Dual-System Theories of Decision Making: Analytic Approaches and Empirical Tests

General Information

Dual-System Theories of Decision Making: Analytic Approaches and Empirical Tests
Aleksandr Sinayev
Publication Type
Dissertation (Bachelor/Master/Phd)
Ohio State University
Dual-system models are popular in the study of decision making. They have faced criticisms, especially for being vague and lacking specific predictions. In order to address these criticisms, three categories of dual-system models are reviewed: parallel-competitive (in which intuitive, system 1, and deliberative, system 2, processing happen at the same time and both influence the response), default-interventionist (in which system 1 executes first and then system 2 may or may not override system 1), and interactive (in which both systems process information at the same time, but they are allowed to influence each other in complex back-and-forth interactions). Tests are conducted of the former two categories.

Default-interventionist dual-system models predict that individual differences in reflectiveness should be associated with less biased decision making. The Cognitive Reflection Test (CRT) is thought to measure monitoring of system 1 intuitions such that, if cognitive reflection is high enough, intuitive errors will be detected and the problem will be solved. However, CRT items also require numeric ability to be answered correctly and it is unclear how much numeric ability vs. cognitive reflection contribute to better decision making. In two studies, CRT responses were used to calculate Cognitive Reflection and numeric ability; a numeracy scale was also administered. Numeric ability, measured with the CRT or the numeracy scale, accounted for the CRT's ability to predict more normative decisions (a subscale of decision-making competence, incentivized measures of impatient and risk-averse choices, and self-reported financial outcomes); Cognitive Reflection contributed no independent predictive power. Results were similar whether the two abilities were modeled (Study 1) or calculated using proportions (Studies 1 and 2). These findings demonstrated that correlations of decision performance with the CRT are insufficient evidence to implicate overriding intuitions, and, therefore, to support default-interventionist theories in the decision-making biases and outcomes we examined. Numeric ability appeared to be the key mechanism instead.

After failing to find evidence for a default-interventionist model of decision making, a mathematical parallel-competitive model, first introduced by Mukherjee (2010), was reviewed and tested. This model was expanded in order to improve its fit to choice data. An experiment was conducted, in which participants made risky choices. Half of the participants made choices under heavy cognitive load, and the other half under no cognitive load. The model’s key parameter indicating the use of system 1 vs. system 2 changed in the predicted way, but the model failed to account for important features in the data. The model’s assumed system 1 process was changed to be sensitive to probability, rather than outcome, after which the model was able to account for qualitative patterns of data more effectively and fit the data better, suggesting this is a better representation of system 1. In the final chapter, I conclude that some dual-system models are specific enough to be mathematically identifiable and to make a number of very specific predictions. I then discuss the possibility of generalizing the analytic approaches I employed to other decision problems and the limitations of these approaches.