Computational mechanisms for context-based behavioral interventions: A large-scale analysis

Abstract

Choice context influences decision processes and is one of the primary determinants of what people choose. This insight has been used by academics and practitioners to study decision biases and to design behavioral interventions to influence and improve choices. We analyzed the effects of context-based behavioral interventions on the computational mechanisms underlying decision-making. We collected data from two large laboratory studies involving 19 prominent behavioral interventions, and we modeled the influence of each intervention using a leading computational model of choice in psychology and neuroscience. This allowed us to parametrize the biases induced by each intervention, to interpret these biases in terms of underlying decision mechanisms and their properties, to quantify similarities between interventions, and to predict how different interventions alter key choice outcomes. In doing so, we offer researchers and practitioners a theoretically principled approach to understanding and manipulating choice context in decision-making.

Keywords: behavioral interventions; computational modeling; context effects; decision-making.

Fig. 1.

(A) Screenshots for exp.1, with an example of the baseline condition (Left) and an option salience intervention (Right). (B) Illustration of the DDM, with hypothetical changes to the starting point, drift rate and decision boundary parameters.

Fig. 2.

Observed and simulated intervention effects on choice probabilities and RTs in exp.1 and exp. 2. Gray points correspond to changes for individual participants. Colored labels correspond to aggregate changes for interventions, averaged over participants. Displayed correlations capture the relationship between observed and simulated changes on the individual level. Note: Participants with behavioral shifts outside of the range of the x and y axes of this figure are shown in SI Appendix, Fig. S11.

Fig. 3.

Effects of behavioral interventions on (A) the start point, (B) the drift rate, and (C) the decision boundary. Positive (negative) starting point and drift rate effects correspond to biases favoring the high-quality (low-price) option. HQ and LP denote interventions selectively targeting the high-quality or low-price options in a trial. Results are based on group-level parameters in each condition.

Fig. 4.

(A) Three-dimensional space of behavioral interventions based on their absolute, standardized effects on the starting point, drift rate, and decision boundary parameters. (B) Cognitive effect sizes of the interventions. These are based on the distance between an intervention and the origin of the space. (C) Hierarchical clustering of intervention vectors (averaged across exp.1 and exp. 2). Results are based on group-level parameters in each condition.