Introduction
During adolescence, the social environment undergoes dramatic changes. As adolescents increase the time spent with peer, they also increase the engagement in risk behaviors. Even though such observations have commonly been interpreted as adolescents being specifically susceptible to peer influences (Albert et al.,
2013), this study set out to directly model the extent to which adolescents are influenced by peer presence during risky decision-making. Most investigations have not employed pure measures of risk-taking tendencies, as many risky decision-making tasks are about about how to deal with uncertainty about risks and possible positive and negative outcomes (Do et al.,
2020b). In fact, many risk-taking behaviors, whether in the laboratory or in real-life, are not only about risk-taking tendencies, but how information is utilized to guide decision-making (Silva et al.,
2016). Therefore, the purpose of this study is to disentangle which cognitive processes are influenced by the presence of a virtual peer when dynamically changing risk probabilities and benefits can only be experienced.
In the decision-making literature, risk-taking is commonly defined as the tendency to choose options with the greatest variance in outcomes. For example, adolescents are considered risk-prone if they tend to choose options that imply multiple possible outcomes over options with a certain outcome (Figner & Weber,
2011). Furthermore, experimental investigations of risky decision-making allow researchers to directly manipulate the social context of risk-taking behavior. Self-described peer resistance has shown to be low in adolescence and to increase with age (Steinberg & Monahan,
2007). To experimentally quantify and to test developmental differences in peer presence effects, studies introduced either direct peer interaction, passive presence, or observation during risky decision-making. When comparing the effect of peer advice and observation, one study showed increases in risk-taking tendencies in both, adolescents and adults, when a peer encouraged risk-taking in a gambling task. Peer observation, however, only increased adolescents’ risk-taking tendencies, highlighting the importance to distinguish between different peer presence effects (Haddad et al.,
2014). Passive peer observation, such as the observation by a friend (Somerville et al.,
2019), or unfamiliar and only virtual peer (Haddad et al.,
2014, Smith et al.,
2014), has been shown to modify adolescent risk-taking tendencies. However, when a peer did not explicitly observe behavior, mere peer presence was sometimes not sufficient to affect risk-taking tendencies in adolescence (Somerville et al.,
2019). As such, recent findings about the benefit of peer influence on risk taking have been inconclusive depending on the type of risk and social context of the tasks at hand.
One reason for the disparate findings in the literature may be the fact that social sensitivity might not apply to all adolescents but might be an individual disposition across development (Do et al.,
2020b). For instance, individual differences in peer resistance explained variance in findings about peer presence increasing (e.g., Chein et al.,
2011) or decreasing the number of risky choices (Kessler et al.,
2017) depending on the task context used. Some reviews in recent years have pointed out that adolescents’ tendency to choose risky options is overly sensitive to diverse aspects of the task context (Defoe et al.,
2019; Romer et al.,
2017; Shulman et al.,
2016). Apart from the social context, the sensitivity to risk probabilities has been shown to guide risk-taking tendencies during decision-making (Defoe et al.,
2019). In classical experimental settings, participants may be able to deduce specific task outcomes, leading to unrealistic measures of risk-taking tendencies. To better approximate real-life decision-making, researchers increased uncertainty by obscuring outcome probabilities. Furthermore, researchers can create more dynamic version of traditional decision making tasks by manipulating the trial-by-trial probabilities of positive or negative outcomes. Using similar dynamic approaches, recent findings indicate that adolescents readily increase risky choices as opposed to situations under known risks (Defoe et al.,
2019; Lorenz & Kray,
2019), also known as ambiguity tolerance. Such as heightened ambiguity tolerance in youth, peer presence effects interact with adolescent decision-making only in situations where the probabilities of outcomes are ambiguous instead of explicit (Lloyd & Döring,
2019; but see Smith et al.,
2014). However, peer presence effects do not fully explain adolescent decision making, leading to an increase of risky choices in experience based tasks (e.g., Chein et al.,
2011), decrease of risky choices in tasks with dynamically changing risk probabilities (Kessler et al.,
2017) or having no effect on risky choice altogether (Reynolds et al.,
2014).
Adding upon the sensitivity to the task context in adolescence, previous findings have highlighted that younger and older adolescents have divergent risk-taking tendencies. Consequently, inconclusive findings could also derive from the diversity in the age ranges and groups used to test peer presence effects. According to the suggested quadratic age trends in risk-taking (Shulman et al.,
2016), adolescents showed greater tendencies for risky choices than adults in ambiguous risk situations. In contrast, adolescents showed similar risk-taking tendencies as children (for a review, see Defoe et al.,
2019). Simultaneously, the boundaries of adolescence are shifting, a development that can be observed in science but also in legal and health policies. On the one hand, there is an ever earlier beginning of puberty and, on the other hand, neurodevelopmental findings push the upper threshold of the adolescent phase into the early 20s (see Ledford,
2018). To test developmental differences in peer presence effects, most studies have assessed peer presence effects form one narrow adolescent age group and have compared adolescent groups with adult groups at utmost. Yet, a few studies have provided further evidence that there is a high variability in peer presence effects between decision contexts even within the adolescent period. Effects of peer presence (aged 13–25 years, Somerville et al.,
2019) and advice (aged 12–22 years, Braams et al.,
2019) increased or decreased the number of risky choices dependent on the risk uncertainty in a given task context and the developmental stage throughout early to late adolescence.
