Resistance to Peer Influence Scores Predict Social but not Non-social Risk
Behaviour in Adolescence.
Maaike van der Rhee
Student Number 12644781
University of Amsterdam, Vrije Universiteit Amsterdam
Research Master Brain & Cognitive Science
Research Project 1, 12th June 2020
Supervisor: Barbara Braams (VU)
Examiner: Wouter van den Bos (UvA)
Abstract
Influence of peers has been linked to an increase in risk taking in adolescents. Determining when and to what extent peer influence affects risk taking could contribute greatly to preventing unnecessary risks in adolescents. It is expected that a greater susceptibility to influences of peers is related to more risk behaviour. Yet, to date, a standardised index of sensitivity to peer influence (Resistance to Peer Influence Questionnaire) has produced inconsistent results. These inconsistencies have been attributed to a mismatch in context of risk measures and the Resistance to Peer Influence Questionnaire. To substantiate this notion, this paper examines the effect of Resistance to Peer Influence on risk taking in different contexts in Dutch adolescents. Risk taking was operationalized using a new questionnaire and the Holt & Laury gambling task. Regression analysis showed a negative linear relationship between the Resistance to Peer Influence and social risk taking as measured by cheating, lying, and standing up for someone, but not between the Resistance to Peer Influence and biking risks or the Holt & Laury task. The results indicate that a measure of general social risk tendencies (Resistance to Peer Influence Questionnaire) is a good predictor for a measure that is also socially embedded, but not for a measure that does not involve such an environment. Thus, this paper highlights that the context of the measured risk behaviour should be considered for future research into adolescent risk taking and peer influence.
Introduction
As a schoolteacher and child psychologist once expressed: “Adolescence can be a time of turmoil and turbulence, of stress and storm. Rebellion against authority and against convention is to be expected and tolerated for the sake of learning and growth.” (Ginott, 1971, p. 24.). Like the above quote illustrates, adolescence is a time of rapid changes and development. Adolescence starts with the biological changes of puberty, when the growth rate accelerates and secondary sex characteristics, such as pubic hair and maturation of genitalia, develop (Spear, 2002). Cognitive functions progress as well, for example, processing speed, response inhibition and working memory all increase during adolescence (Luna, Garver, Urban, Lazar & Sweeney, 2004). Similarly, planning skills improve during this time period, as measured by the performance on the Tower of London (TOL) Test (Luciana, Collins, Olson & Schissel, 2009). The TOL test evaluates planning skills, as it requires individuals to manipulate objects to conform to visible spatial solutions. Additionally, the higher order cognitive control of thought, action, and emotion (which is called executive function or EF) evolves in adolescence (Prencipe et al., 2011). Tasks that investigate EF can generally be divided into two categories, “hot” and “cool” tasks (Zelazo & Carlson, 2012). Hot tasks are said to be more motivationally or emotionally salient, while cool tasks are more abstract and tend to take place in a more affectively neutral context. It has been suggested that improvements of cool EF (as measured, for instance, by the Stroop Test and Digit Span) occur earlier than improvements of hot EF (as measured, for instance, by the Iowa Gambling Task and Delay Discounting; Prencipe et al., 2011). In addition to these physiological and cognitive changes, adolescents spend increasingly more time with their peers, and peer relationships become more intense and complex (Brown, 2004).
Coinciding the aforementioned changes, it is observed that, compared to childhood and adulthood, risk taking is heightened in adolescence. It has been argued that this risk taking can be adaptative for survival and sexual selection (Steinberg, 2008), as well as allow adolescents to learn more boundaries and become more independent (Coleman, & Hagell, 2007). However, risk taking can also lead to severe physical or mental problems in adolescence. Indeed, from childhood to late adolescence, morbidity and mortality rates increase 200%, though adolescents are physically fit (Dahl, 2004). This is canonically referred to as “the health paradox of adolescence”. The causes behind the inflated morbidity and mortality rates in adolescence are primarily behaviourally generated, meaning they are predominantly caused by vehicle use, sexual behaviour and substance abuse (Irwin & Millstein, 2014). The primary causes of death and disability among adolescents are preventable forms of injuries, such as accidents, violence or suicide (Dahl, 2004). This evokes quite a gruesome picture, however it has been noted that
these increases in rates aren’t as dire as implied, as i.e. less than 1 % of adolescents in the USA and Canada die (Willoughby, Good, Adachi, Hamza & Tavernier, 2014). Nevertheless, these increases are significant and should be taken seriously, as every death or disability at this particular age is tragic. These increased rates also direct us towards posing threats for this age group. In addition, many forms of risk behaviour that are initiated in adolescence elevate the risk for said behaviour in adulthood (for instance, drinking), and some forms of risk-taking by adolescents put individuals of other ages at risk (for instance, reckless driving; Steinberg, 2004). A reduction in adolescent risk taking could therefore lead to a meaningful improvement in the overall well-being of the population.
In the previous section, it was discussed that adolescents distinctively take more risks than their younger and older peers. Researchers have set out to investigate whether this could be due to underdeveloped cognitive abilities that assess risk. Perhaps unexpectedly, this is not the case. Adolescents’ cognitive abilities in risk assessment appear to be on a similar level as that of adults: for example, adolescents do not underestimate the chance of contracting a STI when having unsafe sex or getting lung cancer when smoking (Furby & Beyth-Marom, 1992; Reyna & Farley, 2006; Steinberg, 2008). They also do not perceive themselves as more invulnerable than adults do, and they have a similar degree of optimistic bias. Furthermore, compared with adults, adolescents sometimes overestimate rather than underestimate risk (Millstein & Halpern-Felsher, 2002).
Rather than exploring the risk-related cognitive capacity of adolescents, other scientists have focussed on the context in which adolescents take risks. The context of risky behaviour refers to the situational factors around the behaviour. Research into the context tries to investigate whether there are, for instance, differences in risk taking when adolescents are under pressure, online or offline, alone or with others. Among others, this research has proposed peer influences to be important in adolescent risk taking. Peer influence is defined as social processes with peers that lead to changes in a attitudes and behaviour (Arnett, 2007). Adolescents can learn social behaviour from their peers, according to the social learning theory (Bandura, 1986). Common forms of peer influence, such as display and reinforcement, is a way in which adolescents can acquire social norms of the peer groups (Brown, Bakken, Ameringer & Mahon, 2008). Besides the indirect influence (by their perception of group norms, social acceptance, and status associated with the behaviour) adolescents can also be influenced by active persuasion to engage in a behaviour (Simons-Morton & Chen, 2006). In society, peer influence commonly has a negative connotation and is often referred to as “peer pressure”. Similarly, peer influence has most frequently been studied from the perspective of risk taking and reward processing. Before this perspective is addressed
further, it is important to note that peer influence in adolescence isn’t solemnly negative. It has been demonstrated that adolescents show greater prosocial behaviour when provided with positive feedback from peers about prosocial behaviour (Van Hoorn, Van Dijk, Güroğlu, & Crone, 2016; van Hoorn, van Dijk, Meuwese, Rieffe & Crone, 2016). In this sense, peer influence provides both opportunity and threat, it is able to encourage healthy as well has harmful behaviour.
