The future of research into decision-making under stress

The book Thinking, Fast and Slow, by Daniel Kahneman, put forward the idea that we reason using not one but two systems, System 1 and System 2, which are fast and slow, respectively. According to Kahneman’s model, we use System 1 when stress is very low or very high and we use System 2 when stress is moderate. Studies that explore how we use these systems for decision-making in various stressful conditions have recently been summarized in a systematic review that asks three questions: What experimental protocols should be used? How should physiological responses to stress be measured? How can physiological measurements be used to represent stress quantitatively? The open access review article, titled “Measurement and Quantification of Stress in the Decision Process: A Model-Based Systematic Review”, was published Sep. 18 in Intelligent Computing, a Science Partner Journal.

Review motivation
Although decision-making is important and commonly studied, many reviews have focused on studies based on subjective evaluations of stress, which are quantitative but subject to bias. In contrast, measuring the automatic, irrepressible physiological signals that result from the body’s fight-or-flight response to stress has the potential to reveal stress levels objectively. Measurable physiological signals include heart rate, blood pressure, respiration rate, cortisol level, and galvanic skin response, among others. These signals are increasingly able to be measured using convenient wearable devices, and modern machine learning methods are increasingly being used for the processing and analysis of data.

Insights into experimental design
The review presents information about the participants of the study, highlighting the role of biological sex in stress response. It catalogs and describes the types of stressors used as the experiment tasks. Many studies used some form of public speaking or some form of mental arithmetic. Many studies used a version of the Trier Social Stress Test, which involves both. In this task, the experimental participant has 5 minutes to prepare a presentation, then has 5 minutes to give the presentation to a non-reactive panel of listeners, then has to do mental arithmetic for 5 minutes. Other studies used the Stroop test, which requires the experimental participant to say aloud the color in which a color word is displayed, rather than the word itself. Still others, fewer in number, used ill-defined and open-ended tasks that required creative problem-solving. These tasks, being less artificial, better reflect real-world tasks, but participant selection and data collection for such tasks is more difficult.

Insights into stress measurement
One effective tool for measuring physiological response to stress is functional near-infrared spectroscopy. It measures blood oxygenation levels in the brain, and can pinpoint activity in different brain regions in high resolution. Another tool is electroencephalography, which uses wired or wireless systems connected to varying numbers of scalp electrodes to capture electrical activity in the brain. Activity is tracked according to where and when it occurs, and at what frequency bands. Eye-tracking can also reveal stress through measurements of eye movements, pupil diameter and blink rate. Electrocardiography offers several metrics for tracking different components of the cardiac cycle, including but not limited to heart rate. Electrical responses can also be measured via a device on the wrist or hand that measures skin conductivity, a reflection of the sweat that is reliably produced as a stress response. Blood pressure, which temporarily increases with acute stress, was also measured in some studies. Endocrinological measurements can be used to detect stress response, as glands produce hormones in response to stress. To assess endocrine response, testing levels of salivary cortisol is a popular technique, as it is non-invasive compared to hormone testing that requires the collection of blood or urine. The review comments on how convenient these measurements are and how commonly they are used and summarizes how each is related to stress response.

Insights into stress quantification
Stress is commonly believed to be related to four factors: perceived task workload, affect, skills and knowledge. However, no exact mathematical relationship has been established. In particular, stress can be quantified as present or absent, as having two levels, or as having three or more levels. These levels should be standardized for the sake of making comparisons of results for the same individual at different times or different individuals at the same time. Machine learning, which is increasingly being used for data analysis, may help establish accurate baselines. The random forest, support vector machine, and k-nearest neighbor methods were the most common, reflecting a preference for traditional classifiers, which can achieve high levels of accuracy. However, a diversity of methods should be explored.

Future directions
While the study revealed an increase in research on stress in decision-making, it also revealed important gaps, suggesting directions for future research. For example, System 2 experiments are relatively few, due to the complexity and unpredictability of the experimental tasks. The review culminates in ten unresolved questions that can be fruitfully explored as research into human decision-making under stress continues to evolve, enabled by advancements in hardware and software tools as well as insights from previous studies.