Shifting patterns of variance in adolescent alcohol use: Testing consumption as a developing trait-state

Addictive Behaviors 55 (2016) 25–31

Contents lists available at ScienceDirect

Addictive Behaviors journal homepage: www.elsevier.com/locate/addictbeh

Shifting patterns of variance in adolescent alcohol use: Testing consumption as a developing trait-state Logan J. Nealis a, Kara D. Thompson c, Marvin D. Krank d, &, Sherry H. Stewart a,b,⁎ a

Department of Psychology and Neuroscience, Dalhousie University, 1355 Oxford Street, PO Box 15000, Halifax, Nova Scotia B3H4R2, Canada Department of Psychiatry, Dalhousie University, 5909 Veteran's Memorial Lane, Halifax, Nova Scotia B3H2E2, Canada Psychology Department, St. Francis Xavier University, 2323 Notre Dame Ave, Antigonish, Nova Scotia B2G2W5, Canada d Irving K. Barber School of Arts and Sciences, Department of Psychology, University of British Columbia, 1147 Research Rd, Kelowna, British Columbia V1V1V7, Canada b c

H I G H L I G H T S • • • •

Adolescent alcohol use is both a state (fluctuating) and a trait (stable over time). Alcohol use becomes more stable over time, particularly for alcohol quantity. Trait-like stability increases over development for both males and females. The increased stability of use has important implications for prevention.

a r t i c l e

i n f o

Article history: Received 29 June 2015 Received in revised form 12 November 2015 Accepted 16 December 2015 Available online 18 December 2015 Keywords: Alcohol Adolescence Variance decomposition Cohort-sequential Sex differences

a b s t r a c t While average rates of change in adolescent alcohol consumption are frequently studied, variability arising from situational and dispositional influences on alcohol use has been comparatively neglected. We used variance decomposition to test differences in variability resulting from year-to-year fluctuations in use (i.e., state-like) and from stable individual differences (i.e., trait-like) using data from the Project on Adolescent Trajectories and Health (PATH), a cohort-sequential study spanning grades 7 to 11 using three cohorts starting in grades seven, eight, and nine, respectively. We tested variance components for alcohol volume, frequency, and quantity in the overall sample, and changes in components over time within each cohort. Sex differences were tested. Most variability in alcohol use reflected state-like variation (47–76%), with a relatively smaller proportion of trait-like variation (19–36%). These proportions shifted across cohorts as youth got older, with increases in trait-like variance from early adolescence (14–30%) to later adolescence (30–50%). Trends were similar for males and females, although females showed higher trait-like variance in alcohol frequency than males throughout development (26–43% vs. 11–25%). For alcohol volume and frequency, males showed the greatest increase in trait-like variance earlier in development (i.e., grades 8–10) compared to females (i.e., grades 9–11). The relative strength of situational and dispositional influences on adolescent alcohol use has important implications for preventative interventions. Interventions should ideally target problematic alcohol use before it becomes more ingrained and trait-like. © 2015 Elsevier Ltd. All rights reserved.

1. Introduction Adolescence is a period of rapid change, with marked increases in alcohol use during this critical developmental period. Data from the Cross-Canada Report on Student Alcohol and Drug Use indicated marked increases in alcohol use from grades 7 to 12, with the prevalence of heavy episodic use (i.e., N5 drinks on one occasion) within the past month rising from under 5% in grade 7 to over 50% in grade ⁎ Corresponding author at: Department of Psychology and Neuroscience, Dalhousie University, 1355 Oxford Street, PO Box 15000, Halifax, Nova Scotia B3H4R2, Canada. E-mail address: [email protected] (S.H. Stewart).

http://dx.doi.org/10.1016/j.addbeh.2015.12.012 0306-4603/© 2015 Elsevier Ltd. All rights reserved.

