The Association between Changes in Peer Pressure and Changes in Alcohol and Cannabis Use: A Prospective Cohort Study among Men during Emerging Adulthood

Alcohol and other drug use by young people typically occur in social environments [1]. One feature is the impact of peers, usually conceptualised as influence and selection processes [2]. Peer influence describes the process through which individuals are influenced by their peers to use or not to use a substance. Peer selection describes the process through which individuals seek peers with similar behaviours. In the literature, the term peer pressure has often been used as an umbrella term, but described very different processes of peer influence or even peer selection [3]. One is modelling [4‒6], which is the imitation of another’s behaviour. A second influence comes through perceived drinking norms, which encourage people to use substances, e.g., to gain social acceptance and status [2, 7, 8] within a peer group. Social norms can be descriptive (how many peers use substances), or injunctive, reflecting the approval of peers [3, 9‒11]. Modelling and social norms are seen as indirect (passive) influences, which may include self-selection into a peer group. Growing up, the influence of parents become smaller, and young people have a strong need to affiliate with peers [5] to develop their identities. The need to “fit-in” or be accepted by one’s peers may encourage young people to behave in a group-like manner, to be liked and respected [5, 12, 13]. Peer influence is more a passive than an active thing [4, 14]. Modelling of peers may be more important than overt pressure [15]. The literature on direct (active) influence, such as offering a drink or buying rounds, is rare [3], and measures of it are rarely overt [2], such as directly forcing others to drink in drinking games.

Although conformity with peers has a number of beneficial effects on the development of identity and physical and mental well-being [15], there is little research on the positive aspects of peer influence. Peer pressure has a negative connotation of persuasion [16], whereas peer influences have a more positive connotation, also referring to the transmission of skills and competencies. As argued by some authors, the view of being peer-pressured is when it contravenes public health ideals (e.g., the use of substances). “Pressure” is not used when it comes to influence beneficial behaviour. However, young people do not feel being pressured to use a substance, and do not see its use as a moral wrong [5, 14, 17, 18] and risk-taking is rooted in individual and rational assessments [19].

Studies on beneficial influences are rare. Most work on potential beneficial effects of peer influence around substance use and substance-use outcomes has been cross-sectional. For example, Studer et al. [20] showed that peer pressure indexed through the misconduct subscale (substance use, unsafe sex, delinquent behaviours) of the Peer Pressure Inventory (PPI, [21]) was associated with increased drinking volume and frequency of heavy episodic drinking (HED) in a sample of young Swiss men. The inverse relationship was found for the peer conformity (e.g., conformity to dressing, hairstyles, music tastes) and peer involvement subscales (involvement in social activities, spending time with friends, attending parties, and social events). Conformity and compatibility with peers have been shown to have benefits for interpersonal, physical, and mental well-being [15, 16]. Similar beneficial influences have been reported in cross-sectional studies of adolescents [22], and adults [23‒25]. However, these cross-sectional studies cannot be used to determine whether peer influence affects subsequent substance use nor can they be used to distinguish between peer influence and peer selection.

Previous longitudinal studies have found that peer influences exert effects on substance use [2]. However, most of these studies have involved indirect measures characterised as positive answers to questions such as approval of use by peers or the number or percentage of friends who participants believed use the substance [2, 26]. Further, these studies do not report on influence to consume less among those who have already initiated substance use, which has been shown to be associated with reduced consumption, at least in the domain of alcohol consumption [24] and cannabis [27].

As regards alcohol use, studies mostly showed that both peer influence and peer selection operate bidirectional (reciprocally) [2]. However, the evidence is less clear for other substances such as cannabis but seems to point towards selection effects [7, 28]. Often, in studies on peer influence and substance use, composite measures of different substances have been used, resulting in mixed findings. As argued by Becker et al., [28], this calls for studies that examine the unique effects of peer processes on different substances. In one review of eight longitudinal studies using network analysis [6], friends’ influence on their own cannabis use was commonly found. However, the five studies controlling for peer selection effects found that adolescents also select friends similar to their own cannabis use behaviour. With the increasing legalization of cannabis around the world, the relative impact of peer influence and peer selection may change. At the time of the present study, there were relatively strict regulations for medicinal cannabis prescription in Switzerland and recreational cannabis use was illegal. In fact, only since 2023 have their first pilot trials in some cities and with a limited number of participants have started.

