Hindsight Bias in Gustatory Judgments Rüdiger F. Pohl, 1 Stefan Schwarz, 2 Sabine Sczesny, 2 and Dagmar Stahlberg 2 1 Justus-Liebig University Giessen, Germany and 2 Sonderforschungsbereich 504, University of Mannheim, Germany Abstract. Being in hindsight, people tend to overestimate what they had known in foresight. This phenomenon has been studied for a wide variety of knowledge domains (e.g., episodes with uncertain outcomes, or solutions to almanac questions). As a result of these studies, hindsight bias turned out to be a robust phenomenon. In this paper, we present two experiments that successfully extended the domain of hindsight bias to gustatory judgments. Participants tasted different food items and were asked to estimate the quantity of a certain ingredient, for example, the residual sugar in a white wine. Judgments in both experi- ments were systematically biased towards previously presented low or high values that were labeled as the true quantities. Thus, hindsight bias can be considered a phenomenon that extends well beyond the judgment domains studied so far. Key words: judgment, hindsight bias, anchoring, expertise Hindsight bias refers to the phenomenon that people do not only feel wiser after the event (which would reflect successful learning) but that they also think that they were more sure about the outcome than they actually had been (Fischhoff, 1975). In order to quantify the degree of this illusion, participants in one type of experimental design (called the “memory design”) were asked to give a confidence rating be- fore the event and later remember this value after the outcome of the event was given. Typically the remembered confidence for the true outcome was higher than the original rating had been. In another type of experiment (called the “hypothetical de- sign”), participants received the outcome right away and were then asked to rate their confidence in the given outcome had they not known it. Compared to ratings that were given without outcome knowledge, the posterior confidence of the true outcome was again increased. This illusion has been studied across The reported research was partly supported by the Deutsche Forschungsgemeinschaft within the special re- search domain SFB 504 “Rationality, decision making, and economic modeling” at the University of Mannheim (to the fourth author) and through research grant Po 315/6Ð3 (to the first author). For her help in running Experiment 2 we would like to thank Cornelia Glagla. Helpful comments on an earlier version of this paper were provided by Ulrich Hoffrage and an anonymous reviewer. DOI: 10.1027//1618-3169.50.2.107 ” 2003 Hogrefe & Huber Publishers Experimental Psychology 2003; Vol. 50(2): 107Ð115 a variety of domains (see Christensen-Szalanski & Willham, 1991, and Hawkins & Hastie, 1990, for overviews) leading to the conclusion that hindsight bias appears to be one of the most robust distortions of human memory. Indeed, attempts to reduce or even eliminate the bias almost always failed (see Pohl & Hell, 1996, for examples). Independent of the procedural design, several ex- periments have used numerical estimates instead of confidence ratings (e.g., Erdfelder & Buchner, 1998; Pohl, 1998). For example, a participant (in the hypo- thetical design) who was given the solution “93 me- ters” to the question “How high is the Statue of Lib- erty?” may generate a posterior estimate of “80 me- ters,” while another participant who has not received the solution may give an estimate of “70 meters.” This result would indicate a distortion of ∆ = 43.5% [= (80Ð70)/(93Ð70)] towards the solution (see Ja- cowitz & Kahneman, 1995). Another variant of this experimental procedure is to use two different anchor values for different participants. For example, one participant is told that the height of the Statue of Liberty is “90 meters,” while another participant is told that it is “50 meters.” As a consequence, the first participant may estimate “80 meters,” while the second one may estimate “60 meters.” Such a result would indicate a bias of ∆ = 50% [= (80Ð60)/(90Ð 50)].