Stress changes risk-taking by altering Bayesian magnitude coding in the parietal cortex
Poster No:
908
Submission Type:
Abstract Submission
Authors:
Maike Renkert1, Gilles de Hollander2, Gökhan Aydogan1, Saurabh Bedi3, Christian Ruff1
Institutions:
1University of Zurich, Zurich, Zurich, 2UZH, Zürich, Switzerland, 3Department of Neuroeconomics, University Zurich, Zurich, Switzerland
First Author:
Co-Author(s):
Introduction:
Acute stress is an inevitable aspect of life, with long-lasting consequences for physical and financial well-being (DeLongis et al 1988). Previous research documents a link between acute stress and altered risk preferences (Buckert et al. 2014), with suggestions that stress may contribute to the perpetuation of poverty (Haushofer & Fehr 2014). However, despite ample research, little is known about the neurocognitive processes that translate stress into altered risky decision-making. Here, we shed light on these processes in an experimental study of risky choices under laboratory-induced stress. We employed a perceptual account of risky choice, assuming that decision-makers make financial decisions based on noisy and biased perceptual representations of payoffs (Khaw et al., 2020; Garcia et al. 2023). This allowed us to decompose stress-induced shifts in risk preferences to latent Bayesian perceptual processes, specifically to either noisier sensory representation or altered beliefs. Importantly, those cognitive variables have neural surrogates that relate to how numerosities are represented in the parietal magnitude processing system that can be derived from fMRI data.
Methods:
Participants (n=50) performed a risky gamble task in a first baseline fMRI session, before being randomly assigned to either a stress or control group. For the risks task, each payoff is presented separately as cloud of coins to enable (numerical) population receptive field mapping later on.
To induce stress in the second fMRI session, we interleaved the risk task with an adapted version of the well-established MIST with additional social evaluative threat. Cortisol levels from saliva samples across 6 timepoints were collected to obtain a physiological measure of stress (overview of study design in fig.1).
Behavior was analysed with a cognitive model that assumes Bayesian perception of payoffs (noisy logarithmic coding model - NLC). Crucially, this enables disentanglement of noise and bias from decisions, which we could relate to neural measures in the next step. For that, we used an encoding/decoding framework that builds on the idea of neural populations that are tuned to numerosity, so-called numerical population receptive fields (nPRFs), which lie predominantly in the intra-parietal sulcus (IPS; example in fig. 2).
Here, the precision with which one can decode the number from neural data given the model parameters relates to the noise with which numbers are represented in the parietal magnitude processing system. Bias, in turn, was measured as a shift in neural coding, quantifying how much neural population representations of numerosities change in a specific direction for one subject between sessions.
To induce stress in the second fMRI session, we interleaved the risk task with an adapted version of the well-established MIST with additional social evaluative threat. Cortisol levels from saliva samples across 6 timepoints were collected to obtain a physiological measure of stress (overview of study design in fig.1).
Behavior was analysed with a cognitive model that assumes Bayesian perception of payoffs (noisy logarithmic coding model - NLC). Crucially, this enables disentanglement of noise and bias from decisions, which we could relate to neural measures in the next step. For that, we used an encoding/decoding framework that builds on the idea of neural populations that are tuned to numerosity, so-called numerical population receptive fields (nPRFs), which lie predominantly in the intra-parietal sulcus (IPS; example in fig. 2).
Here, the precision with which one can decode the number from neural data given the model parameters relates to the noise with which numbers are represented in the parietal magnitude processing system. Bias, in turn, was measured as a shift in neural coding, quantifying how much neural population representations of numerosities change in a specific direction for one subject between sessions.
·Study design
·Numerical population receptive fields of an example subject
Results:
The stress induction was successful, as cortisol levels were significantly higher in the stress group compared to the control group (p<0.001).
Stress led to a systematic shift in risk preferences, with more risk-seeking behavior under stress (p=0.03). Inspection of model parameters revealed that the shift in risk preference under stress was induced by more optimistic (and mostly more realistic) prior beliefs about the magnitude of the risky payoffs (p<0.01). By contrast, the noisiness of the inferred payoff representations was unaltered. Neural results aligned with this, as there was a large shift of neural coding only for the stressed group (two sample T-Test on group specific shifts: p=0.002) but no change in neural precision. The shift in neural coding correlated with the shift in risky prior (r=0.31,p=0.034) and each of them with the individual cortisol response.
Stress led to a systematic shift in risk preferences, with more risk-seeking behavior under stress (p=0.03). Inspection of model parameters revealed that the shift in risk preference under stress was induced by more optimistic (and mostly more realistic) prior beliefs about the magnitude of the risky payoffs (p<0.01). By contrast, the noisiness of the inferred payoff representations was unaltered. Neural results aligned with this, as there was a large shift of neural coding only for the stressed group (two sample T-Test on group specific shifts: p=0.002) but no change in neural precision. The shift in neural coding correlated with the shift in risky prior (r=0.31,p=0.034) and each of them with the individual cortisol response.
Conclusions:
In conclusion, our Bayesian perceptual approach together with nPRF modelling allowed us to provide a mechanistic perspective on the cognitive and neural processes driving the effect of stress on risky choice. Our results indicate that stress does not lead to noisier processing of information, but to shifts in prior beliefs and, correspondigly, the underlying neural representations of numerosities.
Higher Cognitive Functions:
Decision Making 1
Space, Time and Number Coding 2
Modeling and Analysis Methods:
Bayesian Modeling
Perception, Attention and Motor Behavior:
Perception: Visual
Keywords:
Computational Neuroscience
Other - Stress
1|2Indicates the priority used for review
Provide references using author date format
Buckert et al. (2014), 'Acute stress affects risk taking but not ambiguity aversion', Front Neurosci. 2014; 8: 82
Haushofer & Fehr (2014), 'On the psychology of poverty', Science, 2014 May 23;344(6186):862-7
Khaw et al., (2020), 'Cognitive imprecision and small-stakes risk aversion', The Review of Economic Studies, Volume 88, Issue 4, July 2021, Pages 1979–2013
Garcia et al. (2023), 'Individual risk attitudes arise from noise in neurocognitive magnitude representations', Nat Hum Behav. 2023 Sep;7(9):1551-1567