In regard to adolescent risk taking and influence of peers, general peer influence predicts risky behaviour such as drinking (Eisenberg, Golberstein & Whitlock, 2014; Weitzman, Nelson & Wechsler, 2003) and substance use (Bahr, Hoffmann & Yang, 2005). In an experimental context, set-ups have been created in which risk behaviour can be observed in a virtual game. For example, in a car driving simulation, participants had to drive as far as possible, but they come by a series of yellow lights. Participants could either wait at the traffic light, or ignore it, at the risk of a car accident. Participants consisted of adolescents, undergraduates, and adults, who played this video driving game alone or in the presence of two friends. The peers were allowed to give advice about stopping or continuing to drive, and the player was explicitly instructed that he or she was free to follow or ignore given advice. Comparison showed that the presence of peers doubled the number of risky behaviours in adolescents (in undergraduates there was a rise of 50%, in adults a rise was absent; Gardner & Steinberg, 2005). Both adolescents and undergraduates, but not adults, engaged in greater risk taking and rated the benefits of various risky behaviours relatively higher in the presence a peer.
Another task that is frequently employed in this paradigm is the Balloon Analogue Risk Task (BART). In this task participants have to inflate a balloon in simulation, for every puff of air they receive a small amount of money. Participants can stop a trial at any point and collect the money. However, if the balloon explodes, the money is lost. The amount of puffs it takes for the balloon to explode varies per trial. The output of the BART is the number of explosions and average pumps. This task is based on the real-life concept that risk-taking is rewarded up to a point until further risk-taking leads to negative outcomes and said to measure risk propensity (White, Lejuez & de Wit, 2008). In a study by Cavalca and colleagues, participants would either perform the BART alone, or in the presence of a computer simulated peer (2013). In the peer condition, participants were told that their performance would be observed online by another adolescent, who would provide suggestions regarding pumping strategies with a single statement in a chat room style text box. It was made explicit that following peers’ instructions would not affect whether they got paid or not. A picture of an adolescent (“the one who was giving feedback”) was display on the screen as well. This study found that adolescent smokers caused more explosions
in the peer BART condition, and they thus take more risks in the computer-simulated presence of a peer (Cavalca et al., 2013). Additionally, it was found that peer influence can also affect the perception of risk taking (Knoll, Magis-Weinberg, Speekenbrink & Blakemore, 2015). In a study, participants were asked to rate certain risk behaviours (“Crossing the street on a red light”). Afterwards, they would be presented with their own scores, or the scores as rated by other adolescents or adults. Next, participants were able to re-rate the risk behaviour. The second ratings were significantly influenced by the risk ratings of other people. In this sample, this socially driven change decreased steadily from late childhood to early adulthood (Knoll et al., 2015).
Collectively, this body of research proposes the influence of peers to be an important factor of risky behaviour in adolescence. In order to establish a standardised index in sensitivity of peer influence for research, Steinberg & Monahan developed an index called “Resistance to Peer Influence” (from here on “RPI”, 2007). It measures the general susceptibility to peer influence by asking respondents to pick one of two statements that fits them best (for example, “Some people go along with their friends just to keep their friends happy” or “Other people refuse to go along with what their friends want to do, even though they know it will make their friends unhappy”). The RPI is a widely used measure, validated in nearly 4000 individuals ranging in age from 10–30 and varying in ethnicity and socioeconomic status (Pfeifer et al., 2011). Furthermore, an important advantage of the RPI, is that it does not focus on antisocial or deviant influences. Instead it targets primarily neutral influences and could minimize socially desirable responding (Sumter, Bokhorst, Steinberg & Westenberg, 2009). Additionally, not all adolescents are under pressure to engage in antisocial or deviant behaviour (Brown, 2004). For instance, some peer groups might hold a norm of performing well in school (instead of a deviant norm like doing drugs). The neutral aspect of the RPI can thus allow it to be utilised in a broader population, one that doesn’t necessarily only hold deviant influences. Longitudinal data suggests that, on average, resistance to peer influence is lowest in late childhood and early adolescence, but then increases linearly between the ages of 14 and 18 (Steinberg & Monahan, 2007). It is interesting to note that periods of elevated risk taking and lowest RPI scores in adolescence seem to cooccur. Combining the factors discussed so far, it depicts adolescence as a time where (A) the presence of peers can increase risk behaviour in adolescents in experimental setting, (B) increasingly more time is spent with peers in real life, (C) a standardised measure of peer influence (the RPI) is low, and (D) elevated risk behaviour is observed in real life. Based on the combination of these factors, it is argued here that the RPI plays an important role in adolescent risk taking. Empirical data has demonstrated that peers can cause riskier behaviour. In real life, adolescents spend more time with their peers, which creates greater opportunities for such
inciting influences on risky behaviour to occur. Due to generally low RPI-scores, adolescents tend to be more susceptible to said influences. This creates a scenario in which adolescents have more exposure and vulnerability for peer influences, which could ultimately lead to the observed inflated risk behaviour. Furthermore, individual differences in RPI-scores could (in part) explain individual variability in risk behaviour. In sum, the RPI presents a promising factor in the understanding on adolescent risk taking and requires further investigation.
Indeed, research has been conducted on this topic, but with mixed results. For example, the aforementioned study by Cavalca and colleagues investigated the RPI alongside the BART (2013). As previously stated, participants would either perform the BART alone, or in the presence of a (computer simulated) peer. Adolescent smokers took more risks, scored higher on impulsivity and lower on RPI (Cavalca et al., 2013). However, those with low RPI did not experience greater changes in risk taking in the peer condition. Consistent with Cavalca et al., Kiat and colleagues also found no significant relationship in young adults between RPI and risk level in the BART (Kiat, Straley & Cheadle, 2016). In contrast, when the BART was employed in young adults in an alone versus peer condition, RPI had a significant indirect effect on risk taking via reward responsiveness (Reniers et al., 2017). Reniers and colleagues argue that the absence of a direct effect might be due to context. The context of the experimental task is rather specific: a participant is inflating a balloon digitally in a research building (in some studies even in an MRI machine or with an EEG helmet on). The context of the RPI is very global, it measures general terms and tendencies: it usually reflects behaviour or inclinations within a natural environment or group. Due to the mismatch of a general measure (the RPI) and a specific context (the BART), a direct effect of RPI-score on risk behaviour in the BART might not have been found. Thus, the differential context could have been a limiting factor. If this is the case, one would expect a clear relation between the RPI and a measure of risk that is based on a context more similar to that of the RPI. No relation would be expected between the RPI and a measure of risk that is based on a context more dissimilar to that of the RPI.
This is precisely what this paper aims to investigate, whether the RPI predicts two types of risk behaviour distinctively: firstly, risk behaviour that is more socially embedded and secondly risk behaviour that is more affectively and socially neutral. For this analysis, data will be collected from Dutch adolescents for a two-week (15 days) period. This data will include RPI-scores, self-reported daily risk scores and daily lottery behaviour. The self-reported daily risk scores will be based on questions like “did you stand up for someone today?” and daily lottery behaviour represents choices between two lotteries also known as the Holt & Laury Task. Although comparable data that considers the RPI and Holt & Laury is lacking, it has been indicated before that the Holt &
Laury score does not relate to health risk taking (Szrek, Chao, Ramlagan & Peltzer, 2012), or predicts only very little of the variance of health risk taking (Anderson & Mellor, 2008). These studies suggest that the Holt & Laury may not be extremely fit to be translated into or compared with real life data/general tendencies. Thus, as the Holt & Laury is not embedded in social circumstances but focused on logical decision making in gambling scenarios, it can be argued not to fit the general social tendencies measured by the RPI. Therefore, it is hypothesised that the RPI does not predict Holt & Laury scores. In contrast, risk behaviour that is more socially embedded (the self-reported daily risk), and therefore assumed closer to the context of the RPI, is hypothesised to be predicted by the RPI. It is expected that higher RPI-scores are aligned with lower risk taking. To the author’s knowledge, the RPI has not been investigated in relation to self-reported daily risk, nor has a comparison been made on the relation between the RPI, self-reported daily risk scores and daily lottery behaviour. This research could present insight on whether context of measured risk needs to be considered when employing the RPI. It may also elucidate why previous results have not found a direct link between RPI and experimental risk behaviour, as well as indicate what type of risk behaviour poses a threat to adolescents in regard to peer influence.