12 (Young et al., 2011). Similar data from the United States indicate 71% of adolescents consumed alcohol on at least one occasion by grade 12, with 23% reporting heavy episodic use within the past two weeks (Johnston, O'Malley, Bachman, & Schulenberg, 2011). Measures of alcohol consumption (e.g., frequency of drinking occasions; quantity of alcohol consumed per occasion) increase during adolescence and peak around age 21 (Chen & Jacobson, 2012; Thompson, Stockwell, Leadbeater, & Homel, 2014). Early alcohol use predicts heavy or problematic use (Heron et al., 2012; Irons, Iacono, & McGue, 2015; Liang & Chikritzhs, 2015), antisocial or risky behavior (Duncan, Alpert, Duncan, & Hops, 1997; Stueve & O'Donnell, 2005), long-term changes in neurocognitive functioning (Hanson, Medina, Padula, Tapert, &

26

L.J. Nealis et al. / Addictive Behaviors 55 (2016) 25–31

Brown, 2011; Koskinen et al., 2011), and other adverse social and health consequences (Thompson et al., 2014). Given the myriad negative consequences of adolescent alcohol use, understanding the developmental course of this behavior is clearly important. Most research uses descriptive approaches (e.g., Leatherdale & Burkhalter, 2012) or complex statistical models (e.g., latent growthcurves; Duncan, Duncan, & Strycker, 2006) to describe the average trajectories of alcohol use during adolescence, but not every adolescent follows these average trajectories. The average trajectory of drinking varies between adolescents and within a single adolescent over time. Research has shown variation in individual (e.g., personality) and contextual risk factors (e.g., social and environmental influences) plays an important role in the development of alcohol use. However, it is unclear how much variance in adolescent alcohol trajectories results from stable trait-like variation due to individual differences in drinking patterns, versus state-like variation due to changing contexts and environmental influences over development. Given the rapid changes in alcohol use during adolescence, it is conceivable that contextual factors predominate during this early period, but that more stable individual patterns of use are emerging. This has not been tested, however. Research with emerging adults suggests alcohol consumption is a trait-state, meaning alcohol use has a trait-like component (i.e., stable individual differences) and a state-like component (i.e., statedependent fluctuation) when measured over time (Mushquash, Sherry, Mackinnon, Mushquash, & Stewart, 2014). Using variance decomposition, Mushquash et al. showed slightly more variance attributable to trait-like stability than state-dependent factors (57% vs. 43%) in heavy episodic use among university students, suggesting alcohol consumption is best accounted for by both the dispositional stability and situational fluctuation of use over time. Equivalent research with adolescents is lacking, indicating a gap in current knowledge about the transition from fluctuating, context-dependent initial use to more stable trait-like use as adolescents move toward adulthood. If alcohol use is a trait-state by young adulthood, it is theoretically and clinically important to understand if alcohol use is less stable during adolescence and, if so, when it moves from a fluctuating, contextdependent phenomenon (state-like) to the more stable, enduring pattern of behavior (trait-like) observed in young adulthood. Given sex differences in adolescent alcohol consumption (Chen & Jacobson, 2012; Leatherdale & Burkhalter, 2012; Thompson et al., 2014), the shift from state-like to trait-like use may occur at different times or in different ways for males and females. Understanding if a shift occurs, when it occurs, and for whom, is important for the development and delivery of effective prevention programs that target problematic alcohol use before it becomes more ingrained and trait-like in nature.

Table 1 Sample demographics for each cohort.

Grades (years 1 to 3) Initial sample (N) % retained in year 2 % retained in year 3 Sex Male Female Ethnicity Aboriginal Non-aboriginal Family income Well above average Above average Average Below average Well below average Not reported a

Cohort 1

Cohort 2

Cohort 3

7 to 9 444 89.9% (N = 399) 82.0% (N = 364)

8 to 10 456 90.1% (N = 411) 78.9% (N = 360)

9 to 11 403 82.4% (N = 332) 73.0% (N = 294)

46.2% (N = 205) 53.8% (N = 239)

43.9% (N = 200) 56.1% (N = 256)

44.2% (N = 178) 55.8% (N = 225)

4.1% (N = 18) 95.9% (N = 426)

3.9% (N = 18) 96.1% (N = 438)

