Computational Mechanisms of Social Influence During Adolescence

Abstract

Adolescents are known for their propensity to take risks. They binge-drink (Spear, 2018), use drugs (Defoe et al., 2019), drive risky (Romer et al., 2014) and get into trouble with the law (Viner et al., 2011) more often than people in other developmental stages (Steinberg, 2018). However, a recent meta-analysis on adolescent risk-taking behaviour in controlled laboratory settings showed no evidence of an adolescent risk-taking peak (Defoe et al., 2015). Past research identified two major reasons for the discrepancy between adolescents’ real-life and laboratory risk-taking. First, adolescents are especially attentive to social signals (Blakemore & Mills 2014), and most adolescent risk-taking behaviour in real-life occurs in some kind of social context. Second, adolescents seem to be less inhibited by uncertainty than people in other developmental stages (Tymula et al., 2012; van den Bos & Hertwig, 2017). The risks that adolescents take in real-life are subject to much greater uncertainties than those portrayed in many laboratory experiments. In my dissertation, I investigate how adolescents' sensitivity to social signals and their uncertainties contribute to their propensity to take risks. Chapter 1 gives a broader literature overview that ends in pointing out how understanding the intersection between social susceptibility and uncertainty is crucial to understanding adolescent risk-taking. Chapter 2 points out that the developmental processes underlying adolescents’ social sensitivity remain poorly understood, despite extensive research and theorisation. I emphasise that while many theories assume different mechanisms behind adolescent social susceptibility, they are all consistent with a broad range of evidence from laboratory experiments. I thus propose a formal framework depicting these theories in mathematical equations that make precise predictions that can be evaluated against one another and show that doing so synthesises seemingly disparate results. Chapter 3 introduces a novel experimental paradigm that allows manipulating uncertainties and proposes a model that understands social influence as a learning process wherein social information is more impactful when people are more uncertain. This chapter shows that developmental differences in peoples’ uncertainties can partially explain developmental differences in social impact during risk-taking. Chapter 4 shows how social norms contribute to real-life risk-taking in the general population and provide evidence that adolescents overestimate the normative character of many risky behaviours. Chapter 5 points out that people in different developmental stages live in different environments that expose them to different risks. Using agent-based modelling, I show that an adolescent-peak in risky behaviour can, in combination with a tendency for exploration, be emergent form transitioning from a relatively safe childhood into a risky adult environment. Chapter 6 provides a summary and conclusion of the preceding chapters and points to future research questions that my dissertation opens up.