Forschungskolloquium

Many everyday decisions, such as buying a phone or choosing a hotel, require integrating information from multiple attributes to identify the best option. Our recently proposed Multi-Attribute Search & Choice (MASC; Gluth et al., 2024) theory suggests that this process follows a hierarchical Bayesian belief-updating scheme, where posterior beliefs about attribute values, and consequently about options, are iteratively updated based on information sampled during fixations. Once one option is deemed sufficiently superior, a choice is made. In this EEG and eye-tracking study, 57 participants performed a multi-attribute choice task, choosing between two smartphones based on star ratings for three attributes (battery capacity, screen size, storage space). We fitted MASC to individual eye-tracking data and, by simulating the model with the best-fitting parameters, derived predictions about the strength of belief updating at each fixation. Using linear deconvolution modeling, we used these model-derived estimates to predict EEG activity evoked by fixations, while also controlling for overlapping activity evoked by successive fixations and confounds such as saccade amplitude. Our findings demonstrate that MASC not only captures key behavioural patterns but also provides meaningful predictions at the neural level. Regression analyses, corrected for multiple comparisons using threshold-free cluster enhancement, revealed that predicted belief updating at the attribute level was associated with a transient, P3-like modulation over central electrodes, while belief updating at the option level corresponded to a later, more sustained centroparietal positivity. These findings offer converging behavioural and neural evidence that multi-attribute decision-making involves dynamic, fixation-driven belief updating consistent with Bayesian inference as formalized by MASC.

Compulsive-use disorders like binge eating affect millions and impose substantial costs through psychosocial impairment and health risks. Despite their prevalence, the cognitive mechanisms maintaining compulsive behavior remain poorly understood. Individuals with binge eating preferentially pursue highly palatable foods while forgoing alternative rewards, suggesting aberrant reward processing. However, prevailing models fail to explain how this altered reward processing maintains compulsive eating or how emotional distress modulates these processes.This talk presents a neurocomputational framework proposing that impaired reward discriminability—the ability to detect differences between potential rewards—drives maladaptive over-selection of highly palatable foods in binge-type eating disorders. I will first present behavioral and neural evidence showing that healthy adults exhibit enhanced discriminability for high-value versus low-value rewards, motivating a computational model of value-sensitive decision-making. Next, I will demonstrate how negative emotions bias food choices toward taste over health considerations, with this effect amplified in bulimia nervosa. Finally, I will discuss ongoing work testing whether individuals with binge eating show reward-type-specific discriminability deficits: enhanced for food rewards but impaired for non-food rewards. Together, these findings illuminate the neurocomputational mechanisms underlying maladaptive reward seeking and suggest novel intervention targets for eating disorders.

Humans exhibit flexibility in how they search for information and make choices to align with different decision goals. One possible mechanism underlying this flexibility is the transformation of option values from preference-based to goal-congruent representations. Previous studies support this view, showing that the relationship between the overall value (OV) of the available options and response times (RTs) is modulated by the current decision goal: people tend to make faster decisions when choosing between goal-congruent options than goal-incongruent options. However, these studies mainly focused on two decision goals—choosing the best or worst option—their findings only suggest that values are transformed linearly. Consequently, it remains unclear whether such value transformations can be non-linear under certain goals, such as choosing the most mediocre or extreme option. In this talk, I will first use a computational model to demonstrate how goal-congruency might be involved in information processing. Later on, I will show our online study in which we asked participants to make decisions based on four decision goals: choosing the best, worst, most mediocre, or most extreme option. The preliminary regression analyses revealed that choices were fastest when the overall value of the options was closest to the target for the choice goal. Specifically, replicating previous findings, RTs decreased with increasing/decreasing OV when the target was to choose the best/worst option. Importantly, participants responded fastest at intermediate OV level when the target was to choose the most mediocre option. This U-shape relationship between RTs and OV was reversed in the extremity condition - responses were fastest when OV was at either end of the rating scale. Together, our findings suggest that flexible decision-making relies on goal-dependent value representations that can be transformed in both linear and non-linear ways.

Evidence accumulation models (EAMs) have become the dominant theoretical framework of human speeded decision-making, serving as both a theory capable of explaining a range of empirical trends, and a measurement tool capable of estimating cognitive constructs in applied settings. While several different model variants within the EAM framework contain fundamentally different ideas of how the decision-making process operates, previous assessments have found that these models display a high level of mimicry, which has hindered the ability of researchers to contrast these different theoretical viewpoints. Here, I discuss the idea of moving beyond the standard, singular-time measurements of response time and accuracy, and moving towards more continuous-time measurements of decision-making. Importantly, the additional information contained in measurements from multiple points in time help to distinguish between the predictions of existing models, and allow for the creation of more sophisticated models of the decision-making process, which I discuss through examples from previous work on double responding and EMG recordings. Finally, I discuss the potential for truly continuous-time measurement, what this would mean for models of decision-making, and how these measurements might be practically achieved.

In many real-world settings, individuals make decisions between options that differ not only in risk and reward but also in complexity. However, the role of complexity in preferential decision-making remains poorly understood. Building on the drift diffusion model, we propose a process-level account in which complexity reduces the signal-to-noise ratio of evidence accumulation. This reduction prompts decision makers to deliberate longer and to favor simpler options before and during the decision process when such options are available. We tested these ideas in three preregistered experiments, manipulating complexity through multiple outcomes (Study 1), mathematical expressions of outcomes (Study 2), and compound probabilities (Study 3). Across all studies, complexity led to slower responses and noisier choices. Simpler options were chosen more often when available in Studies 1 and 2, but not in Study 3. Model comparisons showed that these effects were best captured by reduced signal-to-noise ratios and higher decision thresholds when both options were complex, and by additional starting-point biases and drift rate adjustment towards the simpler option when present. These findings clarify how complexity shapes risky choice.