F

P

NTa

b

c

d

e

a

ARRAA

KAADLD

1

hoHfli

s

of

(

0h

Drug and Alcohol Dependence 135 (2014) 133– 139

Contents lists available at ScienceDirect

Drug and Alcohol Dependence

jo ur nal homep ag e: www.elsev ier .com/ locate /drugalcdep

ull length article

redictors of drinking patterns in adolescence: A latent class analysis

icki Jacksona,∗, Simon Dennyb, Janie Sheridanc, Terry Flemingb, Terryann Clarkd,asileta Teevalee, Shanthi Ameratungaa

Section of Epidemiology and Biostatistics, School of Population Health, The University of Auckland, Auckland, New ZealandDepartment of Paediatrics: Child and Youth Health, The University of Auckland, Auckland, New ZealandCentre for Addiction Research, The University of Auckland, Auckland, New ZealandSchool of Nursing, The University of Auckland, Auckland, New ZealandDepartment of Pacific Health, School of Population Health, The University of Auckland, Auckland, New Zealand

r t i c l e i n f o

rticle history:eceived 15 August 2013eceived in revised form 21 October 2013ccepted 29 November 2013vailable online 12 December 2013

eywords:dolescentlcoholrinking patternsatent class analysiseterminants

a b s t r a c t

Background: Uni-dimensional measures of alcohol consumption may be unable to fully capture the com-plexity of adolescent drinking and experience of alcohol-related harms. Latent class analysis provides anempirical method to understand different adolescent drinking patterns.Methods: Latent class analysis was used to create typologies of drinking among the 5018 current drinkersin the national Youth ‘07 survey. Determinants of drinking patterns were identified using multinomiallogistic regression.Results: Four latent classes were identified, demonstrating an overall increase in risk of alcohol-relatedoutcomes from increasing consumption. One class strongly deviated from this pattern, having moder-ate consumption patterns but disproportionately high levels of alcohol-related problems. Multinomiallogistic regression found that the strongest predictors of belonging to high-risk drinking typologies werehaving a positive attitude to regular alcohol use, buying own alcohol, peers using alcohol, and obtain-ing alcohol from friends and/or other adults. Other significant predictors included being male, having astrong connection to friends, having parents with a low level of knowledge of their daily activities and

poor connection to school. Class membership also varied by ethnicity.Conclusion: The latent class approach demonstrated variability in alcohol-related harms across groupsof students with different drinking patterns. Longitudinal studies are necessary to determine the causesof this variability in order to inform the development of targeted policy and preventative interventions.Legislative controls, such as increasing the legal purchase age and reducing the commercial availabilityof alcohol, will continue to be important strategies for reducing harm in young people.

. Introduction

Effective, preventative interventions to reduce adolescent alco-ol use must address the most significant risk and protective factorsf underage drinking. The social development model (Catalano andawkins, 1996) guides the identification of relevant contextual

actors within a target population by suggesting that adolescentsearn patterns of behavior from their primary units of socialization

ncluding peers, family, and school.

A number of systematic reviews have identified the mostalient peer (Leung et al., 2011), family (Hayes et al., 2004) and

∗ Corresponding author at: Faculty of Medical and Health Sciences, The Universityf Auckland, Private Bag 92019, Auckland, New Zealand. Tel.: +64 9 923 6722;ax: +64 9 3737503.

E-mail addresses: nicki.jackson@auckland.ac.nz, nickijackson@hotmail.comN. Jackson).

376-8716/$ – see front matter © 2013 Elsevier Ireland Ltd. All rights reserved.ttp://dx.doi.org/10.1016/j.drugalcdep.2013.11.021

© 2013 Elsevier Ireland Ltd. All rights reserved.

school-related factors (Fletcher et al., 2008) in the initiation anduse of alcohol during adolescence. However, given the hetero-geneity in the measurement of alcohol use in these reviews itremains uncertain as to how each of these factors is associatedwith different levels (or types) of drinking.

Understanding drinking typologies is necessary to inform thedevelopment of appropriate intervention and policy approaches.Researchers (Carey, 2001; Gmel et al., 2011; White, 1987) areincreasingly calling for a move away from broad classifications ofdrinkers (e.g., binge drinkers) to considering more distinct typesof drinkers. For example, it is claimed that the averaging acrossbinge drinkers in studies is likely to combine different types ofdrinkers with very different frequencies of binge drinking and/oramounts consumed, masking potentially different levels of risk

of alcohol-related harms. Consequently, there are compelling rea-sons to undertake research that can differentiate regular heavydrinkers from those who demonstrate risky episodic patterns ofdrinking.

1 hol De

ds2madatlT

phaaWaitvDmiiaar

ssdt

2

a2qi

pZrrwHmG

iadsbsuod5imsw

34 N. Jackson et al. / Drug and Alco

Researchers also argue that such insensitivity to differences inrinking patterns will obscure the understanding of risk relation-hips associated with lower levels of drinking (Thompson et al.,012). Studies have found that drinkers who consume low andoderate volumes with occasional episodes of risky binge drinking

ccount for the majority of alcohol-related problems (e.g., acci-ents, problems at work; Dawson, 2011; Gmel et al., 2001). Amongdolescents and young adults, the association between consump-ion and alcohol-related concerns may begin with consumptionevels as low as two drinks per occasion (Gruenewald et al., 2010;hompson et al., 2012).