One additional reason for inconclusive findings on peer presence effects might be the traditional calculation of dependent measures. Most studies rely on the mean number of risky choices that have been tightly associated with risk-taking tendencies. Even the names of many decision-making tasks refer to the term risk, like the Balloon Analog Risk Task (BART, Lejuez et al.,
2002). Thereby, risk-taking is traditionally calculated by the mean number of risky choices for trials that resulted in positive outcomes (adjusted mean number of risky choices) to compensate for censoring, as trials end early in case of negative outcomes in the BART. But in contrast to decision-making under known risk where trials are independent probabilistic events, peers might influence how adolescents use information to update subsequent decisions, especially in uncertain and dynamically changing risk environments. Taking the inherent variability of the sample, developmental differences in behavioral adjustment to risk uncertainty throughout pre- to late adolescence were considered in this study. In order to model a realistic environment, a sequential decision-making task under risk uncertainty was used, the BART (Lejuez et al.,
2002). The BART is a dynamic version of experience-based decision-making where participants are instructed to inflate virtual balloons. Each pump signifies a simultaneous increase in potential monetary outcomes, as well as a risk of bursting and the loss of all previous earnings.
Reinforcement learning analyses are particularly well suited to modeling such sequential decision-making (for the BART, see Wallsten et al.,
2005) as they allow to distinguish between processes that have previously been discussed to contribute to behavioral adjustment, like exploration and learning. To model choice behavior, reinforcement learning models assume that participants estimate the value of each choice option and update these estimates according to experiences made. In the example of the BART, a parameter is calculated that indicates the participant’s belief in success when pumping a balloon, i.e., the individual belief that pumping a balloon may lead to monetary gains. Based on the belief in success, an updating rate is calculated that quantifies how fast participants adjust their beliefs in success to actual choice outcomes. Risk-taking tendency in this example is the tendency to pump above the perceived value of pumping or saving previous earnings. Moreover, the so-called inverse temperature estimate determines the extent to which the different values of choice options guide choice behavior. High inverse temperature values mean that the difference in outcomes of choice options is exaggerated and individuals stay with a specific choice pattern. In contrast, a low inverse temperature score can reflect exploration tendencies, i.e., the tendency to try out different options and to gather information when in uncertainty (Nussenbaum & Hartley,
2019). In sum, formal models allow for more specific hypotheses on behavioral adjustment that could help to uncover the effect peers have on risk-taking but also on adaptive behavior. In this sense, reinforcement learning analyses have successfully been applied to identify subgroups with addictive tendencies (Wallsten et al.,
2005).
More specifically, constructs assessed through the parameters of formal models can be compared between different choice architectures, groups, or states (Nussenbaum & Hartley,
2019), like peer presence. One potential hypothesis would be that peer presence would provoke impulsive behavior and associated risk-taking by increasing the perceived potential for rewards during adolescence. This is the case because adolescents have been assumed to be specifically reward sensitive due to only gradual increases in cognitive control abilities but a rise in susceptibility to rewarding cues (Shulman et al.,
2016). Unlike adults, adolescents showed greater activity in reward-related brain systems while engaging in a higher number of risky choices in a study using simulated driving under peer observation (Chein et al.,
2011). Despite evidence on the neural level, findings on the behavioral level have shown evidence against the suggestion that adolescents are reward-sensitive in all situations (for a review, see Kray et al.,
2018). When a peer observed choice behavior in the BART, adolescents were even more cautious and reduced the number of risky choices in trials after positive outcomes (Kessler et al.,
2017). An alternative hypothesis could be that a high number of risky choices rather reflects exploration tendencies and learning from experiences instead of rash and impulsive choice behavior (Romer et al.,
2017). Late adolescents (aged 18–21 years) explored choice options and used both, positive and negative feedback, to adjust behavior towards long-term goals when a peer observed choice behavior in a gambling task. When risks and outcomes could only be experienced, adolescents increased adaptive decision-making through exploration behavior and learning from its outcomes (Silva et al.,
2016). Thereby, decreasing exploration tendencies with age across the lifespan is one of the most robust findings in developmental studies applying reinforcement learning models (Nussenbaum & Hartley,
2019).
Finally, an additional factor that has been shown to interact with exploration tendencies and peer susceptibility is gender. Female adolescents have been shown to be less sensation-seeking than male adolescents, and thus, less inclined to show exploratory behavior and associated risk-taking (Cross et al.,
2011). Similarly, male adolescents have engaged in a higher number of risky choices than female adolescents in several decision-making tasks (de Boer et al.,
2017; Cazzell et al.,
2012; Lejuez et al.,
2002; but not Lejuez et al.,
2003) and self-described risk propensity measures (Byrnes et al.,
1999). Given that peer presence can be a particularly arousing situation, male adolescents might also engage in more risk-taking than female adolescents in social situations. Some studies indeed have found risk-heightening effects of peer presence only for male adolescents (Defoe et al.,
2020), or have found male adolescents to be at least more influenced by peer advice than female adolescents (Boer et al.,
2017). However, other studies found no gender differences in peer presence effects on experimental risk-taking at all (Boer & Harakeh,
2017; Harakeh & Boer,
2018). In this sense, a qualitative review on gender differences in adolescent susceptibility to deviant peer pressure suggested male adolescents to be more influenced by peers than female adolescents in only 46% of all studies investigated (see McCoy et al.,
2019). Together with exploration tendencies, developmental, and individual differences in social susceptibility, gender identity seems to add to the number of factors that exert a non-linear, dynamic influence on adolescent risk-taking in peer presence.