The relationship between age at first drink and later risk behaviours during a period of youth drinking decline

Abstract

Aims

Alcohol consumption among teenagers in many high‐income countries has steadily declined since the early 2000s. There has also been a steady increase in the average age at first drink, a reliable marker of later alcohol problems. This study measured whether young people who initiated drinking early were at increased risk of alcohol problems in young adulthood in recent cohorts where early initiation was comparatively rare.

Design

Analysis of six waves of a repeated cross‐sectional household survey (2001–16).

Setting

Australia.

Participants

A total of 9576 young adults (aged 20–25 years) who had initiated drinking before the age of 20 years.

Measurements

Respondents were classified into three groups based on their self‐reported age at first drink (< 16, 16–17, 18–19 years). Outcome variables were self‐reported experiences of memory loss while drinking, risky and delinquent behaviour while drinking and monthly or more frequent drinking occasions of 11 or more 10‐g standard drinks.

Findings

Later initiators reported lower levels of all outcomes [e.g. odds ratios (ORs) for memory loss were 0.56, 95% confidence interval (CI) = 0.50, 0.63] for those who first drank at 16 or 17 years compared with those who first drank at age 15 or younger). Significant interactions between age at first drink and survey year showed that early initiation was more strongly associated with harms (e.g. for memory loss, OR = 0.97, 95% CI = 0.94, 0.99 for 18–19 versus 15 or younger) in young adulthood for recent cohorts where early drinking was less common.

Conclusions

The decline in youth drinking may have contributed to a concentration of risk of alcohol problems among those young people who consume alcohol in early adolescence. Early initiation of drinking may be an increasingly important marker of broader risk taking as alcohol becomes less normative for teenagers.

INTRODUCTION

Alcohol consumption among adolescents has long been a subject of public health concern, both because of the immediate risks [1] and because it is associated with a range of later negative outcomes [2]. Thus, in Australia, for example, national guidelines suggest that people under the age of 18 years should not drink alcohol [3] to prevent both immediate and longer‐term harms. However, alcohol consumption has historically represented a key behaviour that young people engage in on their pathway to adulthood [4] and, in high‐income countries, alcohol consumption during adolescence has been relatively normative. For example, in the 1999 European School Survey Project on Alcohol and Drugs, more than 70% of 15–16‐year‐olds reported past‐year drinking in 29 of the 30 European countries surveyed [5]. Similarly, past‐month alcohol consumption among Australian school students was reported by 70% of 16–17‐year‐olds in the late 1990s [6].

Despite the relative ubiquity of late‐adolescent drinking in high‐income countries, researchers have repeatedly emphasized the need to consider variation in the development of drinking during this stage of life. In a ground‐breaking early study, Jessor & Jessor [4] showed that earlier initiation of drinking was associated with a range of attitudinal and personality factors and that young people who initiated early were more likely to engage in other risk behaviours that were generally linked with the transition to adulthood (e.g. sex and illicit drug use). Since this early work, a large volume of research has focused upon age of initiation of drinking, linking age at first drink with a range of later negative outcomes [7]. Thus, for example, early initiation into drinking has been linked with earlier initiation of binge drinking [8], with subsequent diagnosis of alcohol use disorder [9, 10], with unintentional injuries [11] and suicide attempts [12] and with a range of other alcohol‐related problems [13]. While these associations are strong and consistent, there are good arguments that these relationships are unlikely to be causal, with early initiation to drinking instead reflecting the clustering of a range of personality, family and social risk factors that have common links with most of the outcomes studied [7, 14]. Thus, while interventions focusing specifically on delaying initiation into drinking may not necessarily be effective at reducing later harms, young people who initiate drinking early are clearly a risk group who should be the focus of preventative interventions to reduce these risks.

In recent years, adolescent drinking has declined sharply in many high‐income countries [15, 16], and there is growing research exploring its underlying causes and longer‐term implications [17, 18]. In Australia, for example, the prevalence of life‐time abstention from drinking among 14–17‐year‐olds has more than doubled, from 28% in 2001 to 66% in 2019 [19]. Alongside these declines in drinking prevalence has been a clear shift towards later initiation into drinking [20, 21, 22]. In Australian survey data, the mean age of first drink for recent young drinkers increased from 14.7 to 16.2 between 2001 and 2019 [19]. These results are supported by findings in other surveys [23], and early analyses suggest that declines in adolescent drinking have been at least partly maintained into adulthood [24, 25]. These shifts probably mean substantial public health benefits, in terms of reduced harm in the short term and via longer‐term reductions in heavy drinking throughout the life‐course [22, 26].