The failure of uni-dimensional measures to capture the com-lexity of adolescent drinking and experience of alcohol-relatedarms has resulted in the use of more sophisticated methods, suchs latent class analysis, to identify drinking typologies and risksssociated with varying types of drinking (Percy and Iwaniec, 2007;ells et al., 2004). This categorical, person-centred approach is

rgued to be more appropriate for alcohol-related questions, ast focuses on the relationships among individuals who are similaro each other in their drinking pattern and different from indi-iduals in other drinking typologies (Muthen and Muthen, 2000).espite the opportunities afforded by these methods to describeore nuanced typologies of drinking, there has been inconsistency

n the selection and definition of variables (e.g., frequency of drink-ng, binge drinking, amount typically consumed and experience oflcohol-related outcomes) to construct drinking typologies (Cablend Sacker, 2008). Consequently, comparisons between studiesemain difficult.

Using a comprehensive data resource from a nationally repre-entative health and well-being survey of New Zealand secondarychool students, the current study aims to identify adolescentrinking profiles and to determine if these profiles are differen-iated by levels of risk and protective factors.

. Methods

The anonymous, computer-assisted survey was self-dministered to 9107 New Zealand secondary schools students in007, using hand-held internet tablets. The survey contained 622uestions covering a broad range of health and wellbeing issues,

ncluding alcohol use.One hundred and fifteen schools were randomly selected for

articipation, from the 389 eligible secondary schools in Newealand. Eighteen per cent of eligible students (or 30 students if theoll was less than 166) were randomly selected from each school’soll and invited to participate. The school and student response rateere 84 and 74% respectively. The survey instrument (Adolescentealth Research Group) and a detailed description of the surveyethodology are available elsewhere (Adolescent Health Researchroup, 2008).

Latent class analysis was used to examine the underly-ng structure of co-occurring drinking behaviors and associatedlcohol-related problems among the current drinkers. Currentrinkers were defined as students who stated that they had con-umed alcohol and did not indicate they no longer drank. Seveninary parameters were used to create homogeneous groups withimilar drinking patterns, three of which were consumption meas-res: (1) frequency of consumption in the past 4 weeks (not at allr up to 3 times, once a week or more); (2) number of standardrinks typically consumed during a drinking occasion (1–4 drinks,+); and (3) frequency of binge drinking (5 or more drinks) dur-

ng the past 4 weeks (none to once in the last 4 weeks, twice orore). Four self-reported problems associated with the young per-

on’s drinking (referred from hereon as ‘alcohol-related outcomes’)ere also included: having performance at school or work affected,

pendence 135 (2014) 133– 139

having unsafe sex and/or unwanted sex, doing things that couldhave resulted in serious trouble (e.g., stealing), been injured and/orinjuring another and/or being involved in a car crash. Students wereasked to report whether they had experienced any of the outcomesin the previous 12 months.

2.1. Predictors of latent class membership

A range of individual, peer, family, and school factors relatedto adolescent drinking were included in the analysis. All com-posite scales relating to peer, school, and family connection weredrawn from the Adolescent Resilience Questionnaire (Gartlandet al., 2011).

2.1.1. Individual factors.

(a) Demographics: Age, gender, and ethnicity were determined byself-report. Ethnicity was categorized and prioritized into oneof four ethnic groups, according to the New Zealand census eth-nicity question (Statistics New Zealand, 2007): Maori, Pacific,Asian, and New Zealand European or Other ethnicity.

Socioeconomic status (SES) was determined using a combina-tion of measures including the 2006 New Zealand DeprivationIndex (Salmond et al., 2007) and other indicators of deprivation(i.e., use of a sleep-out for a bedroom, perception that parentsworry about having enough money to buy food, number oftimes the young person has moved home and absence of TV,computer, mobile, phone, or car at the family home). The com-bined SES measure was found to have an acceptable reliability(Cronbach’s alpha = 0.67).

(b) Other individual factors included having a regular job outside ofschool hours in the last year, belonging to a sports team or cluboutside of school time or in the weekend, buying their own alco-hol, and having a permissive attitude to regular alcohol use bypeople their age. Students who responded “yes” were comparedto students who answered “no”.

2.1.2. Peer factors. Three peer variables were analyzed: peer useof alcohol, acquisition (i.e., social supply) of alcohol from friendsand/or another adult (i.e., other than parents), and connection tofriends. The latter measure was determined by responses from anumber of survey questions, including number of friends, abilityto make and keep friends, having a group of friends to hang outwith and have fun, having close friends, perceiving that friends care,having friends that like doing the same thing, and having friendsthat help out ( = 0.77).

2.1.3. Family factors. Four family factors were included: parentalalcohol use, parents supplying alcohol to the young person, con-nection to family, and parental knowledge of youth daily activities.Connection to family was a composite measure of questions relat-ing to the frequency of the family having fun together, perceptionsof family relationships and closeness, and time spent with parents( = 0.84). Parental knowledge of youth daily activities (hereonreferred to as ‘parental knowledge’) was composed of three ques-tions relating to how well the young person’s parents knew abouttheir friends and where they go after school and at night ( = 0.67).