4.7% (N = 19) 95.0% (N = 383)a

8.3% (N = 35) 29.6% (N = 125) 55.9% (N = 236) 5.0% (N = 21) 1.2% (N = 5) 5.0% (N = 22)

7.7% (N = 33) 38.2% (N = 164) 47.3% (N = 203) 5.4% (N = 23) 1.4% (N = 6) 5.9% (N = 27)

8.6% (N = 34) 32.7% (N = 129) 47.8% (N = 189) 9.4% (N = 37) 1.5% (N = 6) 2.0% (N = 8)

Data missing for 1 case.

stable patterns of use (Van Der Vorst, Vermulst, Meeus, Deković, & Engels, 2009). We hypothesized females would show stronger traitlike variance than males early in adolescence, but that males would develop trait-like use later in adolescence as their patterns of use started to stabilize. 2. Method 2.1. Participants Our sample included three cohorts of students who participated in the PATH study, a three-year longitudinal study on adolescent risk behaviors spanning a five-year developmental period from grades 7–11. In the first year of data collection, the three cohorts included students from grade seven (n = 444), eight (n = 456), and nine (n = 403) enrolled in a large school district in western Canada. All students in the school district were invited to participate. Of the 1315 students who provided parental consent and student assent, 1303 completed at least 50% of study measures. Analyses indicate attrition of 21.9% by the third year of the study, with failure to complete predicted by aboriginal status, χ2(1, N = 1302) = 9.02, p = .003, and the absence of a father in the household, χ2(1, N = 1303) = 24.04, p b .001.1 Table 1 shows demographic information for participants in each cohort. Detailed sample information for the PATH study is described elsewhere (e.g. Fulton, Krank, & Stewart, 2012; Krank et al., 2011).

1.1. Objectives and hypotheses 2.2. Measures The purpose of this research was to test the overall pattern of state versus trait variance in alcohol use in this age group, change in statelike and trait-like variance over time during this developmental phase, and sex differences in how state-like and trait-like variance shift during adolescence. We used data from the Project on Adolescent Trajectories and Health (PATH), a cohort-sequential design spanning grades 7 to 11, and applied the variance decomposition approach described by Mushquash et al. (2014) to identify state and trait components of alcohol consumption. Based on Mushquash et al. (2014) and research showing rapid changes in drinking during adolescence (Chen & Jacobson, 2012) we hypothesized adolescent alcohol consumption would have a larger proportion of state-like variation compared to trait-like variation. We hypothesized the state-like component of use would predominate early in development, but trait-like components would emerge and increase in magnitude across adolescence as alcohol use became more stable (Mushquash et al., 2014). Males tend to have a fluctuating and escalating pattern of use compared to females, who show more frequent,

Alcohol consumption was measured using alcohol frequency, quantity, and volume. Frequency of alcohol consumption was measured with the question “If you drank alcohol in the past 30 days, how many days did you drink alcohol?” and typical alcohol quantity was then measured with the question “Think of a typical drinking situation. How many drinks would you normally have?” Both questions used an openended numerical response format (i.e., number of days; number of drinks). These single-item measures of alcohol consumption are commonly used and have shown acceptable reliability and validity in previous research (Bloomfield, Hope, & Kraus, 2013; Chung et al., 2012). Alcohol quantity and alcohol frequency were positively correlated

1 Analyses included sex, starting grade, aboriginal status, family income, foster care, mother's and father's education, presence of mother and father in the household, and presence of step-mother and step-father in the household. Bonferroni correction was used to control type 1 error (α = .05) across these 11 comparisons.

