Take-the-best and the influence of decision-inconsistent attributes on decision confidence and choices in memory-based decisions

ABSTRACT

Take-the-best (TTB) is a decision strategy according to which attributes about choice options are sequentially processed in descending order of validity, and attribute processing is stopped once an attribute discriminates between options. Consequently, TTB-decisions rely on only one, the best discriminating, attribute, and lower-valid attributes need not be processed because they are TTB-irrelevant. Recent research suggests, however, that when attribute information is visually present during decision-making, TTB-irrelevant attributes are processed and integrated into decisions nonetheless. To examine whether TTB-irrelevant attributes are retrieved and integrated when decisions are made memory-based, we tested whether the consistency of a TTB-irrelevant attribute affects TTB-users’ decision behaviour in a memory-based decision task. Participants first learned attribute configurations of several options. Afterwards, they made several decisions between two of the options, and we manipulated conflict between the second-best attribute and the TTB-decision. We assessed participants’ decision confidence and the proportion of TTB-inconsistent choices. According to TTB, TTB-irrelevant attributes should not affect confidence and choices, because these attributes should not be retrieved. Results showed, however, that TTB-users were less confident and made more TTB-inconsistent choices when TTB-irrelevant information was in conflict with the TTB-decision than when it was not, suggesting that TTB-users retrieved and integrated TTB-irrelevant information.

KEYWORDS:

• Decision-making
• memory-based decisions
• take-the-best strategy
• evidence accumulation
• parallel constraint satisfaction

Disclosure statement

No potential conflict of interest was reported by the authors.

Notes

1 Specifically, participants were told that 20 eyewitnesses had been interrogated and consensus was highest for Attribute 1 (e.g., shoes), second highest for Attribute 2 (e.g., coat), and so on. We provided participants only with the ranking of validities and not with numerical validities (e.g., 0.8, 0.7, etc.), because the latter may have biased participants towards either a compensatory or a non-compensatory interpretation of the attribute weights. The validity ranking, in contrast, allowed for the possibility that some participants would give most weight to the best discriminating attribute (TTB) whereas others would weigh the attributes more equally (WADD or EQW).

2 Because the classification method is comparative, the classification of a participant as TTB-CC only indicates that the choices of this participant were more likely to be produced by a TTB- rather than a WADD-/EQW-strategy. It does not tell, however, whether the proportion of TTB-inconsistent choices of a participant classified as TTB-CCs significantly deviated from zero. A more conservative approach for classifying a participant as TTB-CC would be to only consider choices to the 12 diagnostic pairs (6 different pairs, each presented twice) and to conduct a binomial test. The binomial test becomes significant when a participant makes TTB-inconsistent choices in three or more cases (p < .05). When applying this more conservative classification criterion, of the 41 participants classified as TTB-CCs with the Bröder and Schiffer method, 27 also had a non-significant binomial test, indicating that they chose the non-TTB option by (non-systematic) mistake. As analysing only these 27 TTB-CCs rather than the 41 TTB-CCs did not change the present results, we decided to report the analysis of TTB-CCs including all 41 participants because this classification method is widely used in this research area. The average proportion of TTB-inconsistent choices was 17% for the 41 TTB-CCs (i.e., 2 out of 12 pairs, which corresponds to a probability of p < .05 using the binomial criterion).

3 Separate analyses were conducted because our primary research goal was to specifically examine the effects of TTB-irrelevant attributes on the decision behaviour of TTB-CCs. Considering strategy group as between-subjects factor would have yielded significant interactions with both within-subject factors. For sake of clarity, therefore, only results from the separate analyses are reported.

4 The correlation between the two confidence ratings was .70, suggesting that participants were quite consistent in their confidence judgements.

5 All single comparisons were significant, ps < .042.

6 Note, however, that neither the interaction between the two predictors nor the contrast between the consistent and the non-discriminating condition was significant. The contrast between inconsistent and consistent pairs approached significance, p = .053.