2.1.4. School factors. A composite measure, assessing connectionto school, was chosen as the predictor within the school setting.This measure assessed the young person’s perceptions about how

they feel about school, whether they feel part of the school, whetherpeople at school care about them, whether teachers go out of theirway to help students and whether they show a special interestin the young person’s culture or ethnic group. The measure also

hol Dependence 135 (2014) 133– 139 135

ia

2

smeAmtncft

tmilmeitmimsero

dMmctams(

setr

3

w

Table 1Distribution of drinking behaviors and alcohol-related outcomes, in adolescentdrinkers.

Measure n Sample (%)

Frequency of drinking alcohol in the last 4 weeksUp to 3 occasions 3541 70.6At least weekly drinking occasions 1477 29.4Typical amount consumed1–4 drinks 2685 54.05 or more drinks 2288 46.0Frequency of binge drinking (5+ drinks) in the last 4 weeksNot in the last 4 weeks or once 3320 67.0Twice or more 1632 33.0Alcohol-related outcomesPerformance affected 478 9.8Unsafe and/or unwanted sex 806 16.5Troublea 968 19.9Injury and/or car crash 1174 24.0

TF

Nl

N. Jackson et al. / Drug and Alco

ncluded students’ participation in school sports teams, social clubs,nd activities that help others at school ( = 0.67).

.2. Analysis

Beginning with a parsimonious one-class model, we fitted aeries of models with increasing number of classes to find theodel that provided the best fit to the data. As described by Lanza

t al. (2007) model assessment took a heuristic approach, using thekaike Information Criterion, sample size adjusted Bayesian Infor-ation Criterion, and Lo–Mendell–Rubin adjusted likelihood ratio

est as tools to guide the determination of how well the differentumber of classes separated the different drinking patterns. Otheronsiderations of model fit included being able to assign meaning-ul labels to each class and having no individual class which wasoo small in size.

The maximum number of latent classes examined was six, usinghe method of maximum likelihood in Mplus (Version 7.11). This

ethod is iterative, with the estimation algorithm ideally converg-ng on the parameter values associated with the single largest logikelihood. However, in latent class analysis the maximum reached

ay in fact be the best solution in a local neighborhood of param-ter space, but not the global maximum (Uebersax, 2009). Thiss because the latent class algorithm cannot distinguish whetherhe maximum is a global or local, a problem that is more com-

on in complex models with a high number of latent classes andndicators. This problem has been described as akin to climbing a

ountain in the dark, with the first peak you reaching not neces-arily being the highest peak of the mountain (Uebersax, 2009). Tonsure that the global maximum was reached in this study, severaluns with random starting values were performed for each numberf latent classes.

Multinomial regression was used to explore the significant pre-ictors of latent class membership. The new R3STEP method inplus (Version 7.11) was utilized, as this method incorporates theeasurement error (from the determination of most likely latent

lass) in the independent evaluation of the relationship betweenhe latent class variable and the predictor variables (Asparouhovnd Muthen, 2013). This has many advantages over traditionalethods of assigning individuals to the most likely latent class and

ubsequently conducting standard multinomial logistic regressionVermunt, 2010).

Due to the unequal probabilities of selection of students withinchools and likely clustering of responses by school all standardrrors and model fit statistics were adjusted for clustering usinghe Mixture Complex feature in Mplus (Version 7.11). Results areeported as odds ratios with their 95% confidence intervals.

. Results

Of the 8294 students who responded to the alcohol questionsithin the survey, 5018 (60.5%) were current drinkers, of whom

able 2it indices for the latent class analysis of adolescent drinking behaviors and alcohol-relat

Model

2 classes 3 classes

AIC 31461.71 31029.38

BIC 31559.52 31179.36

SSABIC 31511.86 31106.27

LRT p value 0.00 0.00

Entropy 0.80 0.83

ote: AIC: Akaike information criteria; BIC: Bayesian information criteria; SSABIC: Samikelihood ratio test.

a Doing things that could have resulted in serious trouble (e.g., stealing)

57% reported they had participated in binge drinking at least oncein the past 4 weeks. One in every six (17.5%) current drinkers drank10 or more standard drinks during a usual drinking occasion. Thedistribution of drinking behaviors and alcohol-related outcomes inthe entire sample of current drinkers is shown in Table 1.

3.1. Latent class analysis

Fit indices for the latent class analysis are shown in Table 2.Model fit indices, including the Bayesian Information Criterion andLo–Mendell–Rubin likelihood ratio test, indicated that the fourclass model was the best fit for the data. Examination of bivariatemodel fit information revealed no standardized residuals betweenpairs of indicators, indicating that the categorical latent class indi-cators were independent within a latent class.