Methods
This study is part of a larger longitudinal research project (“BrainGames”) at the Vrije Universiteit Amsterdam, which has been approved by the Ethics Board of the Vrije Universiteit. This study entails cross-sectional research on the basis of questionnaires and behavioural tasks.
Participants
This data set is part of an ongoing collection of data, and the total sample is still incomplete. To establish the impact of the smaller nature of the sample on the statistical power, a one-sample power t-test was conducted. The results indicate power of .19, .75 and .98 to detect a small (d = .20), medium (d = .50) and large (d = .80) effect size respectively, with an alpha of .05. Indicating that within this sample medium or large effect size are relatively likely to be found. However, this sample size is too small to detect small effect sizes (approximately 200 participants are required for that). Additionally, a check was performed to establish if enough power is obtained to compare participants of different genders and educational levels. Proportional power tests suggest that the sample is too small to detect large effects between genders and educational levels (d = 0.8, power =.56). Results showed that a total sample of 50 participants with two equal sized groups of N = 25 would be required to achieve a power of .80.
Participants were recruited online, through guest lectures at secondary schools or via personal network. All participants lived in the proximity of Amsterdam, the Netherlands. Recruitment and inclusion were slightly altered towards different age groups and genders throughout the study, to ensure a reasonable ratio in the sample. Every participant was screened before inclusion of the study to determine if A) there was no gambling addiction present, B) no psychological diagnosis had been made, C) the participant had no MRI contra-indications and lastly D) the participant was right-handed. Participants were paid 50 euros plus a bonus (up to 75 euros) based on the money earned in the various games they played. Data were collected between January 2019 and March 2020.
Procedure
All participants went through the following sequence of recruitment and preparatory steps. Participants would be introduced to the study through guest lectures, personal network or find the study online. If the participant was interested in partaking, he or she would provide their personal information. After this initial contact, participants got an elaborate explanation of what participation of the study entailed. They were told the study investigated decision making in adolescents (and not risk taking) to prevent a priming effect. Due to the fact that participants were minors, it was explained that consent would have to be given by their parents. Consent forms were sent to the prospective participant and parents received a phone call with additional information. Once
the participant had filled in an online survey from home (which established the absence of a gambling addiction), returned the necessary forms and parents had given their consent, participant was officially included in the study.
Participation in our study consisted of two parts. Firstly, participants had to fill in an online diary every day for 15 consecutive days. This required about 5 minutes of time each day, and could be done on a personal phone, tablet or laptop. If participants forgot to fill in the diary, they were contacted (usually the next day) and requested to fill it in retrospectively. Secondly, participants were asked to take part in three fMRI-scan during the two-week period in which they wrote the online diary. These scans occurred at the Spinoza Centre for Brain Imaging in Amsterdam in a 3T-scanner. During the scan session, participants completed numerous games and filled in surveys on site. The fMRI scan and data from the games will not be used in this study, however data from the survey that were filled out on site will be considered (demographics and the RPI-questionnaire). The participation period was scheduled for two normal weeks: weeks that did not include unusual events such a school trips or high stress exam periods. This was done to ensure the data would be based on “typical weeks”, and not special circumstances.
Measures
Demographics. Information on age, gender and the level of high school education was collected at the first visit to the Spinoza Centre. For the purpose of analysis, and to ensure adequate cell sizes. Participants (n=30) were 16 or 17 years old (mean= 16.78, SD = 0.58) at the start of the study. About two thirds (63%) of participants was female. Two thirds of participants was enrolled in secondary school of level “VWO” (preparatory for university, 63%), the other participants were partaking in “HAVO” (preparatory for applied sciences, 37%).
Self-reported Daily Risk Behaviour. In the online diaries, we measured self-reported risky behaviour for many types of risk behaviours, such as “Did you come home past your curfew today?”, “Did you skip class today?”, and “Did you drive a scooter without a helmet today?”. For this paper, only a subset of these questions will be analysed. A distinction is made between non-social and social risk within the diary dataset. This distinction is based on the contexts in which the types of risky behaviour occurred. To be categorised as social, risk taking should undoubtedly occur in a setting with other people around or be social by itself (i.e. lying to somebody requires the presence of another, and texting entails an interaction with another). For the non-social risk taking, they clearly did not occur in a social setting, nor were social in itself (i.e. choices in a task performed alone on mobile phone).
Social Self-reported Daily Risk Behaviour. Using aforementioned definition, the social risk was assessed using seven questions: (1) Did you stand up for someone today? (2) How scary was it for you to stand up for
someone? (3) Did you cheat on a test today? (4) Did you bike with three people next to each other today? (5) Did you text while biking today? (6) Did you lie to someone today? (7) How angry would the person whom you lied to be, if they found out? For an overview of all social questions and answering options, see Table 1.
Table 1.
Questions categorised as social, and their answering possibilities
Social Question in Diary Answering Possibilities
Did you stand up for somebody today? 1 Yes 2 No How scary was it for you to do this [standing
up for someone]?
1 Not at all 2 A little 3 Pretty 4 Very 5 NA
Did you cheat on a test today? 1 I didn’t have a test today 2 I didn’t cheat 3 I cheated Did you bike with three people next to each
other today?
1 No 2 Yes, 1-5 Mins 3 Yes, 6-10 Mins 4 Yes, 11-15 Mins 5
Yes, 16-20 Mins 6 >20 Mins
Did you text while biking today? 1 No 2 Yes, 1-5 Mins 3 Yes, 6-10 Mins 4 Yes, 11-15 Mins 5
Yes, 16-20 Mins 6 >20 Mins
Did you lie to somebody today? 1 Yes 2 No
How angry would the person be whom you lied to, if they found out?
1 Not angry at all 2 A little 3 Pretty Angry 4 Very Angry 5
NA
Total score on each item was calculated per participant over the entire participation period (2 weeks). Total score for cheating is presented in a percentage of cheating on tests. This was done to incorporate the fact that some participants had many tests, while others had none. For the two questions regarding biking, a total time of said activity was calculated. For this calculation, the minimum from an answer option was chosen. I.e. if a participant entered “3 Yes [I’ve texted while biking], 6-10 Mins”, the number six was used for calculation. This was done to keep the data as precise as possible: it’s certain the participants had to text while biking for at least 6 minutes to select this answer. If, for example, the median of 8 mins is used in the calculation, a certain degree of certainty is lost. In this case, it’s impossible to be certain the participant texted while biking for 8 mins (and not more or less). The frequency that they did not partake in this risky behaviour was also calculated. The frequency of lying and the sum of severity of its possible outcomes were both calculated and used as a score for the lying behaviour. Both the frequency and severity were considered, as they represent the same axis: on one extreme someone never lies and thus has a severity of zero, in the middle someone frequently uses “small” lies, and on the other extreme someone lies frequently, with sever consequences.
Non-Social Daily Risk Behaviour. Each day, after participants completed the questions about their risky behaviour, they were asked to choose between different gamble scenarios. For the gamble scenarios the 10-point Holt & Laury measure was used. The Holt & Laury task is a well-known risk elicitation task (Crosetto & Filippin,
2016), where one has to choose between two scenarios of a lottery. Scenario A had two moderate payoffs, while scenario B had one high pay-off and one low pay-off. The probabilities changed from item to item, while the payoff remained the same. Hence, the expected values of the two options changed for each consecutive item. For the first four items, the expected value for option A was safer, and option B was riskier. From the fifth item onwards, B had higher expected value. The number of times a participant chose option A, determined their risk attitude (Holt & Laury, 2002). The Holt & Laury was modified from the original written version to a visual pie-chart version for the sake of clarity and comprehensibility. The written version is available via the Science of Behaviour Change Society here. The visual version can be found in the Appendix I and was created for this project, though a visual version of the task has been used successfully before (Doukianou, Daylamani-Zad, Lameras & Dunwell, 2019).