The majority of studies on peer influence come from the USA and have typically involved adolescents in schools or college and university students [12]. Therefore, there is uncertainty in generalising to other cultural backgrounds with variations in use prevalences, social norms, and policies [2]. For example, the legal drinking age in Switzerland is 16 years and not 21 years as in the USA. In the USA, college and universities, the time when students leave their home with decreasing [11], though not vanishing [10], influence from parents has been viewed as the place to drink excessively, particularly in Greek houses with strong descriptive and injunctive norms [12]. Closed systems such as schools may particularly foster peer pressure [2], and may be affected by the substance use (or not use) of the most popular students with high social standing [10, 29]. Studies in colleges and universities in the USA are also biased towards privileged populations.

Peer influence is seen to be strongest in early adolescents with increased individuation [6, 11, 13], when identities are strongest in a state of flux and adolescents want to alter their behaviour. Resistance to peer influence increases over late adolescents (14–18 years; 13), but peer influence persists into early adulthood [13], probably the period with the most new developmental task, such as leaving the home, starting work or studies, and getting married and having children. It is also a period where the influence of peers may be weekend and complemented by influence of intimate partner and their friends [16]. Studies of peer influences should therefore go beyond adolescence. As argued by Morris et al. [12], peer pressure is experienced across the lifespan and more studies among adult populations covering the full socioeconomic spectrum are needed.

Studies using more advanced statistical models such as latent growth models commonly show bidirectional associations, pointing to both peer influence and peer selection as important factors related to both alcohol and cannabis use [30‒32]. However, the mechanisms are yet fully explained. Leung and colleagues [2] argued with their peer process model that early risk factors may lead to early (deviant) peer selection, and the peer influences than take a snowball effect. However, low-risk adolescents may first imitate most popular peers [29] and, with increasing alcohol use, may select peers with similar behaviours. The direction of influence of two parallel processes is of importance in prevention and treatment [33, 34]. Reducing substance use but not the peers may be an indicator of future relapse and recurrence. Peer resistance programs may be useful if the direction of change only goes from peer influence to substance use [7]; however, bidirectional processes of influence and selection may need intervention programs going beyond peer resistance training.

In the present study, we examined whether changes in influence are related to longitudinal changes in the use of cannabis and alcohol. We look at the strength of peer influence and peer selection effects in a representative sample of young men. The second aim was to investigate whether peer influence acts on both sides that is, to use substances more or less often. Overall, we expect that increased influence to consume substances would be associated with increases in substance use over time and, conversely, that influence to not consume substances would be associated with reduced use over time.

We used data drawn from the baseline and first follow-up waves of the Cohort Study on Substance Use Risk Factors (C-SURF); a longitudinal study designed to examine substance use patterns and associated factors among young Swiss men (see [35] for an overview).

Baseline recruitment occurred in three of Switzerland’s six military recruitment centres, located in Lausanne (French-speaking: 57.4% of the final sample), Windisch and Mels (German-speaking: 42.6%), during the military recruitment procedures which are mandatory for all Swiss of biological male sex, irrespective of their gender identification. A total of 13,237 young men were invited to enrol between August 2010 and November 2011, of whom 7,556 (57.1%) gave written consent to be contacted for participation. To use maximum information with regards to different analyses, four different sample sizes were applied in the present paper (see Fig. 1). First, 5,987 (79.2%) completed the baseline questionnaire (wave 1; w1) between September 2010 and March 2013. There were only minor differences between non-consenters and consenters [36] and between participants and silent refusers, that is, people who consented but did not participate [37]. Of these 5,987, 5,479 (91.5%) completed the first follow-up questionnaires (w2) between March 2013 and January 2014, at an average of 15.8 months after the baseline questionnaire. Second, of those 5,479 individuals, 326 participants with missing values on any of the variables of interest were excluded from models needing complete case analysis (e.g., GEE models of pressure towards and away, see below), resulting in a final sample size of 5,153 for the complete case analysis.

In the second wave, in addition to the 5,479 first wave completers, 541 individuals with consent, who did not respond at baseline, responded at follow-up, resulting in a follow-up sample size of 6,020 individuals (5,479 + 541). As some analysis (cross-domain coupling model, see below) used full information maximum likelihood (FIML) estimation, cases with missing values had not be excluded. For this analysis (w1 or w2 observations), the sample size was 6,528 (5,987 + 541).