Overall, the four latent classes represented the continuum oflevels of alcohol consumption (Table 3). The majority (55.0%) ofcurrent drinkers were assigned to Class One, a low-risk groupwith a low prevalence of frequent and heavy drinking and alcohol-related outcomes. One in eight current drinkers was assigned toClass Two, labeled as the moderate-risk group, due to their over-all low frequency of drinking and binge drinking. Classes Threeand Four were labeled high-risk and very high-risk, respectively,due to exhibiting a high prevalence of at-risk drinking behav-iors and alcohol-related outcomes. Although students in Class Twowere characterized by a pattern of low to moderate alcohol con-sumption, the group suffered disproportionately higher rates ofalcohol-related outcomes in comparison to Class Three, the high-risk group. On further analysis, we found a significant difference

in the proportion of moderate-risk drinkers compared to high-riskdrinkers (17.0 vs. 12.6%, p < 0.05) who had ever tried to cut downor had cut down their drinking.

ed outcomes.

4 classes 5 classes 6 classes

30702.88 30677.28 30659.5730905.03 30931.59 30966.0530806.52 30807.66 30816.700.00 0.24 0.380.80 0.78 0.73

ple size adjusted Bayesian information criteria; LRT: Lo–Mendell–Rubin adjusted

136 N. Jackson et al. / Drug and Alcohol Dependence 135 (2014) 133– 139

Table 3Item-response probabilities of drinking measures and outcomes, by latent class membership.

Class OneLow-risk55.0%(n = 2761)

Class TwoModerate-risk12.5%(n = 628)

Class ThreeHigh-risk19.7%(n = 990)

Class FourVery high-risk12.7%(n = 639)

Frequency of drinking alcohol in the last 4 weeksUp to 3 occasions 0.90 0.82 0.42 0.19At least weekly drinking occasions 0.10 0.18 0.58 0.81Typical amount consumed1–4 drinks 0.80 0.39 0.19 0.105 or more drinks 0.20 0.61 0.81 0.90Frequency of binge drinking (5+ drinks) in the last 4 weeksUp to once 0.98 1.00 0.00 0.03Twice or more 0.02 0.00 1.00 0.97Alcohol-related outcomesPerformance affected 0.00 0.22 0.05 0.47Unsafe and/or unwanted sex 0.02 0.30 0.22 0.57Troublea 0.03 0.43 0.17 0.73

0.

3

CTdg2bte

TO

Injury and/or crash 0.04

a Doing things that could have resulted in serious trouble (e.g., stealing).

.2. Multinomial regression

Table 4 presents the results of multinomial regression, withlass One (i.e, the low-risk group) treated as the reference group.he strongest individual-level predictor of belonging to a high-riskrinking group was buying one’s own alcohol, with the high-riskroup and very high-risk group being almost three (OR 2.83 (95% CI.02–3.96)) to six times (OR 5.74 (95% CI 3.76–8.74)) more likely to

uy their alcohol than the low-risk group. In addition, students inhe very high-risk group were three times more likely to be of Maorithnicity (OR 3.16 (95% CI 2.44–4.09)), have a permissive attitude

able 4dds ratios (95% Confidence intervals) of the association between predictors and latent c

Moderate-risk

Individual factorsAgea

14years 1.24 (0.74–2.09)

15years 1.07 (0.61–1.85)

16years 1.75 (1.02–3.01)*

17years+ 1.49 (0.80–2.78)

Genderb

Male 1.36 (1.04–1.79)*

Ethnicityc

Maori 2.90 (1.97–4.26)*

Pacific 3.17 (1.91–5.26)*

Asian 0.29 (0.15–0.56)*

Socio-economic status (SES)d

Low SES 1.10 (0.77–1.57)

Mid SES 0.81 (0.54–1.20)

Regular job 1.63 (1.22–2.19)*

Belong to sports club 0.97 (0.74–1.27)

Positive attitude to alcohol 1.79 (1.30–2.47)*

Buys own alcohol 1.91 (1.17–3.12)*

Peer factorsFriends use alcohol 4.41 (2.13–9.10)*

Peers and/or others supply alcohol 3.99 (2.61–6.09)*

Connection to friends 1.25 (1.10–1.43)*

Family factorsParents supply alcohol 0.78 (0.56–1.07)

Parents use alcohol 0.62 (0.42–0.92)*

Connection to family 0.72 (0.60–0.87)*

Parental knowledge 0.74 (0.63–0.87)*

School factorsConnection to school 0.79 (0.69–0.90)*

a Reference group = Aged 13 years and under.b Reference group = Female students.c Reference group = European and other ethnicities.d Reference group = High SES.* p ≤ 0.05.

49 0.27 0.83

to regular alcohol use (OR 3.36 (95% CI 2.42–4.66)) and almosttwice as likely to be of Pacific descent (OR 1.91 (95% CI 1.11–3.28)).Being male and of older age was also found to be associated withbelonging to the high-risk and very high-risk drinking groups.

Peer-level factors were found to be strong predictors of classmembership across all risky drinking typologies, including peer useof alcohol, obtaining alcohol from friends and/or others, and havinga strong connection to friends. In relation to family and school-level

factors, parental knowledge of youth daily activities and connectionto school reduced the likelihood of membership across all drinkingprofiles, particularly in the very high-risk group.

lass membership (all latent classes are compared to the low risk drinking group).