Resistance to Peer Influence. Resistance to peer influence was measured with the RPI-index developed by Steinberg & Monahan (2007). The RPI is a self-report measure of resistance to peer influence. Participants were presented with two statements and asked to choose the one that fits them best (i.e. “Some people go along with their friends just to keep their friends happy” versus “Other people refuse to go along with what their friends want to do, even though they know it will make their friends unhappy”. Participants also expressed if chosen statement fits them really well or only partly. This led to a 4-point scale with “really true for statement A” on one extreme, and “really true for statement B” on the other. The RPI responses were inverted with item two, six and ten. Scores ranged between 10 and 40, a higher overall score indicates greater resistance to peer influence. As aforementioned, the RPI is a widely validated measure (Pfeifer et al., 2011) and an important advantage of the RPI is that it does not focus on antisocial or deviant influences. Instead it targets primarily neutral influences and could minimize socially desirable responding (Sumter, Bokhorst, Steinberg & Westenberg, 2009).
Data analyses
A number of data checks will be performed before the regression analysis is conducted. The data checks are as follows.
Outliers. The daily self-reported dataset will be tested for outliers based on a z-score. If outliers are present, as indicated by a z-score smaller than minus three, or larger than three, tests will be conducted with and without the outliers. If results do not differ significantly, the outliers will remain included in the dataset. This is done because the dataset is small, and all data points are real data (meaning the outliers were not caused by
instrument error or participants simple skipping over questions). In order to preserve as much data as possible, the outliers will be included if they don’t influence the results significantly.
Normality Check for RPI-scores. A check for normality will be performed with the Shapiro-Wilk test for RPI-scores. As one person did not fill in the RPI subpart of a questionnaire, the sample size for analysis regarding the RPI will be equal to 29. If normality cannot be assumed based on the Shapiro-Wilk test, possible outliers will be investigated via z-scores. If any of the RPI-scores are larger than three or smaller than minus three z-score, they will be classified as outliers. If this is the case, a robust linear regression will be used on the entire dataset. The potential outliers can be included in a robust regression, as this type of regression devises estimators that are not as affected by outliers (Rousseeuw & Leroy, 2005).
Confirmatory Factor Analysis. As previously mentioned, questions from the daily self-reported risks were selected based on an assumption they all hold a social risk component. To test this assumption, a factor analysis will be performed. Factor analysis is a useful tool to investigate underlying structures of data. The strength of this analysis is its ability to construct latent variables (also known as factors) that are not measured directly but are estimated in the model from several measured variables, each of which is predicted to touch the latent variables (Ewalds-Kvist, Högberg & Lützén, 2013). There are two types of factor analyses, the exploratory factor analysis (EFA) and confirmatory factor analysis (CFA). The goal of an EFA is mainly to explore, identify and classify data to generate hypotheses about the underlying structure (Child, 2006). On the contrary, CFA explicitly tests hypotheses about observed variables and latent variables (Jackson, Gillaspy & Purc-Stephenson, 2009). It is common practice to first perform an exploratory factor analysis (EFA) on a subset of the data, and form a model based on the EFA results. Subsequently this theory is tested in a CFA with the rest of the data. This data set is too small to perform both an EFA and CFA. A hypothesis regarding the structure of the latent variables has already been mentioned, in this case the assumption that all chosen questions share an underlying social risk aspect. The presence of this hypothesis fits the conceptual use of a CFA better than that of an EFA. Therefore, it was decided to perform a CFA with a one-factor model were all variables lean onto one latent variable (“global social risk”). If the results from the CFA indicate that the one-factor model does not fit the data well, its fit will be compared to a model with two latent variables. This two-factor model separates biking items from other items in the social risk battery. This division into two latent variables is based on the fact that the context related to traffic is specific for the biking items, other items do not take place is said context.
The model that fits the data (best), will determine how the scores will be regressed with the RPI scores. If the model with one latent variable fits the data well, one “global social risk” score will be calculated (based on the responses of all the items that lean into this one latent variable). This means that all frequencies, severities/scariness, time spent on risky behaviour and percentages cheating will be summated. Because these items were rated on different scales, they will have to be standardised first. Variables that are measured on different scales do not contribute equally to an analysis and may create a bias. For instance, a variable that ranges between 0 and 100 (i.e. percentage cheating) will outweigh a variable that ranges between 0 and 15 (i.e. frequency of lying). Once the variables are standardized, they can be summated. This singular total social risk score will be used in the regression if the one-factor model fits the data.
If the model with two latent variable fits the data well, two different cumulative risk score will be calculated. Firstly, a total score for time spent texting while biking and biking with multiple people next to each other (“biking risk”). As these responses were given on the same scale, they don’t require standardization and can simply be summated. Secondly, a total score was calculated for percentage cheating on a test, frequency of lying, the severity of consequences of lying, frequency of standing up for someone, and how scary standing up was. This will be done by summation of the standardized scores. In sum, if the model two latent variable fits the data well, two separate risk scores (calculated as explained in this section) will be used in the regression.
Comparison of social and non-social variance. Ideally, after the structure behind the social risk taking is presented, a second order CFA would be conducted. In this second order CFA, it is possible to test if the social and non-social data feed into one common latent variable of risk taking. Unfortunately, this dataset is not big enough to perform a second order CFA. In order to still quantify if there’s common variance between the social and non-social risk taking, Pearson’s correlational coefficient will be calculated. This will establish to what extent social and non-social daily risk taking are related. It is expected that the two scores will be correlated, as they both measure a type of risk taking. If they are exceptionally strongly related, caution will be warranted when performing the linear regressions. Entering two strongly related variables as dependent variables in separate linear regression, could lead to two different regressions explaining the same variance. Therefore, if the social and non-social daily risk taking are strongly correlated, caution will be warranted when interpreting the linear regressions.
Regression. Based on the results of the CFA, either two or three regressions will be performed on the RPI-scores. The first linear regression will predict the Holt & Laury mean score based on the RPI score. If the
CFA results indicate the one-factor model, one more linear regression will be performed. In this regression, global social risk (as examined by all the items of the daily risk battery) will be regressed on the RPI-score. If the CFA results present the two-factor model, two more linear regression will be executed: one where biking risky behaviour will be predicted, another where the other social items from the self-reported daily risk will be predicted. If there are outliers in the RPI-scores, robust regression will be used.
Results
Outliers.
The self-reported risk dataset was tested for outliers based on a z-score. A handful of outliers was present (with z-scores ranging from -3.15 to 4.55). Tests were conducted with and without the outliers. The comparison of tests showed that results were not significantly altered with and without the outliers, and thus the outliers remained included in the dataset. This was done in order to preserve as much data as possible, as the dataset is small, and none of the data points were driven by instrument error or inattentive participants.
Normality RPI.
The Shapiro-Wilk test was conducted to test normality within the RPI-scores. The test showed significant departure from normality, W= .92, p= .03. To investigate possible outliers, z-scores were examined. None of the RPI total scores were larger than three/smaller than minus three z-score. Thus, none were classified as outliers. The small nature of the sample probably led the Shapiro-Wilk test to return significant values, and consequently normality was assumed for further analysis. This means that it was not necessary to conduct a robust linear regression, instead a normal regression would suffice.