Alcohol use was measured as alcohol volume and frequency of HED. The self-reported number of standard drinks per week was derived from an extended quantity–frequency questionnaire measuring frequency and quantity separately on workweek days and weekend days to yield volume of drinking as drinks per week [38]. HED was measured using the third question from the Alcohol Use Disorder Identification Test [39], asking about how often participants consumed six or more standard drinks on one occasion. Images of standard drinks (e.g., a glass of beer or wine), corresponding to about 10 g of pure alcohol, were included in the questionnaire for reference. The quantity of the HED-measure corresponded to the common US-measure of 5+-drinks with approx. 12 g per standard drink. As a continuous measure, days per week of HED were used. Cannabis use was measured with one question asking how often during the past 12 months participants have used cannabis. The response options "once a month or less", "2–4 times a month", "2–3 times a week", "4–5 times a week", and "every day or nearly every day" were converted into days per week.

At both measurement points, the three items of the misconduct subscale of the Peer Pressure Inventory (PPI [40]) that were specific to alcohol and cannabis were used. As the wording in the PPI is using the term “pressure,” we used peer pressure hereafter to describe our findings. There was one item for cannabis (peer pressure to use or not cannabis) and two for alcohol, namely (a) pressure to drink or not and (b) pressure to get drunk or not.

For each item, participants evaluated how strongly they perceived pressure from their peers on a seven-point Likert scale, ranging from −3 (“a lot of pressure not to do”) to 3 (“a lot of pressure to do”), with 0 for “no pressure.” For alcohol, the mean across the two items was taken as the alcohol score for each participant.

Percentages of the underlying rating scales (peer pressure, alcohol, and cannabis use) as well as the scale means and standard deviations were given. Analyses were conducted using SPSS 25.

The main analysis used linear latent change score models, more precisely, the cross-domain coupling model [41] with FIML estimation. The advantage of FIML is that it estimates a likelihood function for each individual based on all available data. It therefore does not exclude individuals with missing values on some variables. It has equivalent properties like multiple imputations for missing data [42]. Missing values were excluded from descriptive analyses, and the corresponding sample sizes are indicated. There were generally few missing values. Latent change score models are similar to cross-lagged models with observed variables but have several advantages over models with observed variables (e.g., reducing measurement error in latent variables).

Models were estimated separately for volume of alcohol, HED, and cannabis use. The basic model is shown in Figure 2. The ∆s represent the latent change in scores in substance use and peer pressure. The coupling parameters (βs) can be interpreted like cross-lagged coefficients, indicating whether peer pressure at baseline predicts change in alcohol use or cannabis use (influence), or whether alcohol use at baseline predicts changes in peer pressure (selection). The γs are the paths from baseline measure to the corresponding change. These paths adjust for regression to the mean, which commonly occurs because individuals with high baseline measures tend to have lower follow-up measures on the same construct, and vice versa. Regression to the mean results in a negative association between the initial status and the change [43]. Finally, τ is the correlation between the latent change scores adjusted for coupling pathways (cross-lagged paths) and regression to the mean. τ therefore reflects the adjusted degree of co-occurrence of changes in peer pressure and substance use. We used bias-corrected bootstrap estimates, which take deviations from normality (as is the case for, e.g., alcohol use measures, which are non-symmetric around the mean) into account in confidence intervals (uncertainty intervals) estimation. Bias corrected 95% confidence intervals (CI) were presented.

A second set of models aimed at investigating whether increases in pressure were associated with increases in substance use in the same way as decreases in pressure with decreases in substance use. At baseline and follow-up, we trichotomized peer pressure into pressure away from use (i.e., negative values on the pressure scale), no pressure (i.e., 0 on the peer pressure scales) and towards use (i.e., positive values on the peer pressure scale). Changes in peer pressure between survey waves could therefore take the following forms: (a) no change, (b) increased pressure to use (from peer pressure away from use to no peer pressure or peer pressure towards use, or from no peer pressure to peer pressure towards use), or (c) decreased pressure to use (from peer pressure towards use to no peer pressure or peer pressure away from use, or from no peer pressure to peer pressure away from use). It should be noted that in this analysis, there are “structural zeros” as regards change. Having experienced pressure away at baseline, trichotomized peer pressure can only stay the same (pressure away) or increase (no pressure or pressure towards use at follow-up), but not decrease. Similarly, experiencing pressure towards use at baseline, pressure at follow-up can only stay the same or decrease (no pressure or pressure away), but cannot increase. These structural zeros necessitated stratified analysis according to the pressure status at baseline. Generalised estimating equations (GEE) were used. Outcomes were treated either as ordinal (HED frequency, cannabis use frequency), or count (negative binominal for alcohol drinks per week). Variables included in GEE analyses were time (within subjects, i.e., baseline coded = 0 and follow-up coded = 1), group of change (decrease, increase, no change) and their interaction with adjustment for baseline linguistic region and age (centred) as a time-varying covariate. Time indicates the change in the reference group (coded 0), and group of change (increase, decrease) indicates the difference at baseline compared with the reference group (no change). The interaction coefficient indicates the change in the corresponding group compared with the reference group (no change).