High-risk Very high-risk

1.51 (1.01–2.27)* 1.05 (0.58–1.92)2.38 (1.48–3.83)* 1.53 (0.86–2.74)3.99 (2.51–6.34)* 2.10 (1.13–3.89)*

4.41 (2.64–7.38)* 2.32 (1.23–4.38)*

1.79 (1.47–2.18)* 1.93 (1.37–2.72)*

2.70 (2.08–3.52)* 3.16 (2.44-4.09)*

1.10 (0.65–1.87) 1.91 (1.11-3.28)*

0.32 (0.22–0.46)* 0.15 (0.05–0.45)*

1.27 (0.98–1.64) 1.42 (0.98–2.07)1.05 (0.81–1.37) 0.95 (0.68–1.31)1.22 (1.03–1.45)* 1.35 (0.99–1.83)0.97 (0.81–1.15) 0.79 (0.61–1.03)2.14 (1.74–2.63)* 3.36 (2.42–4.66)*

2.83 (2.02–3.96)* 5.74 (3.76–8.74)*

3.33 (2.17–5.11)* 4.85 (2.28–10.34)*

2.21 (1.73–2.82)* 3.67 (2.34–5.76)*

1.72 (1.45–2.04)* 1.53 (1.26–1.86)*

1.01 (0.80–1.28) 1.09 (0.83–1.43)1.03 (0.82–1.31) 1.08 (0.76–1.53)1.05 (0.92–1.19) 0.71 (0.63–0.80)*

0.74 (0.65–0.85)* 0.56 (0.47–0.66)*

0.79 (0.72–0.86)* 0.62 (0.52–0.74)*

hol De

4

por(2pih

wrcBeOv(3dsfwsdaftacnMstolao

avaWgdTagiedic1db

mtwslse

N. Jackson et al. / Drug and Alco

. Discussion

Our analyses found four latent classes of adolescent drinkingatterns in New Zealand secondary school students. Although overne-half of current drinkers were categorized as having a low-isk drinking pattern, it must be emphasized that health guidelinesDonaldson, 2009; National Health and Medical Research Council,009) recommend that no drinking is the safest option for youngeople under the age of 18 years. Of particular concern was the find-

ng that almost one-third (32.4%) of all current drinkers exhibitedigh or very high-risk drinking patterns.

Interestingly, we also found a group of moderate-risk drinkersho had experienced a disproportionately high number of alcohol-

elated outcomes. This group, which deviates from generalontinuum pattern, was also found in a latent class study ofritish adults (Smith and Shevlin, 2008). There are many possiblexplanations to this finding, all of which require further scrutiny.ne explanation may be that alcohol-related harms occur at lowolumes of alcohol consumption, as has been found by othersGruenewald et al., 2010; Thompson et al., 2012). However, only9% of this group usually drank low volumes of alcohol, defined asrinking less than 5 standard drinks on any occasion. Two othercenarios are more likely to elucidate this anomaly. One is the dif-erence between the reference period for drinking behaviors (i.e., 4eeks) and drinking-related outcomes (i.e., 12 months). As we have

hown, more students in this group had tried to cut down, or had cutown, their drinking in the past twelve months. It is unknown if thisction was taken as the result of experiencing adverse outcomesrom their previous high levels of drinking. This scenario is fur-her described by Gmel et al. (2001), in their study examining whydult drinkers, characterized at moderate risk in terms of averageonsumption and binge drinking occasions, reported a substantialumber of alcohol-related problems. Using the Stages of Changeodel, they found that 40% of moderate drinkers were at the action

tage of beginning to reduce their risky drinking pattern. It was,herefore, suggested that alcohol-related problems may have beenverstated in this group, as the problems may have originated ear-ier when they were high-risk drinkers. However, as stated by theuthors, this scenario cannot fully explain the substantial numberf problems reported by the moderate-risk group.

Another explanation for the disproportionate prevalence ofdverse outcomes is the result of usual drinking patterns (orariability in drinking) moderating the association betweenlcohol-related harm and acute intake of alcohol (Gmel et al., 2006;att et al., 2004). The majority of students in the moderate-risk

roup drank alcohol infrequently and if they did have a bingerinking episode, it occurred on average once per month or less.his would suggest that infrequent drinkers may be more vulner-ble to the effects of consuming a large amount of alcohol on anyiven drinking occasion. Many studies have found this to be truen the adult drinking population (Gmel et al., 2006; Gruenewaldt al., 1996; Treno et al., 1997), showing that drinkers who usuallyrank little, but sometimes drank heavily, were at particular risk of

njury. This was demonstrated also for traffic crashes, where the riskurve increases more steeply in less frequent drinkers (Hurst et al.,994). It is believed that this phenomenon is the result of heavierrinkers having a higher tolerance to alcohol. Further study woulde required to determine if this theory holds true for adolescents.

The difference in alcohol-related outcomes between theoderate-risk group and the high-risk group may also be due to

he environment in which drinking occurs. Again, future researchould be required to investigate if high-risk drinkers (in compari-

on to the other drinking typologies) consume alcohol in differentocations, resulting in them being less exposed to situations whereome harms may occur (Gmel et al., 2006). It must be noted, how-ver, that although the prevalence of alcohol-related outcomes is

pendence 135 (2014) 133– 139 137

lower in the high-risk group in comparison to the moderate-riskgroup, the groups may differ in their frequency of problems.