CFA.
To test the assumption that all selected items from the diary hold a social risk component, a CFA was performed with a one-factor model. Overall the goodness of fit indices suggest that an one factor model does not fit the data well: χ2(26)= 121.49, p= .00 , SRMR = .250, RMSEA = .4 (90% CI =.329-0.473), CFI = .484, TLI = 0.285. The χ2 value is significant, indicating a poor fit. Additionally, the SRMR, RMSEA, CFI and TLI, are also indicative of poor model fit (for an overview of these indicators and their cut-off value, see Hooper, Coughlan & Mullen, 2008). These results indicate that within this sample the selected social risk items do not converge in one latent factor. As explained previously, an alternative two-factor model had been postulated. As the one-factor model did not fit the data well, a subsequent CFA was carried out to examine the alternative model with two latent variables (separating biking risk and specific social risk as measured by percentage cheating on a test, lying and
standing up for someone). The correlation table of all items (see table 2), suggests a moderate to strong relation between the two biking items, which is a promising sign for the fit of the two-factor model. Overall, the goodness of fit indices of the two factor model are indicative of a fair fit: χ2(11)= 14.85, p= .19 , SRMR = .182, RMSEA = .123 (90% CI =
.
00 -.267), CFI = .973, TLI = .948, with a standardised covariance between the latent variables of -.024. The χ2 value is insignificant, indicating a good fit. The CFI and TLI demonstrate great fit, while the SRMR and RMSEA are indicative of poorer model fit. Overall, all the values in the adjusted model are better than the original model.Based on the two-factor model, two cumulative scores, to be used in the regression, were calculated. Firstly, a total score for time spend texting while biking and biking with multiple people next to each other. Secondly, a total score was calculated for percentage cheating on a test, frequency of lying, the severity of consequences of lying, frequency of standing up for someone, and how scary standing up was. This total score was calculated by summation of the standardised scores.
Table 2.
Correlation table of social items in the diary data.
Percentage Cheating Frequency Stand up for someone Scariness Stand up for someone Frequency Lies
Severity Lies Biking with multiple people Texting while biking Percentage Cheating 1 Frequency Stand up -.31 1 Scary Stand up -.35 .88 1 Frequency Lies -.17 .09 .12 1 Severity Lies -.21 -.07 -.05 .94 1 Biking with multiple people .37 -.02 -.12 .24 .19 1 Texting while biking .47 .11 .13 -.07 -.10 .43 1
Note. The correlations for Percentage Cheating are based on n=23 participants (instead of n = 30 as for the other correlations) due to not-applicable data (participants who had no tests).
Comparison of social and non-social variance.
In order to quantify if there’s common variance between the social and non-social risk taking, two correlational tests were performed. One for the relation between the social risk factor (as measured by cheating, lying and standing up for someone) and the Holt & Laury mean (r(28) = -.23, p-value = .22), and another for the biking total and the Holt & Laury mean (r(28)= -.28, p-value = .13). This suggests that social and non-social risk taking are not significantly related.
Regression
Non-social Risk Behaviour.To quantify whether the RPI-scores predict the non-social risk behaviour, as measured by the Holt & Laury task, a linear regression was performed, β = -2.2, t(27) = -1.44, p= .16. This indicates that the RPI is not a significant predictor for the Holt & Laury scores. This is in agreement with the hypothesis.
Social Risk Behaviour. Previously, a CFA indicated the presence of two latent variables within the social risk behaviour data of the diaries. Based on these factors, two totals scores were calculated and used in the linear regressions (see methods section for further detail on how this score was calculated). Firstly, the total biking risk was regressed on RPI-score, β= 0.06, t(27)= 1.19, p= 0.24. These results do not align with the hypothesis, as it was expected that the RPI-scores would significantly predict this score. Instead, the RPI-scores does not significantly explain part of the variance within the biking risk. Secondly, the social risk score (as measured by cheating on a test, lying and standing up for someone) was regressed on RPI-score, β= -.21, t(27)=
-
2.09, p=.046 (see figure 1). RPI was thus a significant predictor for self-reported cheating, lying and standing up for someone. These results confirm the hypothesis, as they indicate that the RPI-scores significantly predict these social scores. The overall fit of the model, F(1,27) = 4.38, p = .05, R² = .14, indicates that the model explains approximately 14% of the variance within the social risk scores. In sum, a negative linear relationship between the RPI and social risk taking was found, but not between the RPI and biking risks or the Holt & Laury task.Exploratory Analysis.
The CFA results indicate that the social items lean onto two different latent variables. An exploratory analysis was performed to see if a difference could be found in perception and likelihood of the items on biking compared to the other items. At the start of this study participants rated each behaviour on how risky they perceived the risk, and how likely they were to partake in in. This data was used to explore the idea that the factorial structure could (in part) be based on a difference in “commonness” or perceived riskiness. Pearson product-moment correlation coefficients were computed, which showed that the rated likelihood of texting while biking only had a (positive) significant correlation with the likelihood of standing up for someone (r(28) = .37, p-value = .04). For the rated riskiness of texting while biking, one significant positive correlation was found between the texting while biking scores and cheating on a test (r(28) = .58, p-value = .0006). The other ratings on texting while biking were not significantly different from the other items for likelihood (for lying: r(28) = .12, value = .53, for cheating on test: r(28) = .008, value = .68) or riskiness (for standing up: r(28) = .12,
p-value = .51, for lying: r(28) = .25, p-p-value = .19). None of the ratings of the item of biking with multiple people were significantly different from any of the other items on likelihood (for standing up: r(28) = -.01, p-value = .92, for lying: r(28) = .14, p-value = .43, for cheating on test: r(28) = .005, p-value = .98) or on riskiness (for standing up: r(28) = -.08, p-value = .66, for lying: r(28) = .08, p-value = .65, for cheating on test: r(28) = -.05, p-value = .77)
Figure 1.
Regression plot predicting social risk based on the RPI-score.
Note. Black dots are observed values. Blue line presents the linear regression slope. Social risk score based on the standardised responses to the items of cheating, lying, and standing up for someone.
Discussion
This study investigated risk taking in adolescents and influence of peers. Peer influence is said to increase risk taking in adolescence in experimental tasks, however a standardized index for resistance to peer influence (RPI) has produced inconsistent results when it’s related to risk taking. These inconsistencies have been attributed to a difference in context between measures. To substantiate this notion, this paper examined the relation between the RPI and two types of risk measures, different in their context. Specifically, self-reported social risk scores and performance on a gambling task (the Holt & Laury Task) were regressed on the RPI. Results indicate that the RPI is a good predictor for part of the self-reported social risk while it does not predict the Holt & Laury score.
