The effects of routine strength on adaptation and information search in recurrent decision making.

The strength of decision routines was manipulated within a computer-controlled micro world simulation involving recurrent decision making. During a learning phase, participants were led to prefer a certain brand of an industrial good either about 15 times in a weak routine or about 30 times in a strong routine condition. In the test phase of Experiment 1, participants were confronted with changes in the microworld that rendered the routine obsolete. Routine maintenance over a series of repeated acquisition decisions was assessed as the major dependent variable. Although new information clearly suggested that a deviation from the routine would be beneficial, strong routine participants were more likely to maintain the routine compared to weak routine participants and a control group in which a comparable option (same outcome probabilities as the routine) carried an unfamiliar brand label. Experiment 2 investigated the effects of routine strength on information search. After having learned the routine, participants were asked to make one final decision involving the routine. The task was either framed as being similar to the learning task or as being novel. Before making the final decision, participants were asked to consider new information about the alternatives. Strong routine participants in the familiar task condition preferred information that favored the routine and avoided unfavorable information. If the task was framed as being novel, such confirmation biases disappeared completely. In contrast, weak routine participants exhibited a moderate confirmation bias in their searches independent from task framing.

A sampling approach to biases in conditional probability judgments: beyond base rate neglect and statistical format.

Conditional probability judgments of rare events are often inflated. Early accounts assumed a general deficit in using statistical base rates. More recent approaches predict improvement when problems are presented in frequency format or refer to natural categories. The present theory focuses on sampling processes. Experiment 1 showed that a seeming advantage of frequency over probability formats is due to a confounded factor, the need to mentally transform stimulus samples. An information search paradigm was used in Experiment 2. When sampling by the predictor, the probability to be estimated, p(criterion/predictor), was conserved in the samples and judgments were quite accurate. However, when sampling by the criterion, the low base-rate event was strongly overrepresented, accounting for the entire bias. Judgments were quite sensitive to the sampled data, but failed to take sampling constraints into account, as shown in Experiments 3 and 4.