Our study also identified ethnic inequalities across drinkingtypologies. Maori students showed increasing odds of belongingto high-risk drinking groups, a finding consistent with previousresearch (Clark et al., 2013). In contrast, Asian students were signif-icantly less likely to engage in risky drinking behaviors, with mixedresults for students of Pacific origin. Identifying the causes of theseinequalities, particularly for Maori (the indigenous population inNew Zealand), is paramount and it is likely that such inequali-ties are socially determined. For example, a separate analysis ofthe same survey utilized in this study found that students (includ-ing Maori) who had experienced ethnic discrimination were morelikely to report an episode of binge drinking in the past four weeks(Crengle et al., 2012). Hence, appropriate interventions to addressalcohol harm for Maori will need to address the wider determinantsof health and, as such, solutions may be found outside of the healthsector.

The strongest predictors of membership across the risky drink-ing latent classes were purchasing one’s own alcohol and socialsupply of alcohol. This suggests that evidence-based policies (Baboret al., 2010), such as reducing the commercial availability of alcoholand increasing the legal purchase age (to have a trickledown effecton social supply), will continue to be the most appropriate strate-gies to reduce alcohol-related harm for most adolescents, not justthose deemed to demonstrate “hazardous drinking” patterns.

Peer use of alcohol was also a significant predictor of member-ship of riskier drinking latent classes. For the moderate-risk group,peer factors and social supply were the two strongest predictors ofmembership. It is uncertain if these factors are associated with thehigher prevalence of alcohol-related harm in this group. Reboussinet al. (2006) and Fergusson et al. (1995) also found peer factors to bestrongly associated with drinking typologies, with the latter studyfinding that the effects of other individual, familial and social factorson alcohol abuse were mediated via their effect on peer affiliationsin adolescence. This is confirmed in a systematic review of peerinfluences on alcohol use, demonstrating that parental acceptanceof adolescent alcohol use influences peer selection and affiliation(Leung et al., 2011). Interventions to address peer factors will,therefore, need to involve multiple components, targeting familiesin addition to adolescents.

The influence of parental knowledge is noteworthy. This studyshowed that parental knowledge of youth daily activities was asso-ciated with a reduced likelihood of membership of higher riskdrinking groups, being particularly protective for the very high-risk group. A systematic review of parenting factors found thatparental knowledge was the factor most consistently and directlyassociated with adolescent alcohol use (Hayes et al., 2004). A lack ofparental knowledge was found to be associated with earlier onsetof drinking and heavier drinking patterns, in addition to having asecondary effect on peer selection, whereby poorly supervised ado-lescents were found to be more likely to associate with “deviant”peers. Interestingly, recent research (Kerr et al., 2012) suggests thatparental knowledge exerts its positive effect through adolescentdisclosure of information, rather than from parents’ direct moni-toring of adolescent behavior. Unfortunately, evidence from highquality interventions to address parental knowledge is scarce. Onefour-year, family-centred randomized controlled trial to improveparental knowledge in very young, high-risk adolescents foundreduced substance use by the first year of high school and main-tenance of parental monitoring behaviors (Dishion et al., 2003).Hence, parenting interventions may need to be considered early

in the life course of adolescence and trajectory of alcohol use.

All of the above findings clearly demonstrate the need forapproaches, such as latent class analysis, to fully understand therisk from, and determinants of, varying patterns of drinking. Using

1 hol De

sipottoycItpmmrsRnafef

scdRplasod

baddgtiecaTpia

A

tpZZHssi

smp(

38 N. Jackson et al. / Drug and Alco

ingle, broad measures of drinking, such as binge drinking or typ-cal amount consumed, would result in the combination of youngeople who frequently drink heavily with those who have a lowverall intake with very occasional binges. This study finds thathe consequences of these drinking patterns is different and needso be fully explored. Despite the advantages of this approach,nly a small number of studies have utilized latent class anal-sis to determine drinking typologies in young people withoutlinical alcohol dependence (Fergusson et al., 1995; Percy andwaniec, 2007; Reboussin et al., 2006). Many have limitations dueo the assigning of individuals to the most likely latent class (arocess which includes measurement error) in order to conductultinomial regression. Furthermore, it has been found that assign-ent of individuals to latent classes, whether this is by modal,

andom or proportional assignment, underestimates the relation-hips between predictors and class membership (Vermunt, 2010).esearchers should take advantage of advances in software whichow permit the regression to take into account this error fromssigning individuals to latent classes. We also recommended thatuture latent class analyses use multi-level designs in order toxamine the upstream determinants of drinking patterns, such asactors at the neighborhood level.

Notwithstanding the knowledge gained, the findings of thistudy must be interpreted in light of several limitations. Theross-sectional design of this study precludes determination of theirection of causation between predictors and drinking typologies.ecall bias may also be present given the retrospective referenceeriods for drinking and its outcomes. Finally, as this study was

imited to students who attended secondary school, the findingsre not necessarily generalizable to young people who are not atchool (or did not attend the day of the survey), home-schooled,r in alternative education. Drinking patterns in these groups mayiffer from the sample population included in this study.