The self-reported daily risk questions were part of a new questionnaire and items were selected for this paper based on an assumed shared social risk component. To test this assumption, a CFA was performed with a one-factor and two-factor model. The CFA results don’t support the one-factor model, in which all items lean into one latent variable. Instead, the CFA results support the two-factor model, in which the items on cheating, lying, and standing up for someone, fall under a different latent variable than the items on biking behaviour. This means that in this sample, the CFA could not confirm that all selected items share a common component, presumably the social risk aspects. Thus, it is not possible to conclude that the items measure the same construct. There is little literature to compare the two-factor model to, due to the novelty of this approach. Nevertheless, speculation on the two-factor structure is possible. The biking behaviour represents a very specific set of risks, that can have severe consequences (accidents, injuries, etc.). It is questionable whether consequences like physical injuries or disabilities due to traffic incidents are similar to social consequences of lying or standing up for somebody. In this sense, both risk behaviours occur in a social setting, but the consequences of biking could be considered as less social than the consequences of the other items. It could therefore be argued that cycling risk behaviour is distinctively different in this sense, and that this may be an explanation why the one-factor model did not fit. Alternatively, cycling risks could encompass behaviours that are considered as socially accepted or “normal” by adolescents. For instance, 80% of Dutch youth use their mobile phone while biking (Ziemerink, 2019), a remarkably high number. Possibly, this cycling risk differentiates itself from others social risk taking by its omnipresence, and therefore the biking items theoretically could measure a construct of risk with a lower threshold to participate. Participants in this study rated each behaviour at the start of the study on how risky they perceived the risk, and how likely they were to partake in in. This data was used to explore the idea that the factorial structure is (partly) based on a difference in commonness or riskiness. However, this exploratory analysis showed that the
rated likelihood of texting while biking only had a (positive) significant relation with the likelihood of standing up for someone. For the rated riskiness, one significant positive relation was found between the texting while biking scores and cheating on a test. Both of these positive relations had relatively high correlations. The item on biking with multiple people was not rated different from any of the other items. Because only two sets of these ratings were significantly different, it is unlikely that the participants perceived the overall likelihood or riskiness differently per latent variable. Considering this, it is difficult to argue how this factorial structure arose exactly. One point that becomes evident from the current factorial organization is that not all risk measures function the same way. This notion may contribute to create awareness regarding the spectrum of risk behaviour and highlights that different approaches might be needed to tackle distinct types of risk within the spectrum. Future research will be needed to see if this the latent variable structure can be replicated and might aid comprehension on how the factorial structure arose.
Additionally, as an alternative to the second order CFA (which was not possible to conduct due to the sample size), a correlation was computed between the Holt & Laury and biking score and the Holt & Laury and other social risk factor. This was done to test the idea that these measures share an aspect of general risk taking. Non-significant relations were found between the Holt & Laury and biking score and the Holt & Laury and specific social risk factor.Unexpectedly, this suggests that social and non-social risk taking are not significantly related. However, the correlation coefficients indicate moderate correlation and p-values are not abnormally high. Within this small sample, it can be argued that lack of power presented these constructs as unrelated, when in fact they are. It would be interesting to see if future research with a larger power is able to detect a significant relation between these constructs.
To quantify whether the RPI-scores predicts risk behaviour three regression were performed. Firstly, the results suggest that the RPI is a good predictor of self-reported social risk-taking, as measured by lying, cheating and standing up for someone. This is in line with the hypothesis. Higher RPI-scores predicted lower social risk scores. These findings are in agreement with existing literature, for instance, smith and colleagues found that participants with relatively low RPI-scores increased rates of monetary gambles on a probability gambling task when they were observed by a peer (Smith, Chein, Steinberg, 2011), and Peake and colleagues established that the degree of heightened risky decision following a social exclusion manipulation was negatively correlated with RPI-scores (Peake, Dishion, Stormshak, Moore, & Pfeifer, 2013). In regard to self-reported risk taking, it has been found that RPI and the indicators of risk behaviour and delinquency (IRBD) scale were significantly
negatively correlated by age 13 (Pfeifer et al., 2011). The latter study is partly comparable to the current study. Both risk measures are self-reported, yet the IRBD assesses substance use and delinquency, such as drinking, stealing, or getting into trouble with the police. Current paper adds to this literature by presenting self-assessed risk behaviour that is less deviant/more neutral, and as not all adolescents engage in this type of delinquent behaviour, the current study could be more applicable to the general population of adolescents.
It should be noted that the RPI explains 14% of the variance of the self-reported risk-taking (measured by lying, cheating and standing up for someone). It can be argued this is due to additional context of the measured risky behaviour. This additional context is unknown but could have potentially affected risk behaviour. For instance, standing up to a best friend can be experienced vastly different than standing up to a group of strangers. Indeed, it has been shown that the identity of the observer/peer that watches a participant matters (Wolf, Bazargani, Kilford, Dumontheil & Blakemore, 2015). Older adolescents showed poorer performance when being observed by their friend relative to an unknown peer. Furthermore, quality of peer relation affects risk taking: peer support has been associated with lower self-reported risk-taking behaviour, while peer conflict was associated with greater risk-taking behaviour (Telzer, Fuligni, Lieberman, Miernicki & Galván, 2015). Interestingly, high levels of peer support served to buffer the association between conflict and risk-taking. These studies highlight the important role of quality of peer relationships in risky behaviour.
Secondly, it was found that the RPI measure was a non-significant predictor of risky biking behaviour. This is unexpected, as it was hypothesized that RPI scores would predict (all constructs within) social risk scores. As aforementioned, the biking behaviour represents a very specific set of risks, that can have severe consequences (accidents, injuries, etc.). The RPI-score is based on questions like, “some people think it’s better to be an individual even if people will be angry at you for going against the crowd” or “some people hide their true opinion from their friends if they think their friends will make fun of them because of it”. The examples given here, intuitively fit behaviour such as standing up for someone (“going against the crowd”) or lying (“hide their true opinion”) better. Accordingly, reviewing these items suggests that the RPI questions might not cover the biking behaviour as extensively as it covers the other risks. In addition to this, it has been discussed that consequences like physical injuries could be very different from the social consequences of lying or standing up for somebody. The RPI is mostly focussed on social settings and social consequences of risk behaviour. It could therefore be argued that cycling behaviour is not well calibrated with the RPI and that this may be an explanation why no insignificant result was found.
To the author’s knowledge, research on adolescent biking risks has not been performed previously. However, driving risks in traffic have been studied more extensively, emphasizing the risks adolescents take in such situations (i.e. Shope, Waller, Raghunathan & Patil, 2001). Furthermore, as previously mentioned the presence of peers doubled the number of risky behaviours in adolescents in a car driving simulation (Gardner & Steinberg, 2005). The majority of this literature is based on data from the USA (for review see Ouimet et al.,2015 or Klauer, Ehsani, McGehee & Manser, 2015), where the legal age for driving is relatively low (in some states a learner’s permit is already granted to 14-year olds). It is not always clear how this data translates to other countries with different legal ages and/or traffic cultures. For example, in the Netherlands a strong bicycle culture is present, and more than one-quarter of all trips made by Dutch residents are travelled by bicycle (Harms & Kansen, 2018). As there is currently no research available on this topic, it would be interesting for future studies to examine RPI-scores and risky traffic behaviour in real life beyond a focus on driving behaviour in the USA. Results from this type of research could have implications for relevant policy making (i.e. legal driving age, restrictions on adolescents partaking in traffic with peers, etc).
Lastly, the Holt & Laury score was regressed on the RPI-scores. Including this measure allows comparison of the RPI as predictor for social and non-social risk taking. No evidence was found that the RPI predicts risky behaviour as measured by the Holt & Laury mean score. This confirms our hypothesis and is in line with studies that have indicated that the Holt & Laury might be unfit to be translated into or compared with real life data/general tendencies (Szrek, Chao, Ramlagan & Peltzer, 2012; Anderson & Mellor, 2008). Thus, this paper emphasises the importance of considering the type and context of a risk measure, as the RPI significantly predicted (some domains of) social but not non-social risk taking.