The decline in drinking and, especially, in early initiation into drinking raises some important questions about the changing profiles of early drinkers. In Jessor & Jessor’s original work, initiation into drinking was a standard development during adolescence, which varied in timing but was broadly normative. In fact, they identified abstainers as a distinct group who stood out from drinkers on a range of characteristics [4]. In Australia and elsewhere, alcohol consumption is now a minority behaviour among late adolescents (e.g. in 2019, fewer than one‐third of Australian 16‐year‐olds reported past‐year drinking), which suggests that early initiators may be more ‘deviant’ than they were in previous generations, where it was a common experience.

While these declines in drinking are generally good news for public health, there is the potential that these declines have led to a ‘hardening’ among remaining young people who drink. This has been argued in tobacco control, where declines in regular smoking in many high‐income countries may have left a group of smokers less susceptible to intervention and at higher risk of harm [27, 28] (although the evidence remains contested [29]). In our context, the declines in youth drinking might mean that the young drinkers who remain have become a ‘riskier’ group who are more important, but potentially more challenging, with whom to intervene. Understanding whether or not this is happening has major implications for prevention and policy, but there has been relatively little work on the topic.

In the 1990s, when youth drinking was increasing, some researchers found that adolescent drinkers became less deviant as youth drinking became more prevalent. For example, Pape et al. [30] showed that heavy drinkers became less likely to engage in problem behaviours as drinking increased between 1992 and 2002. Similarly, in a long‐term study that looked across the whole population, Landberg & Hubner showed that alcohol consumption was less associated with alcohol problems in periods with higher per‐capita consumption [31]. More recently, following a steady declining in drinking by UK adolescents, Oldham et al. found little evidence that teenage drinking was becoming more heavily concentrated among low socio‐economic status (SES) or otherwise marginalized youth. There have also been some attempts to assess whether adolescent drinking has become riskier for those who engage in it as it has declined. For example, in an Australian examination of this question, Mojica‐Perez et al. [32] found no evidence that the link between risky behaviour and drinking among teenagers had strengthened during the period of declining drinking. Similarly, Thor et al. [33] found relatively stable associations between drinking and harms for adolescents as youth drinking declined in Sweden. Both these studies looked at outcomes concurrent with adolescent heavy drinking, but the potential impacts on longer‐term outcomes have not been assessed and the role of drinking early in adolescence not been emphasized.

Given that a key potential benefit of declining youth drinking are reductions in risky and harmful drinking behaviour in adulthood, it is important to assess whether the links between earlier adolescent drinking and later behaviours have shifted as adolescent drinking has decreased. In this study, we aim to assess whether this signal has become stronger as drinking has declined. Our key question is whether or not early initiation of drinking more strongly predicts later alcohol problems as adolescent drinking declines. We are not treating age at first drink as the causal mechanism relating to later harms, but as a relatively robust signal of a cluster of risk factors associated with a wide range of later problems.

METHODS

Data

We use data from the Australian National Drug Strategy Household Survey (NDSHS), which has been conducted every 3 years since 2001. The NDSHS is a representative household survey assessing alcohol and drug issues. Sampling and data collection modes have varied across time, but has generally involved stratified cluster samples, with in‐person recruitment and a ‘drop and collect’ survey form which is completed by the respondent in their own time and then collected by the data collection agency at a subsequent visit. New samples are drawn each wave. A range of modes have been incorporated within this framework over the years (in‐person, telephone, on‐line), but sampling mode differences are low on key measures [34] and NDSHS estimates of consumption have been shown to track objective measures of population consumption relatively well [35].

Data from 2019 were available, but key outcome measures were not included in the most recent questionnaire, so we limited our analyses to the six waves spanning 2001 to 2016. Given that we were interested in changes across recent cohorts and the well‐established recall biases involved in asking older respondents to estimate their age at first drink [36], we limited our analyses to respondents aged between 20 and 25 who reported past year drinking and had initiated drinking before the age of 20. Specific item‐level missing data are noted below. Cases with missing data were excluded from the relevant analyses. It is worth noting that the proportion of respondents aged 20–25 who abstain from alcohol (and are thus excluded) has increased over time, from 9.6% (8.1%, 11.2%) in 2001 up to 17.2% (14.7%, 20.0%) in 2016.

