Cognitive processes that work in hindsight: Meta-cognitions or probability-matching?

The term 'hindsight bias' describes the tendency of peoples´ recalled predictions of events to shift towards the real or alleged outcome, once the outcome of the event is known. Although this phenomenon has often been replicated since it was first investigated in the mid-seventies, the underlying processes are not yet fully understood. Within the last decade, numerous studies led to the conclusion that the hindsight bias is (at least to a large extent) a biased judgment rather than a memory, or a (pure) motivation phenomenon. Some theoretical models were formulated during the last few years, which concentrate on biased judgement processes. The Active Reconstruction Model (STAHLBERG et al. 1990) in particular, was proved to predict hindsight effects very well. The main emphasis of this model deals with the role of meta-cognitions. Another model (Accuracy Assessment Model, WINMAN et al. 1998) was presented recently, which stresses that a probability matching strategy is responsible for hindsight distortions. The uniqueness of this model, which is based on assumptions of 'ecological rationality' of the ABC-research group (see GIGERENZER, TODD & THE ABC RESEARCH GROUP (eds.) 1999), is its comprehension of the theoretical and practical relevance of the hindsight bias. The hindsight bias is said to be no more than a pseudo-phenomenon created in psychological laboratories. The aim of this study is to compare different predictions of the strength and/or direction of hindsight effects when subjects have to work with material that provokes complex knowledge-based decisions. The Accuracy Assessment Model has not yet been tested with this kind of material. In an experiment, 62 advanced psychology students worked on clinical psychology records (diagnostic decisions had to be made). The results (pattern of hindsight effects) revealed more evidence in favour of the Active Reconstruction Model. Specific hypotheses derived from the Accuracy Assessment Model gained no confirmation. The observed data could not be explained by this model. On the other hand, they provided evidence in favour of the relevance of meta-cognitve processes and biased information processing (as mentioned in the Active Reconstruction Model).