The computational basis of following advice in adolescents
Introduction
Advice taking can foster learning and decision making, particularly if decisions are complex and the given information is incomplete (Biele, Rieskamp, Krugel, & Heekeren, 2011). For instance, when you arrive in a new city and must pick a restaurant for dinner, to make a more informed decision, you ask others who have previously visited this city for advice. After following the initial advice, you might also start gathering firsthand experience. For instance, after visiting the recommended restaurant a few times, you might learn that you do not like it that much and start exploring other options.
Taking advice is a common form of social learning1 and influences one’s own subsequent learning experiences (Biele et al., 2011, Biele et al., 2009, Goodyear et al., 2016). However, the degree to which social information and one’s own experience are used in learning varies significantly across development (Rodriguez, Heekeren, Li, & Eppinger, 2018). Adolescents in particular show a specific susceptibility to social influence, particularly to that of their own peers (Albert et al., 2013, Rodman et al., 2017, Sebastian et al., 2010). In this study, we aimed to better understand how this adolescent-specific susceptibility to peer influence shapes adolescents’ subsequent learning behavior and decision making.
Recent experimental studies have started to outline developmental differences in how susceptibility to social influence shapes behavior (Lourenco et al., 2015, Monahan et al., 2009, Steinberg and Monahan, 2007, Steinberg, 2008, Sumter et al., 2009). Consistent with conventional wisdom, these studies show that social information becomes increasingly influential on behavior during adolescence, particularly when it comes from peers (Albert et al., 2013, Blakemore and Mills, 2014, Jones et al., 2014). More specifically, the tendency to rely on social information is particularly prominent during early adolescence (lasting until approximately 15 years of age; see also Blakemore and Mills, 2014, Gardner and Steinberg, 2005, Gunther Moor et al., 2010, Sebastian et al., 2010). Many of these experimental studies compared behavior in the presence or absence of a passively observing peer (Cascio et al., 2015, Chein et al., 2011, Gardner and Steinberg, 2005, Powers et al., 2018, Somerville et al., 2018). However, adolescents often make important decisions in the absence of peers, which raises an important question—how, and for how long, does social information influence learning and decision making when peers are no longer present?
To date, little is known about how or for how long social information from peers affects adolescents’ learning and decision making. To our knowledge, only one study partly addressed this question. In this study, Decker et al. showed that in a learning task adolescents relied less on intentionally false instructions compared with adults (Decker, Lourenco, Doll, & Hartley, 2015). Instead, adolescents relied more on their own experience than on the feedback presented during the learning episode of the task. The authors suggested that instruction biases learning through the top-down influence of the prefrontal cortex and that, due to decreased striatal–prefrontal connectivity (Imperati et al., 2011, Liston et al., 2006), adolescents are less influenced by instructions.2 However, developmental differences during instruction-based learning could also have led to behavioral differences in this study. For instance, developmental differences in explorative behavior during learning can contribute to a reduced instruction bias in children and adolescents. Previous research has shown that children and adolescents explore more options than adults (Christakou et al., 2013). Thus, adolescents might have learned earlier that the instructions were false and that better options were available, which raises another important issue. If we want to understand how social information affects adolescents in the absence of peers, we must understand how it interacts with their own experience, particularly given the well-established findings on developmental differences in learning from experience (Crone et al., 2004, Crone et al., 2006, Eppinger et al., 2009, Ferdinand and Kray, 2014, Hämmerer et al., 2010, van den Bos et al., 2012, van Duijvenvoorde et al., 2008). One consistent result from these studies is that performance increases across adolescence. In addition, these studies show that children have greater difficulty in using negative feedback for learning (Crone et al., 2004, Eppinger et al., 2009, Hämmerer et al., 2010, van den Bos et al., 2012, van Duijvenvoorde et al., 2008). Thus, to understand the unique and lasting effect of social information on adolescent behavior, we also must consider developmental differences in experience-based learning.
To resolve these outstanding issues, we used a specifically designed social reinforcement learning (RL) task in combination with computational modeling. The task involves both social information- and experience-based learning (modified Iowa Gambling Task after Biele et al., 2011). Computational models have the advantage that they can be used to test different theories about the latent processes that underlie social influence and learning (van den Bos, Bruckner, Nassar, Mata, & Eppinger, 2018).
