Judgements of learning (JOL) of paired associates can be made immediately after learning or after a delay, while viewing the first word (cue) only or both words (cue–target) in a pair. Delayed cue-only judgements are more related to subsequent memory performance than delayed cue–target, immediate cue-only, or immediate cue–target judgements. In two experiments we tested students' knowledge of this delayed JOL effect and whether their knowledge increases as a function of task experience (Experiment 2). The majority of the participants did not choose the more effective judgement strategy and they did not systematically alter their behaviour as a function of task experience. Instead, a subset of the participants selected judgement strategies on the basis of a learning goal, that is, a strategy that let them restudy both words in a pair. In sum, most students appear to be unaware of the powerful influence of delayed cue-only JOLs on monitoring accuracy.

We present the Concordant-Ranks (CR) strategy that decision makers use to quickly find an alternative that is proximate to an ideal alternative in a multi-attribute decision space. CR implies that decision makers prefer alternatives that exhibit concordant ranks between attribute values and attribute weights. We show that, in situations where the alternatives are equal in multi-attribute utility (MAU), minimization of the weighted Euclidean distance (WED) to an ideal alternative implies the choice of a CR alternative. In two experiments, participants chose among, as well as evaluated, alternatives that were constructed to be equal in MAU. In Experiment 1, four alternatives were designed in such a way that the choice of each alternative would be consistent with one particular choice strategy, one of which was the CR strategy. In Experiment 2, participants were presented with a CR alternative and a number of arbitrary alternatives. In both experiments, participants tended to choose the CR alternative. The CR alternative was on average evaluated as more attractive than other alternatives. In addition, measures of WED, between given alternatives and the ideal alternative, by and large agreed with the preference order for choices and attractiveness evaluations of the different types of alternatives. These findings indicate that both choices and attractiveness evaluations are guided by proximity of alternatives to an ideal alternative.

In this thesis, some old decision strategies are investigated and a new one that furthers our understanding of how decisions are made is introduced. Three studies are presented. In Study I and II, strategies are investigated in terms of inferences and in Study III, strategies are investigated in terms of preferences. Inferences refer to decisions regarding facts, e.g., whether a patient has a heart disease or not. Preferences refer to decision makers’ personal preferences between different choice alternatives, e.g., which flat out of many to choose. In all three studies, both non-compensatory strategies and compensatory strategies were investigated. In compensatory strategies, a high value in one attribute cannot compensate for a low value in another, while in non-compensatory strategies such compensation is possible. Results from Study I showed that both compensatory (logistic regression) and non-compensatory (fast and frugal) strategies make inferences equally well, but logistic regression strategies are more frugal (i.e., use fewer cues) than the fast and frugal strategies. Study II showed that the results were independent of the degree of expertise. The good inferential ability of both non-compensatory and compensatory strategies suggests there might be room for a strategy that can combine the strengths of the two. Study III introduces such a strategy, the Concordant-ranks (CR) strategy. Results from Study III showed that choices and attractiveness evaluations followed this new strategy. This strategy dictates a choice of an alternative with concordant ranks between attribute values and attribute weights when alternatives are about equally attractive. CR also serves as a proxy for finding the alternative with the shortest distance to an ideal. The CR strategy combines the computational simplicity of non-compensatory strategies with the superior information integration ability of compensatory strategies.

An experiment involving 45 psychology students as participants made it possible to differentiate between strategies used for quickly finding a promising alternative. It was hypothesised (a) that decision makers have an ideal alternative in mind when entering the decision process and (b) that the promising alternative is the one that is most similar to this ideal alternative. For this purpose, the participant’s ideal alternatives were first identified with the think-aloud method. Second, based on this ideal alternative, a computer application generated and presented five alternatives to the participants. Each alternative represented a strategy that individuals can use in order to make decisions. In order to make sure that the alternatives only differed in the applied strategy and not in attribute values, the five alternatives were equal in terms of multi-attribute utility. One alternative was most similar to the participant’s ideal alternative in terms of having shortest Euclidean distance. This alternative represented the strategy of similarity or distance judging. The second alternative had an agreement between the rank-order of attribute values and the ranks of the importance weights of the attributes in the presented alternative as in the ideal alternative. We showed theoretically that if an alternative has the same multi-attribute utility, a choice in line with this strategy is also equivalent to minimizing the weighted Euclidean distance to the ideal alternative. Two alternatives each applied a non-compensatory rule, focussing on the most positive aspects of the alternatives (maximin rule and the lexicographic rule). The last alternative did not apply any decision making rule. It was found that choices as well as preference ratings most often could be predicted from an alternative having an agreement between the rank-order of attribute values and the ranks of the importance weights of the attributes. Think aloud data gave additional support for this conclusion. These findings not only suggest the existence of an ideal alternative in a decision making situation. They also suggest the existence of a newly discovered decision making strategy based on a pattern matching heuristic. In this heuristic, the pattern of attribute weights and attribute values in a given alternative is matched to the corresponding pattern in the ideal alternative. Our data are compatible with the notion that the finding of a promising alternative is guided by this pattern matching heuristic.

It is assumed that in judgment analysis, experts provide better models than non-experts. In this study we challenge this view by showing that data from non-experts might be equally suitable for building models. We show this by modeling the decisions of 21 medical students, 27 general practitioners, and 22 cardiologists on real patient vignettes regarding diagnosing heart failure. The models used were logistic regression and fast and frugal models. Results showed that there were no differences between any of the expertise groups in terms of fit, prediction, information searched, or percent of actual diagnosis in any of the models. Therefore, it seems, at least for the studied conditions, using models built on decision data from non-experts versus experts might be equally valid in judgment analysis.

Using participant data from the medical domain, the robustness of logistic regression (LR) with different cue inclusion levels and two fast and frugal (F&F) models in terms of predictive accuracy and frugality were tested. Two data sets based on judgments of verbally described patients were used: Heart failure (66 analysts), and Hyperlipidemia (38 analysts). In both data sets, when the models were cross-validated, there was a significant decrease in predictive accuracy for all models, especially when all cues were used in LR. The other models had about equal predictive accuracy, also when comparisons were made with actual diagnoses, with a slight advantage for LR in the Heart failure study. LR using the 5% inclusion level was more frugal than F&F. These results emphasize the importance of using cross-validation and of choosing the proper significance levels for cue inclusion and when comparing different judgment models.