PhD thesis: Knowing you, changing me: Individual differences in social learning and influence in discounting tasks
7 minute read
Our decisions are often made within social contexts, and our behaviour and beliefs are subject to change due to the influence of others. One type of social influence is contagion. Contagion is an implicit social influence effect whereby learning about others makes us more like them.
My PhD research focused on using a Bayesian approach to explore individual differences contagion in neurotypical and autistic adults. To explore this, I used discounting tasks (temporal and probability discounting) along with a mix of computational modelling, fMRI, behavioural tasks, and questionnaire measures. I ran both lab-based and online studies.
I use neuroeconomics tasks such as temporal discounting to explore social contagion effects. My behavioural and online research explores individual differences in contagion related to traits (measured by questionnaires), and learning in different behavioural tasks. My fMRI experiment examined the underlying neural basis of social contagion in a neuroeconomics task, with a particular focus on prediction error signalling in the anterior cingulate gyrus (ACCg).
My supervisor, Dr. Joshua Balsters has also written an awesome blog post talking about this research.
Chapter 1: Contagion of temporal discounting value preferences in neurotypical and autistic adults
Find the paper here, and my Twitter thread about the paper here.
This is a behavioural study of individual differences in contagion in two independent neurotypical samples (Study 1 and Study 2), and one sample of autistic adults (Study 3).
These studies examines learning of the value preferences of others, and contagion of temporal discounting (also known as delay discounting!) preferences using a task developed by Mona Garvert. If you want to find out more about temporal discounting and what it is, you can read Josh’s blog post.
We fit Bayesian computational models to precisely quantify choice behaviour on a temporal discounting task, where participants make temporal discounting decisions for themselves before and after learning the different preferences of one more patient agent, and one more impulsive agent. The choices of these agents were modelled, but we told participants that they were the choices of real previous participants.
In both of our neurotypical samples, participants became significantly more patient (i.e., were willing to wait longer to receive a reward) in their choices after making choices on behalf of a more patient other agent (i.e., contagion). We didn’t find that participants became significantly more impulsive!
There were also no significant differences in contagion between the autistic sample, and a matched subset of neurotypical participants, and Bayesian tests of equivalence also provided support for equivalent contagion and accuracy across these two samples.
The findings of this study indicate that, despite social differences being a core trait of autism, contagion of value preferences is one domain of social learning that appears to be intact.
Chapter 2: Does contagion generalise across different tasks? A comparison of temporal and probability discounting
The purpose of this experiment was to explore whether discounting (and contagion in discounting tasks) is domain specific, or domain general (and shares a common framework across tasks).
We explored contagion and discounting in a temporal discounting task and a probability discounting (also known as risk discounting!) task in a sample of 100 neurotypical participants.
We found significant contagion effects (i.e., participants became more like the other agents) for more patient and more risk-seeking others (but not more impulsive or more risk-aversive others).
We also found a correlation between participants own discounting preferences across the two tasks. Participants that were more patient in the temporal discounting task were also more risk-seeking in the probability discounting task. There was also a significant correlation between contagion effects across the two tasks. Together, these findings suggest a commonality of discounting and contagion across tasks. However, we also found that participants choices were more noisy overall in the probability discounting task.
Whilst there may be task-related relationships between participant’s discounting preferences, our results do highlight the importance of distinguishing between value assignment (i.e., discounting preferences) and choice behaviour (i.e., noise in choices) in assessing similarities and differences in discounting behaviour across tasks.
Chapter 3: The power of friendship: Does social distance modulate neural pathways controlling social contagion and learning?
Here, we explored whether the contagion effect was a truly social effect, or was purely the result of learning new information in an fMRI task.
In chapters 1 and 2, contagion was not affected by whether participants believed that they were making choices on behalf of a real agent or not, and previous research had also found no difference in contagion dependent on agent.
Here, we sought to determine whether using real-world examples of close friends and strangers affected contagion. If this is a social effect, we should find stronger contagion effects when participants make choices on behalf of close friends, verses strangers.
We recruited pairs of close friends or partners to take part, and put them in a group with a stranger (an experimental confederate!).
We found significant contagion effects for close AND distant others, suggesting that the effect of contagion wasn’t actually social. We also didn’t find activation related to contagion in any brain area.
We did, however, find a social distance effect on learning and feedback processing. Participants were more accurate at making choices on behalf of their close testing partner than the stranger, and there was also greater activation in the posterior medial prefrontal cortex and the postcentral gyrus related to participants receiving feedback for their choices on behalf of their close testing partner versus the stranger.
Although we didn’t find that contagion was a social effect, these results provide support for differences in learning about others depending on the identity of the other.
Chapter 4: Temporal discounting contagion and probabilistic social learning: Is contagion linked to advice-taking?
This study uses two online tasks exploring the relationship between contagion of temporal discounting preferences and choice behaviour in a probabilistic reward learning task with a social component. The task was set up on Gorilla to look like a card game table, with social information in the corner in a speech bubble.
In an online sample (N=195), we found a strongly significant contagion effect, and participant’s choices became more like those of the other agents.
We didn’t find a relationship between participant’s advice-taking (i.e., following the social information) and contagion in this task.
We’d like to have a look at the results of computational modelling analyses for this chapter too. Stay tuned to find out more!
If you would like to use an online temporal discounting task (also known as delay discounting!) or the probability task in your own research, and are interested in using this on Gorilla, feel free to email me and I can send you the task. My email address is Louisa.Thomas.firstname.lastname@example.org
Additional PhD work
I also started to explore effort discounting, and whether there was a difference in contagion for actual versus hypothetical effort.
I coded an experimental task in which participants exert effort by squeezing a grip force measure (with feedback presented in the form of a bar moving on the screen).
I intended to test this in the scanner before moving to an online study as a result of COVID-19.
I would love to continue working on aspects of effort discounting in the future!