In the current task, participants (children, adolescents and adults) needed to choose one of four card decks that were associated with gains and losses (see Fig. 1A). Unbeknownst to the participants, two of the four decks were associated with higher expected positive values (“good decks”) than the other two (“bad decks”) (see Fig. 1B). At the beginning of the experiment, participants received good advice (i.e., for one of the good decks) from a same-aged peer. After the advice was given, the participants were free to draw from each of the decks as often as they liked. Both good decks were equally good (equal payoff distributions); thus, if the participants explored the other decks, they could over time learn that there was another equally good deck. Furthermore, if the initial advice was completely ignored, participants were expected to draw equally often from each of the good decks. However, Biele et al. (2011) showed that, using the exact same task, adults keep preferring the advised good deck. Thus, the task was designed such that the extent to which the other good deck is selected captures the influence of the advice. In that sense, the other good (nonrecommended) deck served as a “nonsocial” condition. Finally, we used computational modeling to separate the effects of advice, experience (positive and negative), and exploration on learning behavior (for more details on models, see Method). As noted above, each of these aspects could contribute to developmental differences in following advice.
Based on previous studies, we expected the susceptibility to peer influence to peak in adolescents (Steinberg & Monahan, 2007) compared with children and adults. Furthermore, we expected that influence would be particularly strong at the beginning of the task before learning by experience took over (Decker et al., 2015). In addition, we expected that throughout learning, children and adolescents would show more exploration than adults (Christakou et al., 2013, Decker et al., 2015). Thus, the two younger groups should be faster in discovering that there is another (equally) good deck in the experiment. Finally, children were expected to show less optimal performance overall due to their difficulty in using negative feedback for learning (van Duijvenvoorde et al., 2008). We have developed several computational RL models to capture the interaction among these different elements (social influence, exploration, and experience; see Method for detailed model predictions).
Section snippets
Participants
The effective sample of the study consisted of 25 adults aged 18–22 years (13 female; mean age = 20.32 years, SD = 1.15), 24 adolescents aged 13–15 years (12 female; mean age = 13.71 years, SD = 0.75), and 24 children aged 8–10 years (10 female; mean age = 9.08 years, SD = 0.83) (see online supplementary material for a justification of age group selection). Our sample size estimation was based on a study by van den Bos et al. (2012) that used a feedback-based learning task, similar RL models to
Choice behavior
In a first step, we checked to what extent each deck was selected above chance level (i.e., 25%) averaged across all trials (i.e., 210) using one-tailed t tests. As seen in Fig. 2A, the adults and adolescents selected the recommended deck above chance (ts > 2.60, ps ≤ .01), but the children did not, t(23) = 1.53, p = .14. The adolescents chose bad decks below chance (ts > −2.73, ps ≤ .01), whereas the adults and children did not (ts > −2.00, ps ≥ .006) (see Fig. 2A). As expected, most
Discussion
In this developmental study, we used a four-armed bandit task that included peer advice in children (8–10 years old), adolescents (13–15 years old), and young adults (18–22 years old). As expected, all age groups followed the advice at the beginning of the task; however, after a few trials, the behavior of the different age groups started to diverge. The most salient developmental differences suggest that (a) adolescents are initially the most sensitive to advice, (b) adults most consistently
Conclusion
Taken together, our findings show that peer advice guides learning from one’s own experience and that adolescents show the highest initial susceptibility to peer advice. Crucially, higher exploration rates enable adolescents to discover other opportunities. Thus, our results extend previous findings by showing that adolescents’ more exploratory behavior could be—depending on the environmental structure—even more beneficial than less exploratory learning strategies. Taken together, these
Acknowledgments
We would like to thank Valerie Keller, Raphael Schultz and Jann Wäscher for help during data acquisition, Lukas Nagel for programming the task and our participants for their contribution to the study. This work was supported by the Freie Universität Berlin, the Jacobs Foundation, and an Open Research Area grant (ID 176), European Research Council grant (ERC-2018-StG-803338) and the Netherlands Organization for Scientific Research grant (NWO-VIDI 016.Vidi.185.068) awarded to W.VD.B. Julia
References (59)
- et al.
