Memory‐Based Simple Heuristics as Attribute Substitution: Competitive Tests of Binary Choice Inference Models

Abstract

Some researchers on binary choice inference have argued that people make inferences based on simple heuristics, such as recognition, fluency, or familiarity. Others have argued that people make inferences based on available knowledge. To examine the boundary between heuristic and knowledge usage, we examine binary choice inference processes in terms of attribute substitution in heuristic use (Kahneman & Frederick, 2005). In this framework, it is predicted that people will rely on heuristic or knowledge-based inference depending on the subjective difficulty of the inference task. We conducted competitive tests of binary choice inference models representing simple heuristics (fluency and familiarity heuristics) and knowledge-based inference models. We found that a simple heuristic model (especially a familiarity heuristic model) explained inference patterns for subjectively difficult inference tasks, and that a knowledge-based inference model explained subjectively easy inference tasks. These results were consistent with the predictions of the attribute substitution framework. Issues on usage of simple heuristics and psychological processes are discussed.

Familiarity‐Matching: An Ecologically Rational Heuristic for the Relationships‐Comparison Task

“…As for BIC, p denotes the number of free parameters (only one parameter, ε , for each strategy and, therefore, p = 1), and n denotes the number of choice pairs ( n = 60 for each difficulty level). According to previous studies (e.g., Honda et al, ; Raftery, ), we assumed w M 0.99 as “very strong” evidence, 0.95 w M < 0.99 as “strong” evidence, 0.75 w M < 0.95 as “positive” evidence, and 0.50 w M < 0.75 as “weak” evidence. If w M < 0.50, or the value of w M was equal between two or more models, we assumed that one model could not explain her/his inference patterns that was defined as “Not classified.”…”

“…Herbert Simon proposed the metaphor that effective behaviors are generated when context (environmental structures) and cognition (human computational capabilities) fit together like the blades of a pair of scissors—one blade being environmental structures and the other being human cognitive capacity (i.e., “Simon's scissors” metaphor) (Simon, ; see also Kozyreva & Hertwig, ; Lockton, ; Todd & Brighton, ). In fact, many previous works have demonstrated that, using binary choice tasks, the effectiveness of people's subjective memory experiences as inference cues is clearly explained in terms of the real‐world environmental structures (e.g., Goldstein & Gigerenzer, ; Hertwig, Herzog, Schooler & Reimer, ; Herzog & Hertwig, ; Honda, Abe, Matsuka & Yamagishi, ; Honda, Matsuka & Ueda, ; Schooler & Hertwig, ; Xu, González‐Vallejo, Weinhardt, Chimeli & Karadogan, ; as a review, see Gigerenzer & Goldstein, ) . For example, consider a binary choice question: “Which city has a larger population, Tokyo or Chiba?” People can make inferences based on their specific knowledge such as whether there are famous soccer teams in the cities.…”

“…For example, consider a binary choice question: “Which city has a larger population, Tokyo or Chiba?” People can make inferences based on their specific knowledge such as whether there are famous soccer teams in the cities. On the other hand, they can also use heuristics based on their memory such as which city they recognize (e.g., Goldstein & Gigerenzer, ), or which city they are more familiar with (e.g., Honda et al, ). The reason memory‐based heuristics work well in this situation (i.e., ecological rationality) is that heuristics are well suited to environments: Larger cities are often more famous because they appear in the media more often than smaller cities, and thus tend to be recognizable or familiar, so the strategy of choosing the recognized or more familiar city is likely to be rational (for a detailed examination of this issue, see Study 2).…”

“… Although the fluency heuristic (e.g., Hertwig et al, ; Schooler & Hertwig, ) is one of the well‐known heuristics, we did not examine this heuristic for the following two reasons. First, the fluency heuristic is highly difficult to disentangle from the familiarity heuristic in a theoretical sense (Honda et al, ). Second, Honda et al () showed that the familiarity heuristic model could explain participants' inference patterns better than the fluency heuristic model using the model competitive test. …”

Parameter Inference for Computational Cognitive Models with Approximate Bayesian Computation

“…Stocco () compared two competing action selection models, both using the ACT‐R cognitive architecture, and used a thoroughly reported grid search to fit the parameters. Honda, Matsuka, and Ueda () modeled binary choice in terms of attribute substitution in heuristic use, and fitted one parameter with grid search. There are also examples of heuristic extensions to grid search for improving the scalability.…”