Limitations & Future Research
When one considers using an experimental task or self-reported data, there is often a trade-off. In an experimental setting, many factors can be control and the effect isolated. On the other hand, translation to real life can be difficult. In contrast, in real life data there is less control over other factors, and more noise can be present in the data because of this. However, real life data holds an important advantage that it’s already in its “natural” environment, has high ecological validity, and does not require translation. Here, it is argued that research into the RPI should not overlook the importance of real-life data. Neglecting to take this data type into account, could severely undermine one’s understanding of a complex phenomenon. Experimental and real-life data can (and should) compliment and inspire one another, and research involving both types is most optimal. However, as this
paper emphasises the importance of considering the type and context of a risk measure, caution is warranted when interpreting results comparing socially embedded and more logical measures of risk taking. Relations (or lack thereof) may be due to misfit of measure, but not necessarily of constructs.
Additionally, it would be interesting to include subgroups based on gender and educational level in future research. There was adequate power to find moderate to large effect size in the whole sample, however, the sample was not big enough to allow reliable statistics on the subgroups. Thus, gender and educational level have not been considered in this analysis. Consequently, the results are generalizable for the whole population, but potential trends within subgroups cannot be distinguished within this sample. Recently it has been suggested there is a difference in RPI-scores based on gender (McCoy, Dimler, Samuels & Natsuaki, 2019), it would be interesting for future research to investigate this further.
Similarly, other psychological aspects might be distinctively altered in adolescence. For example, it has been proposed that adolescence is a period of increased sensitivity to positive feedback (McCormick & Telzer, 2017), and decreased sensitivity to negative feedback (Ernst, Pine & Hardin, 2006). This is said to contribute to increases in risk taking. Peer influence can be seen as a form of feedback, that can either be positive (“do something more”) or negative (“do something less”). Increased sensitivity suggests that adolescence would be more susceptible to positive feedback from peers. In this dataset, it is unclear to what extent participants received positive or negative feedback from their peers. Theoretically, it could be that two adolescents are equally resistant to peer influence, however one of the adolescents is part of a friend group that exerts a lot of positive feedback. Perhaps this adolescent would show more or less risk than his or her counterpart. In summary, an interaction effect between RPI and type of feedback received from peer could be present.
In addition, social stress has been found to facilitate risk in adolescents. For instance, participants were either asked to deliver a speech and solve math problems alone or under the scrutiny of two interviewers (who provide neutral/negative feedback, experimental set up also known as the Trier Social Stress Test). Adolescents who were under social stress perceived less risk in their environment and engaged in more risk-taking behaviour relative to controls (Jamieson & Mendes, 2016). Thus, social stress can facilitate risk in adolescents. Within this dataset, it is not possible to determine if participants were under social stress and what influence this would have had on risk taking.
Conclusion
In sum, this study examined risk taking in adolescents and influence of peers. As peers tend to increase risk behaviour in adolescence, greater resistance to peer influence was expected to relate to less risk behaviour. Less risk taking in adolescents could be a substantial improvement for society: an adolescent would be less of a danger to themself and others, and be less likely to engage in the behaviour in adulthood (many forms of risk behaviour initiated in adolescence elevate the risk for the behaviour in adulthood). It was demonstrated that an index for resistance to peer influence (RPI) predicts social risk behaviour in adolescents, as measured by cheating, lying and standing up for someone. Greater RPI-scores are associated with lower frequency and intensities of these risk behaviours. The RPI did not predict the self-reported biking risk scores nor the Holt & Laury score. The results for the Holt & Laury score were attributed to the context of the measures: a measure of global social risk tendencies (RPI) is a good predictor for a measure that is also socially embedded, but not for a measure that does not involve such an environment (Holt & Laury score). Additionally, unexpected non-significant results for the biking risk behaviour were theorised to be due to the fact that the biking risk behaviour is not well calibrated with the RPI, based on primary context of consequences (physical versus social risk), as well as the types of tendencies the RPI measures (focussed on a social setting). Based on these findings, this paper argues that consideration of the context of the measured risk behaviour is pivotal in research into adolescent risk taking. Caution is warranted when comparing socially embedded and more logical measures of risk taking. Relations (or lack thereof) may be due to misfit of measure, but not necessarily of constructs. Future research should consider additional influences, such as gender and social stress, in hope to elucidate this further. Conclusively, this research highlights the importance of the RPI in naturalistic risk taking in adolescence and accentuates resistance to peer influence as an important factor in adolescence risk taking.
References:
Alin, A. (2010). Multicollinearity. Wiley Interdisciplinary Reviews: Computational Statistics, 2(3), 370-374. Anderson, L. R., & Mellor, J. M. (2008). Predicting health behaviors with an experimental measure of risk
preference. Journal of health economics, 27(5), 1260-1274.)
Arnett, J. J. (2007). The myth of peer influence in adolescent smoking initiation. Health Education & Behavior, 34(4), 594-607.
Bahr, S. J., Hoffmann, J. P., & Yang, X. (2005). Parental and peer influences on the risk of adolescent drug use. Journal of Primary Prevention, 26(6), 529-551.
Bandura, A. (1986). Social foundations of thought and action. Englewood Cliffs, NJ, 1986.
Brown, B. B. (2004). Adolescents’ relationships with peers. Handbook of adolescent psychology, 2, 363-394.
Brown, B. B., Bakken, J. P., Ameringer, S. W., & Mahon, S. D. (2008). A comprehensive conceptualization of the peer influence process in adolescence. Understanding peer influence in children and adolescents, 17-44.
Cavalca, E., Kong, G., Liss, T., Reynolds, E. K., Schepis, T. S., Lejuez, C. W., & Krishnan-Sarin, S. (2013). A preliminary experimental investigation of peer influence on risk-taking among adolescent smokers and non-smokers. Drug and alcohol dependence, 129(1-2), 163-166.
Chein, J., Albert, D., O’Brien, L., Uckert, K., & Steinberg, L. (2011). Peers increase adolescent risk taking by enhancing activity in the brain’s reward circuitry. Developmental science, 14(2), F1-F10.
Child, D. (2006). The essentials of factor analysis. A&C Black.
Coleman, J., & Hagell, A. (2007). The nature of risk and resilience in adolescence. Adolescence, risk and resilience: Against the odds, 3, 2-16.
Collado, A., MacPherson, L., Kurdziel, G., Rosenberg, L. A., & Lejuez, C. W. (2014). The relationship between puberty and risk taking in the real world and in the laboratory. Personality and individual differences, 68, 143-148. Crosetto, P., & Filippin, A. (2016). A theoretical and experimental appraisal of four risk elicitation
methods. Experimental Economics, 19(3), 613-641.
Dahl, R. E. (2004). Adolescent brain development: a period of vulnerabilities and opportunities. Keynote address. Annals of the New York Academy of Sciences, 1021(1), 1-22.
Doukianou, S., Daylamani-Zad, D., Lameras, P., & Dunwell, I. (2019, July). Reinforcing rational decision making in a risk elicitation task through visual reasoning. In 41st Annual Conference of the Cognitive Science Society (pp. 1662-1668).
van Duijvenvoorde, A. C., Peters, S., Braams, B. R., & Crone, E. A. (2016). What motivates adolescents? Neural responses to rewards and their influence on adolescents’ risk taking, learning, and cognitive control. Neuroscience & Biobehavioral Reviews, 70, 135-147.
Eisenberg, D., Golberstein, E., & Whitlock, J. L. (2014). Peer effects on risky behaviors: New evidence from college roommate assignments. Journal of health economics, 33, 126-138.
Ernst, M., Pine, D. S., & Hardin, M. (2006). Triadic model of the neurobiology of motivated behavior in adolescence. Psychological medicine, 36(3), 299-312.
Ewalds-Kvist, B., Högberg, T., & Lützén, K. (2013). Impact of gender and age on attitudes towards mental illness in Sweden. Nordic journal of psychiatry, 67(5), 360-368.
Furby, L., & Beyth-Marom, R. (1992). Risk taking in adolescence: A decision-making perspective. Developmental review, 12(1), 1-44.