Measures

Age at first drink was assessed via a single item (‘About what age were you when you had your first full serve of alcohol?’). We classified respondents into three groups, based on approximately equal group sizes—early initiators (first full serve before the age of 16, 37% of full sample), middle initiators (first full serve aged 16 or 17, 37% of full sample) and later initiators (first full serve aged 18 or 19, 26%). We categorized this variable for ease of interpretation, but analyses using a continuous measure of age at first drink were also conducted and results are included in the Supporting information.

We assessed the link between age of initiation and four outcome measures, which we selected based on their consistent inclusion in the surveys between 2001 and 2016, and their reflection of problematic drinking or alcohol‐related behaviour in young adulthood:

• Regular very heavy drinking (missing = 525, 5%): A dichotomous measure assessing whether or not the respondent reported monthly or more frequent occasions of drinking involving 11 or more standard drinks (10 g each) in the past 12 months (based on their response to a set of standard graduated frequency items).

• Memory loss (missing = 456, 5%): A dichotomous measure based on a single item (‘In the last 12 months, about how often have you been unable to remember afterwards what happened while you were drinking?’), with respondents who responded ‘never’ classified as having no memory loss and those who reported ‘at least once’ or more frequently classified as having experienced memory loss.

• Risky behaviour score (missing = 280, 3%): A variable tallying how many of five risky behaviours (went to work; went swimming; operated a boat; drove a motor vehicle; operated hazardous machinery) the respondent had engaged in while affected by alcohol in the past 12 months. The distribution of responses to each item is provided in the Supporting information (Table S1).

• Delinquent behaviour score (missing = 280, 3%): A variable tallying how many of five delinquent behaviours (caused a public disturbance; caused damage to property; stole money, goods or property; verbally abused someone; physically abused someone) the respondent had engaged in while affected by alcohol in the past 12 months. The distribution of responses to each item is provided in the Supporting information (Table S2).

• These measures are not validated outcome scales but are consistently collected over a 15‐year period and capture excessive drinking (the first two) and problematic alcohol‐related behaviours (the last two). The last two variables have been used previously and shown to measure distinctly different kinds of behaviour [37]. All models were adjusted for sex and single year of age. Survey year was adjusted so that it ranged from 1 (for 2001) to 16 (2016) to ensure that interaction terms were easily interpretable.

Analysis

The associations between age of initiation and the two dichotomous outcome measures (regular very heavy drinking; memory loss) were analysed using binomial logistic regression. The two score measures (risk score, delinquent behaviour score) were treated as count data and comparisons of negative binomial and Poisson regression models based on the likelihood ratio test resulted in negative binomial models being used for these outcomes (the Pearson dispersion statistics in base models were between 1.3 and 1.7). The survey year term was modelled as a linear, continuous variable after initial comparisons between a variety of specifications (linear, quadratic, categorical) found that it was the most appropriate. In three of the four different outcomes, models with linear year terms had lower Bayesian information criteria (BIC) than the other specifications, indicating a better fit to the data. For the delinquent behaviour score, the quadratic specification had a slightly lower BIC, but we decided on the linear approach for consistency and to provide easily interpretable interaction terms. Simple interaction terms between survey year and age of initiation were used to assess whether the strength of any associations varied over the study period. A further three‐way interaction assessing whether there were differing relationships by sex was also estimated and its utility assessed via likelihood ratio tests comparing it with the simpler two‐way interaction. All analyses were conducted using Stata [38], with the ‘svy’ commands to adjust for the complex design of the NDSHS sample. Analyses were not pre‐registered and should thus be considered exploratory; 95% confidence intervals (CI) are provided in parentheses throughout and all tests of significance were at the 0.05 level.

RESULTS

The initial sample of respondents aged 14 years and over throughout the six waves was 151 341. Once respondents older than 25 and younger than 20 were excluded, we were left with 11 196 respondents. Those who had not drunk alcohol in the past 12 months (n = 1289) were excluded, as were those who initiated drinking at aged 20 or later (n = 331), leaving a final sample of 9576 (see Fig. 1).

