Evidence Accumulation: Multi-Attributes, Multi-Responses, And Complexity

The impact of visual attention on choice processes has been established over the last decades. Several studies are consistent with the view that visual attention increases the subjective value of the attended option. However, a few computational models have been proposed to investigate how attention and subjective values interact in multi-attribute choices. Moreover, these models disagree in terms of whether value is modulated by attention additively or multiplicatively. The additive theory states that the boost up subjective value depends only on gaze duration, and gaze on an option magnifies the subjective value at a constant rate. On the other hand, the multiplicative theory assumes that the magnitude of the attention-driven boost is value-dependent, and gazing at a high-value option yields a more significant boost in subjective value. Although there is a long debate on these two theories, recent studies have shown that both additive and multiplicative interactions between subjective value and gaze time may be essential for explaining empirical data and have suggested hybrid theories. For multi-attribute decisions, however, extant attentional models only consider the multiplicative interaction. This work introduces a new computational framework to account for attention in multi-attribute decisions. Our model assumes a hybrid attentional mechanism for the interaction between subjective values and gaze duration. We have tested the model on four datasets from various domains (e.g., clothing/brand, food/nutrition, food bundle, and money risk tasks). The results from the nested model comparison show that the proposed hybrid model works better than the other computational models.

This is an in-person presentation on July 20, 2024 (14:00 ~ 14:20 CEST).

The literature on value-based decision-making often focuses on single-item selections. However, many decisions like grocery shopping or forming a team, involve choosing multiple items. These “multi-response” decisions require selecting multiple options from a set. Despite their prevalence in everyday life, little work has examined the cognitive mechanisms underpinning multi-response decision-making. Our study extends the Sequential Sampling Model (SSM) framework to test competing strategies that people may use in multi-response choice. In our task, subjects were asked to choose one, two, or three food items out of a set of four. Our subjects generally chose their highest ranked foods (based on independent ratings). Interestingly, they also generally chose higher ranked items first. Subjects’ response times for their first selected items were similar to their response times when choosing a single item. However, subsequent responses were made more quickly. Together, these results suggest that in multi-response choice people may evaluate their options in parallel, allowing them to gather support for multiple items at once, and leading to the fastest responses for the best items. Model simulations confirm that an SSM account can capture these patterns, while several alternative models cannot. Together, our work offers a starting point for studying the dynamics of multi-response decisions.

This is an in-person presentation on July 20, 2024 (14:20 ~ 14:40 CEST).

Response time is often used as an indicator of how complex a problem is in various settings, such as lottery choice, loans, and portfolio selection. This is sometimes justified by is the observation that, in many cases, faster choices tend to be more accurate. Existing evidence suggests a more ambiguous relationship between speed and accuracy, wherein sometimes faster decisions are better, and in other instances slower decisions are more accurate. In this paper, we reconcile this ambiguous relation between speed and accuracy by revisiting a standard Wald problem of optimal stopping. We show that whilst choice quality is monotone in problem complexity (noise-to-signal ratio), expected stopping time is inverse U-shaped. This suggests a nuanced relation between speed and accuracy: in simple problems, people choose fast and well; in slightly more complicated ones, they choose slower and less well; but if they become much more complicated, choices are necessarily worse, but response times are now faster. Extending our model to dynamic effort control, we uncover that this non-monotonicity also suggests restraint in using response times to infer ability: high ability individuals may choose faster in simple problems and slower in more complex ones. Moreover, we show that this non-monotonicity is a generic feature of models with costly information acquisition. Finally, we leverage our results to examine the effect of incentive distortion on behaviour. We find that incentive distortion is more effective in steering choices in more complex problems, opening the way to infer problem complexity from simple manipulations of incentives.

This is an in-person presentation on July 20, 2024 (14:40 ~ 15:00 CEST).

In evidence accumulation models of value-driven choice, drift rates are known to reflect the relative preference of the decision-maker over the alternatives. Although common evidence accumulation models - such as the DDM and LBA - are equally able to fit data for a given choice set, they generate diverging predictions about the effect of increasing the value of an alternative and their calibrated drift rates are not easily comparable. In this paper, I clarify theoretically the relation between drift rates across evidence accumulation models. I characterize evidence accumulation models by their range - the set of choice and response time distributions that they can generate - and their contrast - the extent to which increasing the value of one alternative slows down the choice of another. Common evidence accumulation models have a similar range, as my simulations reveal, but a drastically different contrast. Since the correct level of contrast is an empirical question, I propose a tractable framework to calibrate it on an existing model. I also give general conditions under which this approach is applicable. Evidence accumulation model with a similar range generate similar predictions once their contrast is properly calibrated. I calibrate a LBA model and its contrast on data generated by a DDM with varying alternatives values and find the level of contrast predicted by theory for a DDM. Overall, this paper sheds a new light on the long-lasting debates on model equivalence and mimicry from the perspective of value-driven choice.

This is an in-person presentation on July 20, 2024 (15:00 ~ 15:20 CEST).

The pervasive challenge of information overload often leads decision-makers to avoid complex options in favor of simpler to understand alternatives, even if the complex options are more rewarding. However, previous research only focused on the behavioral phenomenon of complexity aversion. Here, we provide a drift diffusion modeling approach to better understand how complexity affects decision making in a different experimental environments. In our experiment, we use compound lotteries to examine decision making not only between complex and simple options but also in situations where both options are either complex or simple. Through the application of drift diffusion models, we aim to elucidate the cognitive mechanisms at play during the decision-making process. As results we found no support for complexity aversion when contrasting complex with simple options. However, comparing decisions between complex and simple options, we found effects of complexity on decision making in response times, choice consistency, risk taking, and the effect of rare outcomes (or skewness) on decisions. Building on these behavioral effects, we employed drift diffusion models to test the cognitive processes through which complexity affects decision-making. This approach reveals significant effects of complexity on risk preference, choice consistency, and the subjective interpretation of probabilities. The new experimental results and the computational modeling advance our understanding of how complexity impacts risk-taking behavior and delineates the cognitive processes involved.

This is an in-person presentation on July 20, 2024 (15:20 ~ 15:40 CEST).