The Impact of Social Learners on Collective Decision-Making

Post by Lani Cupo

What's the science?

In democratic societies, collective decisions, such as who should hold power, or what action should be taken to address climate change, can drastically impact society. But when people make decisions in groups, the most popular option is sometimes chosen even though it does not have the most merit. This phenomenon is due in part to the presence of those identified here as social learners who adopt the opinions of others instead of critically assessing options for themselves. This week in PNAS, Yang and colleagues developed a mathematical framework for investigating whether there is a critical threshold of social learners that can be present in a collective decision, after which one option may prevail because of popularity, rather than merit. 

How did they do it?

The authors created a dynamical system model which integrated two options (X and Y) with relative merit (m) associated with each option, where m was a number between zero and one. The model incorporated differing proportions of social and independent learners (s) in the population. Finally, it included one parameter as a function that is hypothesized to model two types of conformity, normative (engaging in a behavior because others do it), and informational (engaging in a behavior because it is the right thing to do). The authors derived transition rates between the options for the different types of learners, where social learners will transition between the options based on the popularity of the option, but independent learners transition based on the merit of the option. This allowed the authors to examine the fixed points of the equation, where the proportion of people favoring a given option stops changing. They also investigated how stable these fixed points are when the model is perturbed. Their conclusions remained the same when they changed the model to account for opinion on a spectrum from independent to social, when they only allowed individuals to be impacted only by their local environment, and when they introduced statistical noise to the model. Finally, the researchers simulated a model incorporating the strength of opinion weighting towards option X or Y. 

What did they find?

When X and Y are options with equal merit, there is a critical threshold for the proportion of social learners after which the model bifurcates into two branches, implying either option X or Y could be selected. In the case that the model is not equal between groups, the majority will favor the more meritorious option up to a critical point, but if the proportion of social learners is too high, instability is introduced into the model, meaning there can be cases in collective decision making where the less meritorious option is still chosen. The critical threshold is determined both by the discrepancy in merit between the two options and the conformity function. The threshold that these parameters identify predicts the threshold above which the proportion of social learners harms the decision. Notably, the model is flexible to adapt to different behaviors modelled through the conformity function or to allow parameters, such as the strength of opinions, meaning it represents a flexible tool that can be used to model different situations. 

What's the impact?

This study investigated the impact of social learners on collective decision-making, demonstrating there is a threshold above which social learners may negatively impact the outcome of collective decisions. The outcome of collective decisions can drastically impact daily life—not only in small communities but on a national and global scale as well. The mathematical framework presented provides future studies with the ability to examine social learning in varied and complex scenarios. 

Yan et al. Dynamical system model predicts when social learners impair collective performance. PNAS (2021). Access the original scientific publication here

Decision Processes Leading to Unhealthy Food Choices

Post by Andrew Vo

What's the science?

After making poor dietary choices, we often blame our actions on either a strong preference for tasty (but oftentimes unhealthy) food, or on poor self-control. Traditional computational models characterize such value-based decisions as a dynamic accumulation of evidence that biases us towards one option over another. These models, however, do not account for distinct contributions of separable attributes to a decision (e.g., how health and taste attributes are integrated with different weights and at different times in evidence accumulation). This week in Nature Human Behaviour, Sullivan and Huettel use an updated computational framework to better understand how distinct attributes influence decision processes that could lead to unhealthy food choices.

How did they do it?

The authors recruited a group of young adults who arrived hungry at the lab after a four-hour fast. They were then asked to rate 30 different snack foods based on tastiness, healthiness, and ‘wanting’ attributes. Before beginning the main task, participants received a behavioral primer that emphasized the importance of either healthy or tasty choices. During a main, binary choice task, they were presented with pairs of food items (that they had previously rated) and were asked to indicate which they would like to eat more. Of the 300 self-paced trials, half were designed to be “conflict trials” in which one option was tastier but less healthy than the other, whereas the other half were non-conflict trials in which both options were closely matched.

Participants’ food choices and response times (RTs) were fitted using a multi-attribute, time-dependent, drift diffusion model (mtDDM) (a statistical model). This model has the advantage of distinguishing the various contributions of different attributes to a decision. To do this, it estimates (1) drift slope, which captures the rate of evidence accumulation for each attribute, and (2) drift latency, which describes when each attribute begins to exert its influence during evidence accumulation.