FIGURE 1.

Sample flow diagram

There were some missing data on each of our outcome measures: regular very heavy drinking (525, 5%), memory loss (456, 5%), risky behaviour score (280, 3%) and delinquent behaviour score (280, 3%). A comparison of missing and complete cases is provided in Supporting information, Table S1. Generally, respondents missing on outcome measures were more likely to be female, to have been surveyed in more recent waves and to have initiated drinking at an older age.

Descriptive trends over time in the age of initiation into drinking are presented in Table 1. More than half of 20–25‐year‐old drinkers in 2001 reported initiating drinking before the age of 16, declining to one‐third for 20–25‐year‐old drinkers in 2016. The proportion of respondents reporting their first drink aged 18 or 19 nearly doubled over the same period.

TABLE 1.

Trends in age of initiation to drinking.

	n	Younger than 16 years	16–17 years	18–19 years
2001	2123	51.5% (48.7%, 54.4%)	33.5% (30.8%, 36.2%)	15.0% (13.1%, 17.1%)
2004	2156	49.6% (44.4%, 54.7%)	33.2% (30.5%, 35.9%)	17.3% (12.8%, 22.8%)
2007	1492	50.7% (47.5%, 54.0%)	35.8% (32.7%, 39.0%)	13.5% (11.4%, 15.8%)
2010	1540	46.5% (43.6%, 49.4%)	34.2% (31.6%, 37.0%)	19.3% (17.0%, 21.9%)
2013	1384	37.1% (34.1%, 40.2%)	39.1% (36.2%, 42.1%)	23.8% (21.2%, 26.7%)
2016	1229	33.0% (29.9%, 36.1%)	38.6% (35.4%, 41.9%)	28.4% (25.3%, 31.7%)

Trends in the four outcome measures are presented in Table 2. All measures were lower in 2016 than in 2007, although declines were more striking for early initiation and the two behaviour scores. Baseline models assessing the links between age of initiation and each of the four outcomes are presented in Table 3. Later initiation into drinking was significantly associated with lower levels of all outcomes, as was being female. Once age of initiation was adjusted for trends were largely non‐significant, although both risk and delinquent scores were still significantly lower in 2016 compared to 2001. Age effects between 20 and 25 were relatively minor—both memory loss and delinquent behaviour declined slightly with age.

TABLE 2.

Trends in the four study outcome measures.

	n	Regular very heavy drinking (%)	Memory loss (%)	Risky score (mean)	Delinquent score (mean)
2001	2123	29.5% (27.0%, 32.1%)	46.9% (44.3%, 49.5%)	0.57 (0.52, 0.62)	0.40 (0.35, 0.45)
2004	2156	29.8% (26.4%, 33.4%)	46.7% (44.3%, 49.2%)	0.52 (0.47, 0.57)	0.39 (0.35, 0.43)
2007	1492	35.2% (31.9%, 38.5%)	49.6% (46.0%, 53.1%)	0.51 (0.45, 0.57)	0.44 (0.37, 0.50)
2010	1540	31.3% (28.6%, 34.2%)	52.6% (49.6%, 55.7%)	0.49 (0.44, 0.53)	0.36 (0.31, 0.41)
2013	1384	28.2% (25.5%, 31.0%)	45.5% (42.3%, 48.7%)	0.45 (0.39, 0.51)	0.24 (0.19, 0.29)
2016	1229	28.1% (25.1%, 31.3%)	42.6% (39.2%, 46.1%)	0.38 (0.33, 0.43)	0.15 (0.12, 0.19)

TABLE 3.

Models assessing the relationship between age of initiation and adult drinking and risk behaviours.