The mean age of men was 20 years at baseline. Over 90% reported drinking alcohol at both time points (Table 1). Around one third reported cannabis use in the previous 12 months, and more than 40% reported HED at least monthly at both time points. Individuals experienced more pressure to drink and to get drunk than not to drink or not to get drunk. However, pressure to use cannabis was more often away from cannabis. For example, at baseline, 16.9% were pressured to use cannabis, but 22.1% were pressured not to use cannabis.

All coefficients of the full coupling model were significant at the 5%-level (Table 2). γs were all negative, representing the correlation between initial status and change and pointing to strong regression to the mean effects. τs are the correlations between the change in peer pressure and the change in substance use. The correlation was τ = 0.151 for volume of drinking (95% CI: 0.120, 0.180), τ = 0.130 for HED (95% CI: 0.103, 0.158), and τ = 0.130 for cannabis use (95% CI: 0.106, 0.154).

The coupling parameters showed bidirectional associations of similar magnitude for alcohol, i.e., β = 0.090 for the path from volume drinking to peer pressure and β = 0.089 for the path from peer pressure to volume drinking. The corresponding coefficients for HED were β = 0.062 and β = 0.071. Although both coefficients were significant, for cannabis use the coupling parameter from cannabis use to peer pressure (representing peer selection) was around 5-times larger (β = 0.141, 95% CI: 0.121, 0.161) than the path from peer pressure to cannabis use (β = 0.29, 95% CI: 0.004, 0.054).

Regarding pressure towards and away from substance use (Table 3), regression coefficients went in the expected direction, indicatiog that if peer pressure increased, alcohol use (volume and HED) and frequency of cannabis use also increased. Similarly, negative coefficients indicated that, if peer pressure decreased, alcohol and cannabis use also decreased. Although the signs of coefficients were, with one exception, in the expected direction, not all changes were significant at the conventional level (p < 0.05). The notable exception in the direction of effects was decreased peer pressure from no pressure to pressure away for volume of drinking (b = 0.061; p = 0.556). The effects on group membership (peer pressure increase, peer pressure decrease) indicated that those experiencing an increase in peer pressure had higher alcohol use at baseline, whereas those for which peer pressure decreased had lower consumption at baseline.

Our study showed that peer pressure to use substances is not only relevant in adolescents or closed school systems with more affluent college students, but still operates in young adults in a socioeconomically diverse sample in Switzerland. Although longitudinal studies on alcohol use have mostly shown bidirectional associations of peer pressure and peer selection, it has been hypothesized that peer pressure is stronger in early to mid-adolescence, and is increasingly dominated by peer selection when alcohol use has been initiated [2, 44, 45]. The cross-domain coupling model indicated that, in relation to alcohol use, both peer selection and peer pressure operated with similar strength at the beginning of emerging adulthood, i.e., at times when alcohol initiation has been mostly completed. At baseline, more than 95% of our sample of young men (around 20 years of age) were already consuming alcohol in the past 12 months. Findings therefore suggest that not only peer selection, but also peer influence may be experienced across the lifespan [12] and particularly in emerging adulthood [13].

Literature on peer influences on cannabis use is much scarcer and has produced less consistent results as regards the pre-dominance of selection versus pressure effects [28]. The cross-domain coupling model indicated an almost 5-time larger peer selection than peer pressure effect, which is in accordance with Becker and Curry’s [46] findings and findings in the social network literature [6]. This finding indicates that cannabis is still culturally less accepted, resulting in less peer influence to use it. In Switzerland at the time of the study, recreational cannabis use was still illegal (first pilot trials with recreational cannabis sales only started in 2023) and may be associated with a higher risk profile in general. Contrary to alcohol, pressure for cannabis was more frequently pressure to NOT use it than to use it (Table 1). As argued by Leung et al. [2], the effect of peer selection is more profound among young people with higher risk profiles.

Very few studies have looked at putatively positive peer pressure [27], that is, pressure NOT to use a substance. Studer et al. [20] have shown using the PPI [40] that, in contrast to peer misconduct, there may be putatively positive effects of peer involvement and peer conformity. In the present study we used only items from the misconduct scale. Further analysis of peer conformity and peer involvement may be fruitful to identify their importance for the development of identity and the securing of physical and mental well-being [15]. This was, however, beyond the aims of the present study. Our study indicated that peer pressure can go in both directions, increasing and decreasing subsequent substance use. If peer pressure decreased from pressure towards substance use to no pressure or even pressure away from substance use, substance use decreased. Thus, the longitudinal link between pressure and substance use does not only hold for pressure to use substances but may also be used for preventive purposes to decrease substance use.

