Real-World Decisions

Computational models of behavior are key in enhancing road safety by offering insights into human behavior that extend beyond empirical studies. Particularly, generalized drift-diffusion models (GDDMs), that is, DDMs with time-varying drift rates and decision boundaries, have deepened our understanding of decision making in traffic interactions. However, the application of GDDMs to traffic decisions has mostly focused on scenarios in which the decision-maker is stationary and the other traffic participants move with constant velocities. This restricts the application of such work in real-world traffic contexts in which the decision maker is often moving during the decision-making process and the other traffic participants may exhibit time-varying dynamics. To address this gap, we developed a GDDM-based account of human drivers’ decision-making during overtaking maneuvers. In a driving simulator experiment (N=30), we varied the size of the gap available for the participants to overtake a leading vehicle. We assessed how this gap size and the dynamics of an oncoming vehicle impacted the decision outcomes and response times of the participants. Our empirical findings underscore the critical role of initial distance, time-to-arrival, and velocity of participants’ vehicle in overtaking decisions. By fitting four candidate GDDMs to the observed data, we found that participants’ behavior was best described by a model hypothesizing dynamic drift rate, constant decision boundaries, and a decision bias dependent on the initial velocity of participants’ vehicle. Overall, our findings highlight the potential of drift-diffusion models with time-varying components for further advancing the understanding of human behavior during dynamic traffic interactions.

This is an in-person presentation on July 21, 2024 (10:00 ~ 10:20 CEST).

The shooting of unarmed Black males by police officers is a topic at the forefront of public awareness in the U.S. today. It is widely believed that police shootings reflect racial bias on behalf of officers, which erodes public trust and reduces policing effectiveness. We introduce a novel framework—the Attention-integrated Model-based Shooting Simulator (AiMSS)—to study how race, suspect behavior, and policing scenarios impact police officers' decision to shoot. The AiMSS combines computational models of decision making, visual psychophysics, eye-tracking methods, social measures of affective evaluations, and an immersive decision simulator to map the processes underlying a police officer’s decision to shoot. We will summarize work from across several datasets with police officers completing a first-person shooter task in the AiMSS. Overall our behavioral and cognitive modeling results reveal that (a) policing scenarios and suspect behavior played an essential role in officers' decisions; (b) that errors are higher for unarmed than armed suspects, with some evidence for greater errors for Black vs White suspects; (c) this race effect is determined partly by an initial bias and diminished sensitivity during the decision, which are linked to perceived threat; and (d) training largely mitigated these race effects. This work provides a novel method for understanding the mechanisms underlying the decision to shoot, in terms of how different sources of information are integrated and how they impact different components and stages of decision making.

This is an in-person presentation on July 21, 2024 (10:20 ~ 10:40 CEST).

Cognitive process models such as the diffusion decision model allow researchers to estimate a set of parameters from empirical response times and accuracy data obtained in binary decision tasks. These parameters can then be used to quantify individual differences. They are thought to reflect, for example, a person's speed of evidence accumulation or decision caution in a certain task. Recently, researchers have become interested in how the model parameters might be related to other measures of individual differences, specifically general intelligence. For the drift rate parameter, higher parameter estimates seem to be linked to higher intelligence scores. However, cognitive abilities such as general intelligence are known to predict socioeconomic success (e.g., educational attainment, job prestige, income). If drift rates reflect a type of cognitive ability, they should also exhibit similar patterns. We thus studied the associations of diffusion decision model parameters with several indicators of socioeconomic success in a very large sample of online implicit association test data (Project Implicit; N > 5,000,000). We found robust associations between diffusion decision model parameters and indicators of socioeconomic success marked by small effect sizes. Our results highlight the utility of big data approaches in the field of cognitive modeling that have only recently become practically feasible through novel simulation-based inference methods.

This is an in-person presentation on July 21, 2024 (10:40 ~ 11:00 CEST).

How much debt and how much equity firms choose to finance their operations by balancing the costs and benefits is a fundamental question in corporate finance. The dynamic trade-off theory of capital structure suggests that a firm selects an optimal ratio of its liability to asset value (i.e. an optimal leverage ratio) to balance the dead-weight costs of bankruptcy and the tax saving benefits of debt. In other words, in the presence of adjustment costs, firms try to set relatively stable targets but tolerate deviations from these targets as long as leverage ratios stay within their target zones. Inevitably, such behavior is expected to affect the dynamics of leverage ratios. To assess the extent to which and how leverage dynamics are driven by such behavior, we apply a modified version of the Leaky Competing Accumulator model of decision making, which allows cooperation in addition to competition among firms, to investigate the dynamics of corporate leverage ratios and determine the target leverage ratios. Since the multi-firm joint probability density function of leverage ratios is known in closed form, a likelihood function can be constructed and thus model-fitting against empirical data becomes feasible. In both automotive and integrated oil industries firms are found to cooperate to fix their target ratios, and competition dominates between conventional automotive companies and electric car companies. Moreover, analytical default probabilities of these firms are available and their default risk can be estimated. Hence, the impact of climate-change policy on these firms are examined as well.

This is an in-person presentation on July 21, 2024 (11:00 ~ 11:20 CEST).