	Regular very heavy drinking a	Memory loss a	Risk score b	Delinquent score b
Age of initiation (years)
< 16 (ref)	1.0	1.0	0	0
16–17	0.45 (0.39, 0.51)	0.57 (0.51, 0.63)	−0.64 (−0.73, −0.54)	−0.76 (−0.89, −0.63)
18–19	0.17 (0.13, 0.21)	0.23 (0.20, 0.27)	−1.49 (−1.66, −1.32)	−1.92 (−2.20, −1.65)
Sex
Male (ref)	1.0	1.0	1.0	1.0
Female	0.31 (0.28, 0.35)	0.76 (0.69, 0.84)	−0.55 (−0.63, −0.47)	−0.85 (−0.97, −0.72)
Year (2001 as reference)
2001 (ref)	1.0	1.0	0	0
2004	1.05 (0.85, 1.28)	1.01 (0.87, 1.18)	−0.07 (−0.20, 0.06)	0.03 (−0.10, 0.16)
2007	1.32 (1.08, 1.61)	1.09 (0.92, 1.31)	−0.12 (−0.26, 0.03)	0.10 (−0.09, 0.28)
2010	1.18 (0.98, 1.43)	1.36 (1.15, 1.60)	−0.10 (−0.23, 0.02)	−0.07 (−0.25, 0.10)
2013	1.10 (0.91, 1.34)	1.08 (0.91, 1.29)	−0.10 (−0.24, 0.05)	−0.35 (−0.57, −0.13)
2016	1.17 (0.95, 1.45)	1.03 (0.86, 1.24)	−0.22 (−0.37, −0.08)	−0.76 (−1.01, −0.50)
Age (continuous)	0.97 (0.94, 1.00)	0.95 (0.92, 0.98)	0.00 (−0.02, 0.02)	−0.10 (−0.14, −0.06)

^a Odds ratios;

^b coefficients.

There were clear associations between early initiation of drinking in adolescence and heavy drinking and problem behaviour in early adulthood. To assess whether these associations had changed during the study period, interactions between the age of initiation variable and year (as a continuous variable, to maximize study power) were added to the base models. Three‐way interactions also including sex were estimated, but did not provide significant improvement in model fit and are not presented. The key interaction parameters are presented in Table 4 and then visualized via predicted marginal probabilities in Fig. 2.

TABLE 4.

Interaction terms from models assessing whether the impact of drinking initiation age varies over time.

	Regular very heavy drinking a	Memory loss a	Risk score b	Delinquent score b
Age of initiation (< 16 years as reference) × year
16–17	0.99 (0.96, 1.01)	0.99 (0.97, 1.02)	−0.02 (−0.04, −0.01)	−0.03 (−0.06, −0.01)
18–19	0.97 (0.93, 1.02)	0.97 (0.94, 0.99)	−0.03 (−0.06, 0.01)	−0.05 (−0.10, −0.01)

^a Odds ratios;

^b coefficients.

FIGURE 2.

Predicted marginal probabilities from regression models, showing the changing relationships between early initiation of drinking and later behaviour

For all outcomes, the relative difference in risk between early and later initiators increased during the study period, although this was not statistically significant in every case. This is clearest in the memory loss model (Fig. 1), where the predicted probability of young adult memory loss due to drinking increased non‐significantly for those who first drank aged 16 or younger from 56.6% (54.0%, 59.2%) to 61.2% (57.6%, 64.8%), while for those initiating aged 18 or 19 it dropped non‐significantly from 27.8% (23.4%, 32.2%) to 22.9% (18.7%, 27.1%).

Results using a continuous measure of age at first drink are provided in Supporting information, Table S3 and Fig. S1a–d and support the results described above.

DISCUSSION

Our analyses demonstrate that the early initiators of drinking have become a higher risk category over time as adolescent drinking has declined. Specifically, young people who initiate drinking before the age of 16 have had a higher risk of later heavy drinking and alcohol problems than later initiators throughout the entire study period, but these differences increased among all four measures examined. These changes in risk profile were relatively small, and indeed for some measures (e.g. the delinquency score) rates fell among all categories, simply more steeply for later initiators. This, together with the broad trend towards later initiation found in the sample, suggests that harms overall will continue to decline due to the generally lower risk profiles of later initiators found in more recent cohorts of adolescents. However, our findings highlight the need for increased focus upon groups that are being left behind by recent declines in youth drinking, who are at potentially higher risk of subsequent problems. This is broadly consistent with the recent arguments put forward by Caluzzi et al. [39, 40] regarding the denormalization of adolescent drinking. They argue that adolescent drinkers are potentially at risk of marginalization and stigma as alcohol consumption becomes less normative. In our results, early initiators have shifted from making up more than half the population of young adult drinkers in the early 2000s to just one‐third in 2016. While our study cannot unpack the mechanisms behind the increasingly disparity in risk between early and later initiators, it suggests that early drinkers should be a focus of intervention and prevention to reduce harm.

