Investigating the impact of individual differences and stress on decision-making performance under threat

Abstract

Quality of life largely depends on the outcomes of our decisions. A common model for understanding decision-making is reinforcement learning. Reinforcement learning involves deciding how to behave based on the appraisal of the situation and the post-evaluation of the positive or negative outcomes of decisions. In reinforcement learning, the difference in appraised value between expectations and actual outcomes is referred to as reward prediction error (RPE). Dopaminergic neuronal firing activity in the midbrain has been shown to encode RPE. RPE is used as a signal to guide decisions; for example, when decision outcomes are worse than expected (negative RPE), then those decisions are subsequently avoided. In contrast, when decision outcomes are better than expected (positive RPE), then those decisions are likely to be repeated. There is limited research explaining how individual differences such as age, gender, years of education, history of acute and chronic stress, and personality might impact decision-making performance under threat. For example, although a certain level of stress can be adaptive and improve cognitive and physical performance, including decision-making, prolonged and repeated exposure to stress has been negatively associated with both mental and physical health and longevity. As such, a history of acute or chronic stress might impact decision-making under threat; however, the interrelationship between individual differences, stress and decision-making under threat is still poorly understood. This thesis attempts to synthesise and expand existing knowledge regarding the relationship between decision-making performance, individual differences and stress. Hence, a novel decision-making task was designed and deployed in order to test the ability to learn from positive and negative RPE during safe and threatening conditions. The decision-making task, along with self-rated surveys associated with individual differences in demographics, personality, and history of acute and chronic stress, were delivered both online (N=109, M=37.09, SD= 10.9 years), using a crowd sourcing platform, and within a laboratory setting (N=107, M= 19.42, SD= 3.77 years). In the online experiment we identified several significant linear regression models predicting the overall average of win-stay (i.e. correctly staying with a choice following a positive RPE) and lose-switch (i.e. correctly switching a choice following a negative RPE) performance across both safe and threat conditions. One of such models having age, gender, years of education, personality, acute and chronic stress factors as predictive variables, explained 34.7% of the variance in overall average win-stay performance across safe and threat conditions, and 30.8% of the overall average lose-switch performance across safe and threat conditions. In the lab experiment, we identified significant linear regression models predicting the difference in mean win-stay and lose-switch performance between threat and safe conditions. One of such models having age, gender and years of education as predictive variables, explained 10.2% of the variance in the difference of lose-switch performance between threat and safe conditions. Another model having age, gender, years of education, personality, acute and chronic stress factors, as well as the difference in heart rate between threat and safe conditions, as predictive variables, explained 22.0% of the variance in the difference of win-stay performance between threat and safe conditions. Such findings contribute to the body of knowledge regarding the impact of individual differences and stress on decision-making performance under threat, and could guide the design and development of stress management prevention and intervention decision support systems.