Our findings suggest that peer resistance training alone may have limited impact on preventing substance use [7]. There is increasing literature questioning the focus on interventions portraying peer pressure only as pressure to use substances [14, 17, 18]. Such interventions are “adultist constructions” which replace pressure by peers through pressure of interventionists with public health ideals and may not be meaningful to participants of such interventions. They may even impede the development of autonomy and lead to iatrogenic effects such as deviancy training [47, 48]. Promoting better peer selection but also interventions that strengthen individualisation [30] and focus on self-regulatory strategies and decision-making [45] and the transmission of skills and competencies [7, 16] may be beneficial.

One of the limitations of the present study is that only men were included in the sample. This was related to the enroling of participants, which was linked to Swiss army conscription procedures. Women can join the army on a voluntary basis, but this would not be a representative sample of young women. The advantage is that this process is mandatory for all Swiss with a biological male sex, and therefore, there are no sample selection biases around, for example, socio-demographic background or gender identity. It is important to note, however, that the army conscription process was only used to enrol participants. The study was conducted completely outside the army and included participants independent of which branch of national service they subsequently engaged in (i.e., the army, social service, were exempted, or refused any service). Nevertheless, men and women will likely differ in their peer processes. Men drink more than women and are more susceptible to overt alcohol offers and competitive substance use [3, 12, 16] and peer selection may be stronger for women [45]. Another limitation is that we could not measure the social network of participants. Hence, we do not know whether bidirectional associations were due to maintaining a stable peer group or changes were due to changing peers [2, 30, 49]. In addition, peer processes may have started before the baseline measurement, and thus, earlier influence processes may look like selection effects and vice versa. Although the PPI wording is about pressure, we do not know whether the participants feel overt pressure, or feel pressured through, e.g., norms or attitudes in their peer group. Finally, although we study effects at the beginning of emerging adulthood, the time between the two waves was only about one and a half years. Studies such as of Windle and Windle [45], for example, covered a much longer period from 17 to 33.5 years of age. More studies covering different age frames are needed to study peer pressure as a life-long process and to identify at which stage in life peer selection or peer pressure operates more profoundly to identify age-appropriate prevention and intervention strategies.

According to our findings, peer selection and peer pressure do not only operate in late childhood or early adolescence but continue to impact on substance use at least in emerging adulthood. First, processes seem to work in both directions; increases and decreases in peer pressure were related to increases and decreases in substance use. Second, what impacts the most, selection or pressure, seems to be dependent on the substance used, a culturally long-time accepted substance with early initiation such as alcohol in Switzerland, or – despite increasing acceptance – a substance which is still illegal, such as cannabis, which may indicate a higher risk profile associated with stronger peer selection effects [2]. Third, peer pressure can have putatively positive effects on substance use if the influence goes in the direction towards non-use of a substance.

This study was reviewed and approved by the Human Research Ethics Committee of the Canton of Vaud (Protocol No. 15/07), Approval No. 15/07. All participants signed written informed consent to participate in the study.

The authors have no conflicts of interest to declare.

The Burnet Institute gratefully acknowledges the funding provided under the Victorian Research Operating Infrastructure Fund. The C-SURF study was funded by the Swiss National Science Foundation (FN 33CSC0-122679, FN 33CS30_139467). The funder had no role in the design, data collection, data analysis, and reporting of this study.

G.G. made substantial contribution in the acquisition and data collection of the cohort study as a whole and the analysis, interpretation, and writing up of the present article. S.M. made substantial contributions to the analysis, interpretation, and write up of the present study. J.S. and M.W. made substantial distribution to designing the analysis and interpretation of results. P.D. developed the idea of the present study and made substantial contribution to interpretation of data and writing up of the study. All five authors additionally reviewed critically the study for intellectual content, approved the final version, and are accountable for all aspects of the presented work.

All data, questionnaires and selected syntaxes are publicly available in the repository “Zenodo” (Gmel et al., 2021 Gmel, G., Mohler-Kuo, M., Studer, J., Gachoud, C., Marmet, S., Baggio, S., & Foster, S. (2021); Cohort Study on Substance Use Risk Factors (C-SURF) [Data set]. Zenodohttps://doi.org/10.5281/zenodo.5469953) and can be accessed after verification of potential users and their aims to avoid double publications and access of the industry.