In the theoretical part of the thesis, several models of probabilistic inference will be described with a particular emphasis on the simple heuristics suggested within the adaptive toolbox approach by Gigerenzer, Todd and the ABC Research Group (1999). The numerous critical responses to this approach are subsumed to primarily - though not exclusively - identify two problems: (1) the problem of descriptive validity and (2) the problem of strategy selection. Subsequently, in Chapter 3 the common methodological approaches to judgment and decision making are outlined. In this context, especially relevant to the aforementioned problems is the methodological problem of strategy identifiability (e.g. Bröder, 2000), which is caused by the substantial overlap in the behavioral predictions made by various decision models on the outcome level. Next, the empirical evidence of the Take The Best heuristic (TTB), one of the two most frequently studied simple heuristics, is reviewed extensively. Based on the empirical evidence, it appears that, albeit not a universal theory, TTB is a descriptively valid model of probabilistic inferences, especially under several, favorable conditions such as high information costs, time constraints, non-compensatory structure of environment, and cues recalled from memory (Chapter 4). The varying proportions of different strategies found in previous studies, have been interpreted to potentially result from individual differences. In the present thesis, the relation between spontaneous strategy selection and cognitive style (maximization tendency and preference for intuitive and deliberate decision making) was examined in four studies.
In the first study (N =235), the factor structure and psychometric properties of the German language version of the maximization and regret scale (Schwartz et al., 2002) were assessed using confirmatory factor analyses (CFA). The inner factor structure of the scale was replicated, with model fits roughly similar to those reported for the English language version. However, researchers are discouraged to use the short versions of the questionnaire because of their low reliability. The results on validity of the scale in the context of an online recommender system for last minute travels show that maximizers need longer to decide, and evaluate more options.
In the second study (N = 24), concurrent verbal protocols were used to assess the amount and variability of information recalled. Participants had to infer which of two cities has more inhabitants while thinking-aloud. Decision trials consisted of pairs of either German or Austrian cities. The less-is-more effect was replicated by observing that participants made more accurate inferences for the less familiar German cities than for the more familiar Austrian cities. The verbalized cues reflected highly heterogeneous knowledge. On average participants verbalized one cue per decision trial. The preference for deliberate decision making was positively correlated with the number of verbalized cues. The tendency to maximize was also positively related to the number of verbalized cues, interestingly however, this correlation was only significant in the less familiar domain (German cities).
In the third study (N = 79), an experimental setup was used in which cue value patterns had to be learned. In a subsequent decision phase, participants made inferences between two options while completing a tone monitoring-task in parallel (cognitive load condition). The options were fictitious candidates for the Academy Award as the best female actor in a leading role and were described by four binary attributes. TTB was found to be the most prevalent strategy, and under cognitive load, the proportion of TTB users did not decrease as much as the proportion of participants classified as using more elaborate strategies. This finding provides further support for the computational simplicity assumption of TTB. The tendency to maximize and the preference to engage in deliberate instead of intuitive decision making were only marginally higher for participants who were classified to use additive strategies.
In the fourth study (N = 20), the setting from the third study was used but with systematically missing information in the phase of cue value learning. It was assumed that missing information and, as a consequence, partially unalignable cues would lead participants to depart from one-reason decision making and to seek more confirming evidence for their inferences. Although, one-reason decision making consistent with TTB was found, the proportion was lower than in the previous study.
Evidently, missing information did induce search for more confirming evidence, at least in some participants. In this study, relations between decision style and inference strategies were not found.