The Impact of Different Forms of Decision-Aids on User Best Assessments

In a world where information can be gathered, analyzed, interpreted and diffused much faster than for prior generations, we inquire about optimal schemes for the presentation of predictive models. We asked subjects to make predictions of quarterback ratings by presenting them with different information sets, some information sets including a predictive model prediction (with some different ways, some direct, some indirect, of presenting this information). We investigate (1) actual and optimal model usage by model users, (2) preferences of model users over ways to present information to them, and (3) inter-personal consistency of predictions. We find that (i) subjects are over-confident in their predictions, (ii) that this over-confidence is reduced when the subjects are presented with the predictive model, (iii) that subject s prefer an indirect presentation of the predictive model, where the model is presented as a deviation from a base statistic that is perceived to be relevant and credible, (iv) that their preferences are aligned with their own informal predictive model, and (v) that inter-personal consistency of predictions is fostered by the indirect presentation of the model (as a proposed deviation from a base statistic that is perceived to be credible and relevant).

Keywords: Clinical versus actuarial controversy, clinical synthesis, behavioral economics, information processing, business engineering, over-confidence, anchoring effects.