Modelling the influence of situational uncertainty on risk taking in everyday life

Individuals make countless decisions that involve evaluating uncertain outcomes every day. The resulting behavior, often referred to as risk taking, has been studied for decades, with a strong focus on trait-like predictors of interindividual differences, such as the construct of risk preference. Yet, about 50% of variance in individual’s risky choices cannot be explained with stable predictors, thus raising the question what situational factors affect risk taking. With the current study, we investigate one potential mechanism causing variation in risk taking between different choice-situations, specifically the perception of uncertainty. Individuals intuitively distinguish epistemic uncertainty, referring to lacking knowledge about the world, from aleatory uncertainty, caused by the innate randomness of the world. Previous laboratory research has found that people become increasingly risk-averse when their perception of uncertainty is more epistemic, rather than aleatory, however, it remains unclear if this tendency generalizes to real world decisions. We are tackling this shortcoming by tracking a person’s decisions during their everyday life with an experience sampling study. This allows us to model participants’ decision making throughout the day using Bayesian multilevel models, and show how different levels of epistemic uncertainty can predict risk taking. Additionally, we are collecting data on individuals’ perception of situational uncertainty with a classic, yet often neglected method: Participants record think-aloud protocols, describing decision-situations as they experience them. With the resulting speech-data, we investigate two novel research questions. First, we use participants’ verbal descriptions of choice-situations to quantify the degree of uncertainty an individual faces, and to thus predict variability in risk taking from these estimates. Second, we explore a relatively new way of using natural language data to model which features of a situation are relevant, salient, or accessible to individuals when making decisions. With that, we show how semantic information such as word embeddings can be used for inferring cognitive processes underlying risk taking, or other decision-processes.