with different responses in a given context. To test this hypothesis, we design six experiments, in which we manipulate both the prior and the reward structure, and we examine the degree to which our proposed models of constrained information processing capture the behavioral patterns found in experimental data.

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.