Our results are somewhat counter to earlier work, both in Sweden [33] and Australia [32], both of which found no immediate increase in risky behaviours or harms for heavy drinkers as the rate of heavy drinking declined. The disparity with the earlier Australian study is particularly of note, given that similar data, time‐periods and broadly similar analytical approaches were applied. Our focus upon the longer‐term relationship between early drinking and problematic outcomes suggests that any increased risks associated with adolescent drinking become more marked as young people age into young adulthood, when drinking typically peaks.

In their work, Jessor & Jessor argued that early initiation of drinking ‘may indicate mere precocity of development’ ([4], p. 49). However, the consistent and strengthening associations between early drinking and later negative outcomes suggest that it may be better thought of as a marker of broad risk. Thus, we would argue that targeted interventions with teenagers who initiate drinking early are of potential benefit, although a focus on alcohol‐specific intervention may not be justified (not least given the relatively mixed evidence that these programmes are (in)effective at reducing adolescent alcohol consumption [41, 42, 43]). Instead, adolescents who initiate drinking early may benefit from supportive interventions beyond alcohol‐specific approaches. Interventions with high‐risk teenagers have a robust evidence base, with reviews highlighting the benefits of parent‐focused interventions [44, 45].

Our results should be considered in light of key limitations. As with all studies relying upon survey‐based data, we are using data from a sample with a relatively low response rate (~50% overall for each wave, likely to be lower for younger respondents), meaning that there is a significant likelihood of systematic under‐representation of key risk groups. Missing data were an issue for all our key outcome items, with missing respondents systematically different from those who provided data. This raises concerns about bias, especially as respondents with missing data were simply excluded from analyses, but the relatively low prevalence (between 3 and 5%) of missing data means any bias introduced should be relatively small. Self‐report data from sampled respondents are subject to a further array of potential biases, including recall and social desirability, that may result in misleading estimates of the links between consumption and harm. Age of initiation is particularly susceptible to recall bias [36], but we have tried to limit this effect by only including young adults, whose initiation into drinking was relatively recent. It is well established that survey data markedly underestimate actual consumption [46, 47], but NDSHS data have at least been shown to reproduce broad population trends in drinking [35]. Ideally, any biases will be consistent over time, indicating that our findings reflect real changes, although it is worth noting that attitudes to alcohol have shifted in Australia [48], meaning that this assumption may not hold.

Nevertheless, the use of a large sample of data from a long‐running survey that has used consistent items over a period of substantial change in youth drinking makes this an important contribution to the literature on the consequences of recent changes in youth drinking. There are observations that with the decline in youth drinking, non‐drinking has become normalized [49] and, conversely, a denormalization of youth drinking may be under way. While more research in other settings and using other data sources is critical to assess how the relative riskiness of adolescent drinking has shifted during this period of youth drinking decline, our findings provide support to recent concerns that denormalizing processes may inadvertently risk further marginalizing adolescents who drink in non‐normative ways [39]. Further research to understand both the population health gains, as well as the potential problems, associated with declining adolescent drinking is needed to ensure that harms from drinking do not become disproportionately entrenched in the most vulnerable populations.

DECLARATION OF INTERESTS

None.

AUTHOR CONTRIBUTIONS

Michael Livingston: Conceptualization; formal analysis; funding acquisition. Jonas Raninen: Conceptualization; investigation; methodology. Amy Pennay: Conceptualization; investigation. Sarah Callinan: Conceptualization; formal analysis; investigation; methodology.

Supporting information

Table S1. Prevalence of items making up Risky Behaviour Score

Table S2. Prevalence of items making up Delinquent Behaviour Score

Table S3. Models assessing the relationship between age of initiation and adult drinking and risk behaviours with continuous age of initiation specification

Figure S1a‐d. predicted marginal probabilities from regression models, showing the changing relationships between early initiation of drinking and later behaviour with continuous age of initiation specification

Livingston M, Raninen J, Pennay A, Callinan S. The relationship between age at first drink and later risk behaviours during a period of youth drinking decline. Addiction. 2023;118(2):256–264. 10.1111/add.16036