Gardner, M., & Steinberg, L. (2005). Peer influence on risk taking, risk preference, and risky decision making in adolescence and adulthood: an experimental study. Developmental psychology, 41(4), 625.
Ginott, H. G. (1971). Between parent & teenager. (3rd ed.). Location: Macmillan, 1969.
Harms, L., & Kansen, M. (2018). Cycling Facts. Netherlands Institute for Transport Policy Analysis (KiM). Den Haag: Ministry of Infrastructure and Water Management.
Holt, C. A., & Laury, S. K. (2002). Risk aversion and incentive effects. American economic review, 92(5), 1644-1655. Hooper, D., Coughlan, J., & Mullen, M. R. (2008). Structural equation modelling: Guidelines for determining model
fit. Electronic journal of business research methods, 6(1), 53-60.
Van Hoorn, J., Van Dijk, E., Güroğlu, B., & Crone, E. A. (2016). Neural correlates of prosocial peer influence on public goods game donations during adolescence. Social cognitive and affective neuroscience, 11(6), 923-933.
van Hoorn, J., van Dijk, E., Meuwese, R., Rieffe, C., & Crone, E. A. (2016). Peer influence on prosocial behavior in adolescence. Journal of Research on Adolescence, 26(1), 90-100.
Irwin Jr, C. E., & Millstein, S. G. (2014). Risktaking behaviors and biopsychosocial development during adolescence. Emot Cogn Heal Dev Child Adolesc (PLE Emot).
Jackson, D. L., Gillaspy Jr, J. A., & Purc-Stephenson, R. (2009). Reporting practices in confirmatory factor analysis: An overview and some recommendations. Psychological methods, 14(1), 6.
Jamieson, J. P., & Mendes, W. B. (2016). Social stress facilitates risk in youths. Journal of experimental psychology: general, 145(4), 467.
Kiat, J., Straley, E., & Cheadle, J. E. (2016). Escalating risk and the moderating effect of resistance to peer influence on the P200 and feedback-related negativity. Social cognitive and affective neuroscience, 11(3), 377-386. Klauer, S. G., Ehsani, J. P., McGehee, D. V., & Manser, M. (2015). The effect of secondary task engagement on
adolescents' driving performance and crash risk. Journal of Adolescent Health, 57(1), S36-S43.
Knoll, L. J., Magis-Weinberg, L., Speekenbrink, M., & Blakemore, S. J. (2015). Social influence on risk perception during adolescence. Psychological science, 26(5), 583-592.
Luciana, M., Collins, P. F., Olson, E. A., & Schissel, A. M. (2009). Tower of London performance in healthy adolescents: The development of planning skills and associations with self-reported inattention and impulsivity. Developmental Neuropsychology, 34(4), 461-475.
Luna, B., Garver, K. E., Urban, T. A., Lazar, N. A., & Sweeney, J. A. (2004). Maturation of cognitive processes from late childhood to adulthood. Child development, 75(5), 1357-1372.
McCormick, E. M., & Telzer, E. H. (2017). Failure to retreat: Blunted sensitivity to negative feedback supports risky behavior in adolescents. NeuroImage, 147, 381-389.
McCoy, S. S., Dimler, L. M., Samuels, D. V., & Natsuaki, M. N. (2019). Adolescent susceptibility to deviant peer pressure: Does gender matter?. Adolescent research review, 4(1), 59-71.),
Millstein, S. G., & Halpern–Felsher, B. L. (2002). Judgments about risk and perceived invulnerability in adolescents and young adults. Journal of research on adolescence, 12(4), 399-422.
O'Brien, L., Albert, D., Chein, J., & Steinberg, L. (2011). Adolescents prefer more immediate rewards when in the presence of their peers. Journal of Research on adolescence, 21(4), 747-753.
Ouimet, M. C., Pradhan, A. K., Brooks-Russell, A., Ehsani, J. P., Berbiche, D., & Simons-Morton, B. G. (2015). Young drivers and their passengers: a systematic review of epidemiological studies on crash risk. Journal of Adolescent Health, 57(1), S24-S35.
Peake, S. J., Dishion, T. J., Stormshak, E. A., Moore, W. E., & Pfeifer, J. H. (2013). Risk-taking and social exclusion in adolescence: Neural mechanisms underlying peer influences on decision-making. NeuroImage, 82, 23-34. Pfeifer, J. H., Masten, C. L., Moore III, W. E., Oswald, T. M., Mazziotta, J. C., Iacoboni, M., & Dapretto, M. (2011).
Entering adolescence: resistance to peer influence, risky behavior, and neural changes in emotion reactivity. Neuron, 69(5), 1029-1036.
Prencipe, A., Kesek, A., Cohen, J., Lamm, C., Lewis, M. D., & Zelazo, P. D. (2011). Development of hot and cool executive function during the transition to adolescence. Journal of experimental child psychology, 108(3), 621-637.
Reniers, R. L., Beavan, A., Keogan, L., Furneaux, A., Mayhew, S., & Wood, S. J. (2017). Is it all in the reward? Peers influence risk-taking behaviour in young adulthood. British Journal of Psychology, 108(2), 276-295.
Reyna, V., & Farley, F. (2006). Risk and Rationality in Adolescent Decision Making: Implications for Theory, Practice, and Public Policy. Psychological Science in the Public Interest, 7(1), 1–44. https://doi.org/10.1111/j.1529-1006.2006.00026.x
Rousseeuw, P. J., & Leroy, A. M. (2005). Robust regression and outlier detection (Vol. 589). John wiley & sons. Shope, J. T., Waller, P. F., Raghunathan, T. E., & Patil, S. M. (2001). Adolescent antecedents of high-risk driving
behavior into young adulthood: substance use and parental influences. Accident Analysis & Prevention, 33(5), 649-658.
Simons-Morton, B., & Chen, R. S. (2006). Over time relationships between early adolescent and peer substance use. Addictive behaviors, 31(7), 1211-1223.
Smith, A., Chein, J., Steinberg, L. (2011, November). Developmental differences in reward processing in the presence of peers. Paper presented at the 42nd annual meeting of the Society for Neuroscience, Washington, DC.
Steinberg, L. (2004). Risk taking in adolescence: what changes, and why?. Annals of the New York Academy of Sciences, 1021(1), 51-58.
Steinberg, L. (2008). A social neuroscience perspective on adolescent risk-taking. Developmental review, 28(1), 78-106. Steinberg, L., & Monahan, K. C. (2007). Age differences in resistance to peer influence. Developmental psychology,
43(6), 1531.
Spear, B. A. (2002). Adolescent growth and development. Journal of the Academy of Nutrition and Dietetics, S23. Sumter, S. R., Bokhorst, C. L., Steinberg, L., & Westenberg, P. M. (2009). The developmental pattern of resistance to
peer influence in adolescence: Will the teenager ever be able to resist?. Journal of adolescence, 32(4), 1009-1021.
Telzer, E. H., Fuligni, A. J., Lieberman, M. D., Miernicki, M. E., & Galván, A. (2015). The quality of adolescents’ peer relationships modulates neural sensitivity to risk taking. Social cognitive and affective neuroscience, 10(3), 389-398.
Szrek, H., Chao, L. W., Ramlagan, S., & Peltzer, K. (2012). Predicting (un) healthy behavior: A comparison of risk-taking propensity measures. Judgment and decision making, 7(6), 716.
Weitzman, E. R., Nelson, T. F., & Wechsler, H. (2003). Taking up binge drinking in college: The influences of person, social group, and environment. Journal of adolescent health, 32(1), 26-35.
