Knowing if you would have known : A model of the hindsight bias

Sammanfattning: A common finding in previous research on judgment under uncertainty is that when participants are presented with the correct answer to a general knowledge item or feedback about the actual outcome of an event they respond as if they "knew-it-all-along", as if the presented answer or outcome was easy to predict. The present thesis consists of ten experiments and offers a new theoretical model of this phenomenon, commonly referred to as hindsight bias. The findings of Study I indicated that the way items (here general knowledge questions) normally are sampled can contribute to increased bias. When items were sampled randomly from a naturally delineated reference class, and a within-subjects procedure was employed, bias was entirely eliminated. In Study II a new quantitative model for hindsight bias, the accuracy-assessment model was developed. The model, based on the notion that participants use the strategy of assessing whether or not they would have been correct, predicts a consistent relation between overconfidence and hindsight bias, referred to as the confidence-hindsight mirror effect. Additional predictions are improvement in the calibration score in hindsight conditions and circumstances under which the bias will reverse, observations which other models fail to account for. These predictions were confirmed in four experiments, and a high quantitative fit of the model was observed. Study III showed that situations involving higher cognitive processes are not necessarily associated with more bias. The results were compatible with the accuracy-assessment model. An aggregated analysis of the data from the three studies showed a reasonable quantitative fit. In the final section, theoretical implications are discussed. It is concluded that hindsight bias may occur as a result of erroneous judgmental processes rather than by automatic assimilative memory processes.