Inferring Decision Strategies Based on the Path to a Choice

The simple heuristic framework proposes that people are equipped with a toolbox of decision strategies. Luan, Schooler and Gigerenzer (2014) investigated the theoretical properties of Δ inference, a simple heuristic that ranks alternatives (e.g., banks) according to a criterion of interest (e.g., risk of failure) based on cues (e.g., loan to capital ratio). It makes a decision when the value on a cue exceeds those of other alternatives by a threshold Δ. One question is under what conditions, if any, do people adopt Δ-inference? A methodological challenge is to recover the strategy used by a decision maker to make a choice. We present a machine learning approach that recovers the strategy used by considering not only the final choice but also the steps taken on the path to the choice. Three experiments have been conducted, in which we teach participants to use three decision making strategies: Take The Best, Tallying and Δ-inference. Participants are instructed to use a particular strategy to make decisions and their behaviors during the process are observed. We then build machine learning algorithms to analyze behavior data in order to recover their strategies. The results showed that we could differentiate the strategies that search information in parallel between two alternatives (e.g., Take The Best, Δ inference) from the strategies that summarize all information for all alternatives and compare (e.g., Tallying) accurately. However, the results of differentiating between Take The Best and Δ inference, where the latter is a generalization of the former, still needs improvement. The next step is to explore whether a richer set of features could help differentiate whether people are using Take The Best, Δ inference or some other strategy altogether