Hindsight and Numeric Confidence: Influences on Decision Processes

Abstract

Hindsight bias has long been speculated to increase confidence for subsequent decisions and judgments (e.g., Dawson et al., 1993; Hawkins & Hastie, 1990; Roese & Vohs, 2012); however, to our knowledge, there is no empirical evidence for this consequence of hindsight. The goals of the present dissertation are to: 1) provide evidence of the causal relationship between hindsight bias and increases in both prospective and retrospective confidence, 2) study how the amount of hindsight experience can affect changes in confidence, 3) begin to understand the consequences of elevated confidence due to hindsight bias on decision processes, and 4) investigate possible emotional processes underlying how people learn confidence from experiences. To accomplish these goals, we developed a paradigm that would afford some participants hindsight bias during training for a novel math choice task (Pilot Studies). Then participants completed a set of test trials during which they indicated their confidence in their answers.Using an online sample of participants (N = 842 MTurkers; Study 1), we found that participants who experienced hindsight trials during training reported greater prospective confidence than participants who completed the training in foresight without any feedback (but not more confidence than those with feedback). Additionally, participants who experienced hindsight trials during training reported greater retrospective confidence than participants who completed the training in foresight without any feedback. However, foresight participants appear to have decreased in prospective confidence, rather than hindsight participants increasing in prospective confidence, challenging the speculation that hindsight bias increases confidence. Moreover, the amount of training did not appear to play a significant role in participants’ confidence judgments. Prospective confidence had a parabolic relation with response times (such that people with moderate levels of prospective confidence answered more slowly), suggesting that prospective confidence modulates decision processes. Using an in-person sample of undergraduates (N = 199; Study 2), we replicated key findings from Study 1 and investigated the role of affective responses to feedback (i.e., participants learning that they were correct/incorrect or judging in hindsight that they would have been correct/incorrect) in changes to prospective confidence. Affective responses appear to be important for people to learn prospective confidence and to learn to recognize errors in decision making. Greater positive affect to positive feedback was associated with increases in prospective confidence. On the other hand, greater negative affect to negative feedback was associated with more sensitive retrospective confidence judgments (i.e., participants’ retrospective confidence judgments differed more for correct and incorrect responses). Drift diffusion modeling revealed that hindsight training (versus foresight-with-feedback) reduced participants’ rate of evidence accumulation and their cautiousness in subsequent decisions. Moreover, greater prospective confidence was associated with faster evidence accumulation and more cautious decision making.