Finding useful questions: on Bayesian diagnosticity, probability, impact, andinformation gain.
Journal Title: Psychological Review - Year 2006, Vol 112, Issue 4
Abstract
Several norms for how people should assess a question's usefulness havebeen proposed, notably Bayesian diagnosticity, information gain (mutual information), Kullback-Lieblerdistance, probability gain (error minimization), and impact (absolute change). Several probabilisticmodels of previous experiments on categorization, covariation assessment, medical diagnosis, and theselection task are shown to not discriminate among these norms as descriptive models of human intuitionsand behavior. Computational optimization found situations in which information gain, probability gain,and impact strongly contradict Bayesian diagnosticity. In these situations, diagnosticity's claims arenormatively inferior. Results of a new experiment strongly contradict the predictions of Bayesian diagnosticity.Normative theoretical concerns also argue against use of diagnosticity. It is concluded that Bayesiandiagnosticity is normatively flawed and empirically unjustified.
Authors and Affiliations
Jonathan D Nelson
Keywords
Related Articles
- EP ID EP82481
- DOI -
- Views 16
- Downloads 0
Finding useful questions: on Bayesian diagnosticity, probability, impact, andinformation gain.
Several norms for how people should assess a question's usefulness havebeen proposed, notably Bayesian diagnosticity, information gain (mutual information), Kullback-Lieblerdistance, probability gain (error minimization)...