There was no single predictor in our study which discriminatedetween the different drinking patterns or fully explained the vari-bility in alcohol-related outcomes experienced by students withifferent drinking patterns. It is imperative to continue to explorerinking typologies so that causes of variability in risk betweenroups can be identified and addressed within targeted interven-ions. The current study highlights that, for those at highest risk,t is universal policy approaches that are likely to have the great-st impact on reducing alcohol-harm. This includes reducing theommercial availability of alcohol and increasing the legal purchasege in order to prevent social supply of alcohol from older friends.hese policy initiatives should complement community-actionrogrammes which are also informed by a thorough understand-

ng of the specific risk and protective profiles as they pertain to thedolescents in their community.

cknowledgements

We acknowledge the Health Research Council of New Zealand,he Department of Labour, Families Commission, Accident Com-ensation Corporation of New Zealand, Sport and Recreation Newealand, Health Promotion Agency, Te Puni Kokiri, Vodafone Newealand and the Ministries of Youth Development, Justice, andealth for providing financial assistance to complete the Youth’07

urvey. We also acknowledge the young people from New Zealandecondary schools for sharing their information with us, their fam-lies/whanau and staff at the schools that participated.

Role of funding source: Funding for the analysis of Youth’07

urvey data was provided by the Health Promotion Agency (for-erly Alcohol Advisory Council of New Zealand). The Youth’07

roject was funded by the Health Research Council of New Zealandgrant05/216), the Department of Labour, Families Commission,

pendence 135 (2014) 133– 139

Accident Compensation Corporation of New Zealand, Sport andRecreation New Zealand, the Health Promotion Agency and theMinistries of Youth Development, Justice, Health and Te Puni Kokiri.Vodafone New Zealand provided support for electronic communi-cation. All funding bodies had no further role in the study design;in the collection, analysis and interpretation of data; in the writingof the report; or in the decision to submit the paper for publication.

Contributors: All authors were involved in the design of thestudy. Jackson conducted all analyses with assistance of Denny.Jackson wrote the first draft of the manuscript, with all authorscontributing to and approving the final manuscript.

References

Adolescent Health Research Group, 2008. Youth’07: The Health and Wellbeing ofSecondary School Students in New Zealand. University of Auckland, Auckland.

Adolescent Health Research Group, 2013. Youth’07 Student Health and Well-being Questionnaire, Retrieved from: http://www.fmhs.auckland.ac.nz/faculty/ahrg/ docs/youth07-student-survey-web.pdf [accessed 07.02.2013].

Asparouhov, T., Muthen, B., 2013. Auxiliary Variables in Mixture Modeling: 3-StepApproaches Using Mplus, Mplus Web Notes: No. 15.

Babor, T., Caetano, R., Casswell, S., Edwards, G., Giesbretcht, N., Graham, K., Grube,J., Hill, L., Holder, H., Homel, R., Livingston, M., Osterberg, E., Rehm, J., Room, R.,Rossow, I., 2010. Alcohol: No Ordinary Commodity: Research and Public Policy.Oxford University Press, Oxford, pp. 239–257.

Cable, N., Sacker, A., 2008. Typologies of alcohol consumption in adolescence: pre-dictors and adult outcomes. Alcohol Alcohol 43, 81–90.

Carey, K.B., 2001. Understanding binge drinking: introduction to the special issue.Psychol. Addict. Behav. 15, 283–286.

Catalano, R.F., Hawkins, J.D., 1996. The social development model: a theory of antiso-cial behavior. In: Hawkins, J.D. (Ed.), Delinquency and Crime: Current Theories.Cambridge University Press, NY, pp. 149–197.

Clark, T.C., Robinson, E., Crengle, S., Sheridan, J., Jackson, N., Ameratunga, S., 2013.Binge drinking among Maori secondary school students in New Zealand: asso-ciations with source, exposure and perceptions of alcohol use. N. Z. Med. J. 126,55–69.

Crengle, S., Robinson, E., Ameratunga, S., Clark, T., Raphael, D., 2012. Ethnic dis-crimination prevalence and associations with health outcomes: data from anationally representative cross-sectional survey of secondary school studentsin New Zealand. BMC Public Health 12, 45.

Dawson, D.A., 2011. The challenge of distinguishing high-volume and frequent riskydrinking. Addiction 106, 1046–1047, discussion 1050-1041.

Dishion, T.J., Nelson, S.E., Kavanagh, K., 2003. The family check-up with high-riskyoung adolescents: preventing early-onset substance use by parent monitoring.Behav. Ther. 34, 553–571.

Donaldson, L., 2009. Guidance on the consumption of alcohol by children and youngpeople. Department of Health, London.

Fergusson, D.M., Horwood, L.J., Lynskey, M.T., 1995. The prevalence and risk fac-tors associated with abusive or hazardous alcohol consumption in 16-year-olds.Addiction 90, 935–946.

Fletcher, A., Bonell, C., Hargreaves, J., 2008. School effects on young people’s druguse: a systematic review of intervention and observational studies. J. Adolesc.Health 42, 209–220.

