Evidence-Accumulation Models: Applications II
Dr. Sophie Bavard
Sequential sampling models (SSM) assume that decisions for oneself are made by accumulating evidence for the available options, until the choice of one option has been triggered. As such, it seems plausible that predicting the decisions of another individual could be achieved similarly, by simulating the other person’s evidence accumulation process with one’s own cognitive mechanisms – the ‘simulation hypothesis’. Although SSMs have been successfully used to describe predicted decisions of other agents, that in itself is insufficient evidence in support of the simulation hypothesis. However, one concrete implication of the hypothesis is that, if people use their own mind to simulate and predict the decisions of others, then biasing an individual’s own decision-making process should result in a corresponding bias when they make predictions. To test this, we first biased participants’ risk perception (n = 172) by adapting them to a high-risk/low-risk context when they made decisions for themselves, and subsequently asked them to predict the decisions of a risk-seeking and risk-averse (hypothetical) agent in a medium-risk context. On average, participants in the high-risk adaptation group predicted a more risk-accepting behaviour than those in the low-risk group, indicating that the predicted decisions were made with participants’ own biased decision-making system - consistent with the simulation hypothesis. Surprisingly, this effect was only present during prediction for the risk-averse, but not risk-seeking agent, suggesting that the mechanisms of prediction may depend on the specific characteristics (e.g., similarity) of the other person. We also fit our data with a version of the drift diffusion model devised to account for risky decisions, to understand how the seemingly different mechanisms of prediction across agents are captured in the SSM framework.
This is an in-person presentation on July 21, 2023 (15:20 ~ 15:40 UTC).
Hunger is a biological drive, with the function of motivating a mechanism to eat to reach homeostasis. Hungry participants are particularly likely to choose hedonic food options. Here we apply a version of a sequential sampling model to elucidate the mechanisms underlying the hunger-driven impairment in healthy dietary choice. We implemented a binary food choice task, in which two food images (representing tastiness of the option) and their respective Nutri-Score (representing healthiness of the option) appeared on the screen. Participants completed the task in a hungry and a satiated state (within-subject design) while their eye-movements were being recorded. In line with our hypothesis, behavioral evidence demonstrated that participants were more likely to choose tasty over healthy food items, and this difference was amplified by hunger state. To identify hunger-driven effects on decision processes, we used an extension of the Drift Diffusion Model, the multi attribute-time dependent Drift Diffusion Model (mtDDM) (Sullivan & Huettel, 2021, Nat Hum Behav), which allows the options’ underlying attributes (here taste and health) to influence the decision process with different latencies and different weights. Applying the mtDDM to our data, we found that in both conditions’ health latencies were significantly later than taste latencies and health weights were significantly lower than taste weights. When comparing conditions, there was no significant influence of hunger state on the attributes’ latencies. However, we found that health weights were significantly reduced in the hungry compared to the sated condition, while taste weights were unaffected. This suggests that poor dietary choices under hunger are driven primarily by an impairment in health consideration in the decision process. Notably, our modeling results also revealed that the mtDDM predicts that more than 20% of the estimated responses are made faster than the estimated latencies, that is, before any attribute information comes into play. While the purpose of the mtDDM is to predict multi-attribute choice, we would argue that it may not be suited for tasks in which the underlying attributes are represented distinctly. Further analyses of the eye-tracking data, combined with different implementations of dynamic process model of decision making will extend our understanding of the effects of hunger on attentional dynamics and preference formation in dietary decision making.
This is an in-person presentation on July 21, 2023 (15:40 ~ 16:00 UTC).
Dr. John Buckell
Discrete choice experiments in healthcare typically include only a small number of choice tasks for each participant to minimise the burden on responders. The complexity of healthcare issues means that choice tasks often include many attributes and/or alternatives. Consequently, individual-level models are not possible. Differences in behaviour, preferences, and decision-making processes must be captured with the use of more complex behavioural models. The inclusion of sociodemographic parameters in the model aids the detection of deterministic heterogeneity, while stochastic heterogeneity is typically captured using random parameters or latent class structures. However, neither of these methods are particularly well-suited to disentangling confounding sources of heterogeneity, that is, to detect individuals who make decisions in different ways separately from the identification of different preferences. In the current work, we address this issue using a discrete mixture model. This model estimates probabilities for each individual having particular tastes, and separately estimates probabilities of using different decision-making processes. In particular, we develop a discrete mixture decision field theory (DMDFT) model to capture preference heterogeneity through different attribute importance parameters whilst accommodating the different processing speeds of decision-makers and propensities to be subject to the similarity effect through different process parameters. We apply the model to datasets from discrete choice experiments on tobacco preferences. We demonstrate that (a) DMDFT models outperform latent class DFTs, providing clearer insights on individual differences (b) the models outperform their counterpart econometric choice models, and (c) the models find clear differences in the individuals’ decision-making processes as well as preferences.
