Information amount is a crucial determinant of decision outcomes. But how much information one should collect before arriving at a decision depends on a cost–benefit trade-off: Is the expected benefit of increased decision accuracy that can be gained from additional information higher than the additional information costs? To investigate this trade-off with temporal costs for information, we developed a speed–accuracy trade-off paradigm with sample-based decisions, in which the total payoff was the product of the average payoff per decision and the number of decisions completed in a restricted period. Increasing n served to increase the accuracy of choices, but also to decrease the number of completed choices. Yet, whereas the number of completed choices decreases linearly with increasing n, accuracy increases in a clearly sublinear fashion. As a consequence, the sample-based choice task calls for more weight given to speed than to accuracy. However, overly conservative sampling strategies prevented almost all participants from exploiting the speed advantage despite various guiding interventions. Even when the task was enriched by the social aspect of a teammate or rival, who demonstrated the optimal trade-off, participants remained too focussed on accuracy. We also investigate the cost–benefit trade-off with financial information costs, for which participants’ performance was less biased. We propose this to be related to how evaluable the information’s costs were relative to its benefits. Issues of adaptivity in contrast with optimality are addressed in a final discussion.

To understand and explain sample-based impression formation, it is necessary to consider both the Brunswikian uncertainty caused by sampling from the stimulus environment and the Thurstonian uncertainty arising from the cognitive processing thereof. Impression judgments must be formed in view of both sources of uncertainty. Even when an ecological sample of a target’s traits is held constant, the resulting distribution of target information in the judge’s mind can vary substantively as a function of semantic and affective responses, top-down inferences, and contextual influences. In the research reviewed in the present chapter, we investigate the interplay of Brunswikian and Thurstonian sampling in a person-impression task based on self-truncated trait samples, in which judges can stop sampling at the very moment when their internal mindset optimally prepares them to form a distinct impression. This task setting produces distinct self-truncation effects; the resulting impression judgments are polarized, conflict-free, and typically driven by small samples (after early truncation). When exactly the same traits are presented in a yoked-control design to other judges, who cannot exploit self-truncation effects, their judgment patterns are similar but clearly less pronounced, reflecting the same Brunswikian trait samples detached from the Thurstonian mindset.

Bayes factors – evidence for one model versus another – are a useful tool in the social and behavioral sciences, partly because they can provide evidence for no effect relative to the sort of effect expected. By contrast, a non-significant result does not provide evidence for the null hypothesis tested. If non-significance does not in itself count against any theory predicting an effect, how could a theory fail a test? Bayes factors provide a measure of evidence from first principles. A severe test is one that is likely to obtain evidence against a theory if it were false – to obtain an extreme Bayes factor against the theory. Bayes factors show why cherry picking degrades evidence, how to deal with multiple testing, and how optional stopping is consistent with severe testing. Further, informed Bayes factors can be used to link theory tightly to how that theory is tested, so that the measured evidence does relate to the theory.