Gartland, D., Bond, L., Olsson, C.A., Buzwell, S., Sawyer, S.M., 2011. Developmentof a multi-dimensional measure of resilience in adolescents: the AdolescentResilience Questionnaire. BMC Med. Res. Methodol. 11, 134.

Gmel, G., Klingemann, S., Muller, R., Brenner, D., 2001. Revising the preventive para-dox: the Swiss case. Addiction 96, 273–284.

Gmel, G., Bissery, A., Gammeter, R., Givel, J.C., Calmes, J.M., Yersin, B., Daeppen, J.B.,2006. Alcohol-attributable injuries in admissions to a Swiss emergency room–ananalysis of the link between volume of drinking, drinking patterns, and preat-tendance drinking. Alcohol. Clin. Exp. Res. 30, 501–509.

Gmel, G., Kuntsche, E., Rehm, J., 2011. Risky single-occasion drinking: bingeing isnot bingeing. Addiction 106, 1037–1045.

Gruenewald, P.J., Mitchell, P.R., Treno, A.J., 1996. Drinking and driving: drinkingpatterns and drinking problems. Addiction 91, 1637–1649.

Gruenewald, P.J., Johnson, F.W., Ponicki, W.R., Lascala, E.A., 2010. A dose-responseperspective on college drinking and related problems. Addiction 105, 257–269.

Hayes, L., Smart, D., Toumbourou, J.W., Sanson, A., 2004. Parental Influences onAdolescent Alcohol Use. Australian Institute of Family Studies, Melbourne.

Hurst, P.M., Harte, D., Frith, W.J., 1994. The Grand Rapids dip revisited. Accid. Anal.Prev. 26, 647–654.

Kerr, M., Stattin, H., Özdemir, M., 2012. Perceived parenting style and adolescentadjustment: revisiting directions of effects and the role of parental knowledge.Dev. Psychol. 48, 1540–1553.

Lanza, S.T., Collins, L.M., Lemmon, D.R., Schafer, J.L., 2007. PROC LCA: a SAS procedurefor latent class analysis. Struct. Equ. Modeling 14, 671–694.

Leung, R.K., Toumbourou, J.W., Hemphill, S.A., 2011. The effect of peer influence andselection processes on adolescent alcohol use: a systematic review of longitu-dinal studies. Health Psychol. Rev. 4, 1–32.

hol De

M

N

P

R

S

S

S

N. Jackson et al. / Drug and Alco

uthen, B., Muthen, L.K., 2000. Integrating person-centered and variable-centeredanalyses: growth mixture modeling with latent trajectory classes. Alcohol. Clin.Exp. Res. 24, 882–891.

ational Health and Medical Research Council, 2009. Australian guidelines to reducehealth risks from drinking alcohol. Commonwealth of Australia, Canberra.

ercy, A., Iwaniec, D., 2007. The validity of a latent class typology of adolescentdrinking patterns. Ir. J. Psychol. Med. 24, 13–18.

eboussin, B.A., Song, E.Y., Shrestha, A., Lohman, K.K., Wolfson, M., 2006. A latentclass analysis of underage problem drinking: evidence from a community sam-ple of 16–20 year olds. Drug Alcohol Depend. 83, 199–209.

almond, C., Crampton, P., Atkinson, J., 2007. NZDep2006 Index of Deprivation:User’s Manual. Department of Public Health, University of Otago, Wellington.

mith, G.W., Shevlin, M., 2008. Patterns of alcohol consumption and relatedbehaviour in Great Britain: a latent class analysis of the alcohol use disorderidentification test (AUDIT). Alcohol Alcohol 43, 590–594.

tatistics New Zealand, 2007. Classifications and Related Statistical Standards: UrbanArea Page. Statistics New Zealand, Wellington.

pendence 135 (2014) 133– 139 139

Thompson, K.D., Stockwell, T., MacDonald, S., 2012. Is there a ‘low-risk’ drinkinglevel for youth? The risk of acute harm as a function of quantity and frequencyof drinking. Drug Alcohol Rev. 31, 184–193.

Treno, A.J., Gruenewald, P.J., Ponicki, W.R., 1997. The contribution of drinking pat-terns to the relative risk of injury in six communities: a self-report basedprobability approach. J. Stud. Alcohol 58, 372–381.

Uebersax, J., 2009. LCA Frequently Asked Questions, Retrieved from:http://www.john-uebersax.com/stat/faq.htm#maxima[accessed 10.11.13].

Vermunt, J.K., 2010. Latent class modeling with covariates: two improved three-stepapproaches. Polit. Anal. 18, 450–469.

Watt, K., Purdie, D.M., Roche, A.M., McClure, R.J., 2004. Risk of injury from acute alco-hol consumption and the influence of confounders. Addiction 99, 1262–1273.

Wells, J.E., Horwood, L.J., Fergusson, D.M., 2004. Drinking patterns in mid-adolescence and psychosocial outcomes in late adolescence and early adulthood.Addiction 99, 1529–1541.

White, H.R., 1987. Longitudinal stability and dimensional structure of problemdrinking in adolescence. J. Stud. Alcohol 48, 541–550.