Subjective Probability Interval Estimates: A Simple and Effective Way to Reduce Overprecision in Judgment

摘要: Overprecision in judgment is the most robust type of overconfidence, and one least susceptible to debiasing. It refers people’s excessive certainty accuracy their estimates, predictions or beliefs. Research on overprecision finds that confidence intervals, estimated ranges judges are confident will include correct answer, tend answer significantly less often than what assigned level would suggest. For example, 90% intervals typically about 50% time (Klayman, Soll, Gonzalez-Vallejo, & Barlas, 1999). By this standard, appear too narrow, overprecise. This dissertation focuses effectively reducing bias. In dissertation, I present a novel elicitation method which can reduce overprecision, sometimes eliminating This method, called Subjective Probability Interval Estimates, or, short, SPIES, presents judge with entire range possible values, divided into intervals. The for each interval, probability it includes answer. Since these sum subjective probabilities constrained equal exactly 100%. This work six experiments, organized two parts. Part use SPIES eliciting quantitative tests against other methods three experiments. Experiment 1 included within-subject comparison methods, namely 5th 95th fractile found produce interval estimates higher hit-rates methods. 2 varied task spanned number it, outperformed across all configurations. 3 tested robustness effect different value 5 scales, variations extremity true values range. again produced consistently more inclusive better calibrated intervals. In II, whether improve formats. Participants made multiple using some others. 4 calibration after having used prior estimate before practiced SPIES. held even when had no shared content, suggesting influence estimation process, rather merely increase amount relevant information already memory making second estimate. Experiment subsequent types estimates. When participants could retrieve relatively homogeneous set but were asked likelihoods wide outcomes, they responded by improving inclusiveness However, first was diverse, such retrieving evidence easy, observed suggests do not simply generalize process Rather, might react conflict between knowledge make. may doubt, leading an adjustment account uncertainty. Experiment 6 manipulated existence participants’ distribution structure task, varying set’s exposure time. very long, assign its according likelihood without conflict, came mainly from itself. did either conditions. only moderate, as 5, result improved intervals. Together, results experiments show effective judgment. allows those variety types. addition, enact changes judges’ elicited traditional display accuracy. These features make tool Judgment Decision Making’s biases.