Advice taking and decision-making: An integrative literature review, and implications for the organizational sciences
Organizational Behavior and Human Decision Processes
(2006) - et al.
A heart rate analysis of developmental change in feedback processing and rule shifting from childhood to early adulthood
Journal of Experimental Child Psychology
(2006) - et al.
Developmental changes in performance monitoring: How electrophysiological data can enhance our understanding of error and feedback processing in childhood and adolescence
Behavioural Brain Research
(2014) - et al.
Pubertal development and behavior: Hormonal activation of social and motivational tendencies
Brain and Cognition
(2010) PsychoPy—Psychophysics software in Python
Journal of Neuroscience Methods
(2007)- et al.
Cognitive components underpinning the development of model-based learning
Developmental Cognitive Neuroscience
(2017) - et al.
Social brain development and the affective consequences of ostracism in adolescence
Brain and Cognition
(2010) The adolescent brain and age-related behavioral manifestations
Neuroscience & Biobehavioral Reviews
(2000)A social neuroscience perspective on adolescent risk-taking
Developmental Review
(2008)- et al.
The developmental pattern of resistance to peer influence in adolescence: Will the teenager ever be able to resist?
Journal of Adolescence
(2009)
Computational neuroscience across the lifespan: Promises and pitfalls
Developmental Cognitive Neuroscience
The teenage brain: Peer influences on adolescent decision making
Current Directions in Psychological Science
Fitting linear mixed-effects models using lme4
Journal of Statistical Software
The computation of social behavior
Science
Computational models for the combination of advice and individual learning
Cognitive Science
The neural basis of following advice
PLoS Biology
The role of puberty in the developing adolescent brain
Human Brain Mapping
Is adolescence a sensitive period for sociocultural processing?
Annual Review of Psychology
Buffering social influence: Neural correlates of response inhibition predict driving safety in the presence of a peer
Journal of Cognitive Neuroscience
Peers increase adolescent risk taking by enhancing activity in the brain’s reward circuitry
Developmental Science
Neural and psychological maturation of decision-making in adolescence and young adulthood
Journal of Cognitive Neuroscience
Understanding popularity in the peer system
Current Directions in Psychological Science
Developmental change in feedback processing as reflected by phasic heart rate changes
Developmental Psychology
Experiential reward learning outweighs instruction prior to adulthood
Cognitive, Affective, & Behavioral Neuroscience
From creatures of habit to goal-directed learners: Tracking the developmental emergence of model-based reinforcement learning
Psychological Science
Developmental differences in learning and error processing: Evidence from ERPs
Psychophysiology
Computational modeling of cognition and behavior
Peer influence on risk taking, risk preference, and risky decision making in adolescence and adulthood: An experimental study
Developmental Psychology
Auditory working memory and Wisconsin Card Sorting Test performance in schizophrenia
Archives of General Psychiatry
Cited by (23)
Environmental statistics and experience shape risk-taking across adolescence
2023, Trends in Cognitive SciencesUnderstanding patch foraging strategies across development
2023, Trends in Cognitive SciencesSensitivity and specificity in affective and social learning in adolescence
2023, Trends in Cognitive SciencesDiminished reinforcement sensitivity in adolescence is associated with enhanced response switching and reduced coding of choice probability in the medial frontal pole
2023, Developmental Cognitive NeuroscienceSocial learning across adolescence: A Bayesian neurocognitive perspective
2022, Developmental Cognitive NeuroscienceEffects of advice on experienced-based learning in adolescents and adults
2021, Journal of Experimental Child PsychologyCitation Excerpt :This last finding corroborates the finding described in the previous section that advice suppressed the learning rate during early trials, especially when working memory load was low (note that our models assumed constant update rates and hence could not specifically capture advice effects during early trials). Previous studies have investigated effects of advice on experience-based learning using instrumental learning tasks, which involve repeated choices between two or more stimuli (Biele et al., 2011; Decker et al., 2015; Doll et al., 2009; Lourenco et al., 2015; Rodriguez Buritica et al., 2019). In these tasks, both exploratory choice behavior and advice-related modulation of the learning process can influence advice following, which makes developmental differences in these two processes difficult to disentangle.