This is an in-person presentation on July 21, 2023 (16:00 ~ 16:20 UTC).
Often in our lives, we need to take a series of sequential actions rather than making a single, isolated decision. A typical everyday example is to decide which and how many items we want to put in our shopping basket. Multiple previous studies have investigated the temporal dynamics of these sequential decisions using a virtual shopping paradigm (Wolf et al., 2018, 2019; Xu et al., in press). Specifically, these studies have examined the probability and speed of dietary decisions with various constraints but did not further study the underlying cognitive mechanisms. Here, we show how a process model based on the sequential sampling modeling framework can elucidate those mechanisms and provide a comprehensive account of the reported choice and response time (RT) patterns. In the virtual shopping paradigm, participants decided for sequentially encountered items whether or not to add them into their shopping basket (“pick” or “leave”). The number of food items that could be added to the basket was limited and manipulated in one experiment. In some conditions, participants were given the opportunity to defer the food item by placing it on a waiting list. Importantly, “pick” decisions tended to be longer than “leave” decisions, and this difference was most pronounced when fewer food items could be selected. Decisions to wait were slower than both “pick” and “leave” decisions. On the inter-individual level, the RT difference between “pick” and “leave” decisions was higher for participants who rejected more options. We account for these choice and RT patterns using a new variant of the Feed-Forward Inhibition model (FFI) (Shadlen & Newsome, 2001) with three separate accumulators for the decisions "leave", "pick", and "wait". The “leave” and “pick” accumulators, but not the “wait” accumulator, inhibit each other. Due to the limited basket space, we assume that participants have a default preference for leaving the food item, which we implement by shifting the starting point of the accumulation process in favor of “leave” vs. “pick” decisions as a function of the basket size. Our simulations show that this model can account for the various choice and RT patterns in the virtual shopping paradigm, including the RT difference in “leave” vs. “pick” decisions, the excessively long “wait” decisions, and the dependency of these differences on the basket size. In addition, when inspecting our simulations, we see that the drift rate of “pick”-choices is considerably higher than average, mimicking the fact that only high-value food items are chosen for the basket. In contrast, the drift rate of “wait”-choices is only slightly above average, suggesting that food items which are of a somewhat higher value are put on the waiting list. Essentially, by simulating the data, we were able to reproduce the main effects of the above-mentioned studies and provided mechanistic explanations for sequential shopping decisions. Until the start of the conference, we also plan to replicate the paradigm and fit our model to the data. Shadlen, M. N., & Newsome, W. T. (2001). Neural basis of a perceptual decision in the parietal cortex (area LIP) of the rhesus monkey. Journal of neurophysiology, 86(4), 1916-1936. Wolf, A., Ounjai, K., Takahashi, M., Kobayashi, S., Matsuda, T., & Lauwereyns, J. (2018). Evaluative processing of food images: a conditional role for viewing in preference formation. Frontiers in Psychology, 9, 936. Wolf, A., Ounjai, K., Takahashi, M., Kobayashi, S., Matsuda, T., & Lauwereyns, J. (2019). Evaluative processing of food images: longer viewing for indecisive preference formation. Frontiers in Psychology, 10, 608. Xu, J., Jin Y., Lauwereyns, J. (in press). The Framing of Choice Nudges Prolonged Processing in the Evaluation of Food Images. Frontiers in Psychology
This is an in-person presentation on July 21, 2023 (16:20 ~ 16:40 UTC).
Dr. Marieke Van Vugt
Prof. Hamidreza Jamalabadi
Major depressive disorder is characterized by among others difficulty in letting go of negative and self-deprecating thoughts. To allow for a deeper mechanistic understanding of rumination and depression more generally, cognitive tasks and computational modeling play a major role. We have shown that thinking that is difficult to disengage from, such as rumination, can be captured in tasks that track spontaneous thinking. In fact, we demonstrated that this sticky thinking is associated with increased alpha oscillatory power. Moreover, using alpha power in the drift diffusion model shows that high alpha is associated with a lower drift rate, suggesting that sticky thinking disrupts the decision making process. Using a very different kind of modeling, a dynamical systems analysis shows that sticky and self-related thinking is more difficult to control.
This is an in-person presentation on July 21, 2023 (16:40 ~ 17:00 UTC).