What did they find?

The authors found faster RTs for conflict versus non-conflict trials, as participants made fast unhealthy choices over healthier ones. Those participants who were primed with health information were found to put less weight on taste information, which marginally increased their likelihood of making healthy choices.

The mtDDM estimated that taste drift slopes were larger (steeper) than health drift slopes and taste drift latencies were earlier than health drift latencies. These results suggest that bias towards tasty versus healthy food choices is due to a greater weighting and earlier entry of taste information into evidence accumulation. To test whether slope and latency independently influenced food choices, multiple linear regressions of drift slope and latency differences (i.e., taste minus health) were performed. Both drift slopes and latencies predicted individual differences in the likelihood of healthy food choices. Finally, examining the relationship between trial-by-trial RTs and healthy choices in conflict trials, the authors found that longer RTs were associated with healthier food choices. This suggests that longer RTs allow time for slower-processed healthy information to influence evidence accumulation.

What's the impact?

This study demonstrates how the influence of different attributes on decision-making processes might explain our food choices. The results provide insight into how we can augment our thinking to make better decisions for our long-term benefit, such as considering the healthiness alongside the tastiness of a food item or taking more time to seek out health information on a food choice. Understanding the timing of decision processes in the brain might also be key to creating effective interventions that help people make better choices — not just in terms of diet but also in financial decisions, for example. Much like how you should look before you leap, consider pausing before you place that next restaurant order.

Sullivan & Huettel. Healthful choices depend on the latency and rate of information accumulation. Nature Human Behaviour (2021). Access the original scientific publication here.

We Arrive at Negative Conclusions More Easily Under Threat

Post by D. Chloe Chung

What's the science?

We accumulate information over time to make important decisions, sometimes even in highly stressful situations. As there is often endless information available to us, we need to decide when to stop gathering information in order to make judgments. However, under stressful, threatening conditions, we are prone to make decisions even with a small amount of information (e.g. hearing a faint sound in a dark alleyway, perceiving it as a threat, and concluding that the environment is dangerous). This week in the Journal of Neuroscience, Globig and colleagues investigated how we process information in threatening situations.

How did they do it?

A total of 83 participants were divided into a threat manipulation group and a control group. The threat manipulation group was informed that they would later have to deliver a speech on an undisclosed topic in front of judges. Then, they were asked to solve difficult math problems in a limited time. These anticipated threats were designed to increase the anxiety and stress levels in participants. On the contrary, the control group was told that they would later have to write a short essay on a random topic that would not be judged and were given easier math problems to solve. After the manipulation, both groups played the “Factory Game”: Participants had to determine whether they were in a telephone or television factory based on the number of telephones or televisions shown on a moving conveyor belt that they were observing. During this test, for each participant, one type of factory was randomly assigned as the “desirable” factory and the other one as the “undesirable” factory. Participants were told that they would earn points when they visit the desirable factory but lose points when they visit the undesirable factory. They were also told that they would earn points when making a correct judgment about the factory type but would lose points upon a wrong judgement. The two payments were independent of each other.

What did they find?

After first checking that the level of anxiety was successfully increased in the “threat manipulation group” participants, the authors observed that the threat manipulation group tended to determine that they were in the undesirable factory (for example, the television factory) after observing a smaller proportion of undesirable items (televisions) compared to the control group. This means that with perceived threats, participants were more likely to draw conclusions based on less evidence, but only when drawing conclusions about the undesirable condition specifically. When it came to the desirable factory, the presence of perceived threats did not change the amount of evidence required to draw conclusions about the desirability of the situation. To understand how threat affects evidence accumulation specifically the authors used a computational modelling approach, finding a higher relative rate for negative evidence in the threat manipulation group. They found that threat biases the way in which participants weigh valenced (positive or negative) information.

What’s the impact?

This work confirms that the way we collect and process evidence can be greatly impacted by threats present in our situation, suggesting that we draw conclusions about our undesirable situation quickly, and with little evidence. Findings from this study suggest that, for those who are more sensitive to threats due to anxiety or other mood disorders, this process of accumulating negative information faster could be harmful as it may lead to an overly negative assessment of their situation.

Globig et al. Under threat weaker evidence is required to reach undesirable conclusions. Journal of Neuroscience (2021).Access the original scientific publication here.