L.J. Nealis et al. / Addictive Behaviors 55 (2016) 25–31

27

Table 2 Means and standard deviations of alcohol use measures for each cohort over time. Cohort

1

2

3

Outcome

Volume Frequency Quantity Volume Frequency Quantity Volume Frequency Quantity

Grade 7

Grade 8

Grade 9

Grade 10

Grade 11

M

SD

M

SD

M

SD

M

SD

M

SD

0.15 0.28 0.51 – – – – – –

0.91 1.21 1.14 – – – – – –

2.44 1.62a 1.36 0.98 0.75a 1.46 – – –

11.84 4.37 3.12 3.35 1.84 2.78 – – –

2.12 1.63 2.49 2.73 2.05 2.51 2.04 1.55 2.36

8.84 3.42 3.22 6.40 4.07 3.57 6.34 3.40 2.60

– – – 4.01 2.23 3.91 2.99 2.01 3.14

– – – 10.10 3.94 4.24 6.28 3.18 3.62

– – – – – – 4.66 2.56 4.46

– – – – – – 9.39 4.38 3.98

Note: Superscript letters indicate significant differences (p b .05) in outcomes between cohorts during grades with overlapping cohorts (i.e., grades 8–10).

(r = .54, p b .001). As in previous research (e.g., Thompson et al., 2014), we also derived a measure of weekly alcohol volume, which was calculated as a product of the quantity and frequency measures. Consistent with Thompson et al. (2014), this value was divided by four to represent the average weekly volume of alcohol (in number of drinks) consumed over the past month. Alcohol volume shows a strong correlation (r = .77) with heavy episodic drinking in previous research with adolescents (Thompson et al., 2014).

2.3. Procedure The PATH study was approved by the ethics board of the Okanagan University College (now the University of British Columbia, Okanagan Campus). Participants completed a large battery of paper-and-pencil questionnaires in groups of 20–70 students during a one-hour period. Follow-up assessments were conducted 12-months and 24-months following completion of the initial questionnaires. Participants completed questionnaires in May to June of each year.

3. Results 3.1. Descriptive statistics and cohort linkage Table 2 shows means and standard deviations for each cohort at each measurement occasion. All drinking measures showed an increase across the three years of measurement for each cohort and an overall upward trend across the entire developmental period across cohorts. Pairwise comparisons were conducted for alcohol use measures between cohorts on overlapping years using independent-samples ttests. Bonferroni corrections were applied to control Type I error (α = .05) across five comparisons for each outcome measure. No significant differences were shown for volume or frequency measures. Only one comparison was significant for quantity, with higher quantity of use in Cohort 1 than Cohort 2 during grade 8, t(476.28) = 3.60, p b .001. Only minor differences in mean level of alcohol consumption across cohorts supports the use of a single developmental trajectory in our data. 3.2. Variance components across development

2.4. Data analysis Consistent with Mushquash et al. (2014), we used generalizability theory to estimate the proportion of trait-like and state-like variance in each alcohol use measure. Generalizability theory (or G-theory) extends classical test theory by estimating variance components beyond the true score and error term. We have adapted this framework to estimate the variance components associated with the person (i.e., the trait-like component), year (i.e., test-retest variance), cohort (i.e., between-cohort differences), and the person-by-year interaction (i.e., the state-like component). The cohort-by-year interaction was also included, as this is thought to indicate non-uniformity of trajectories across multiple overlapping cohorts (Miyazaki & Raudenbush, 2000). Variance components were estimated as mixed-effects models with restricted maximum likelihood estimation using the lme4 package in R (Bates, Maechler, Bolker, & Walker, 2014). All factors were treated as random effects. Significant differences between variance components are indicated by 95% confidence intervals, which were calculated from bootstrapped parameter estimates (500 repetitions) using the confint function in the lme4 package. Data were missing from 3.9% of cases (N = 135) for alcohol quantity and from 5.7% of cases (N = 197) for alcohol frequency and volume.2 Only complete cases were included in analyses.

2 Missing data is described for students who completed at least one study measure at all three measurement occasions (n = 955) and does not include missing data due to attrition.

Table 3 shows overall variance estimates. For weekly alcohol volume, the state-like component (person-by-year) was most predominant, representing 76.30% of the total variance. This suggests that for most youth, their pattern of alcohol use varies significantly from year to year. However, 19.32% of the total variance was trait-like, suggesting some consistency in alcohol use within individuals over time. When alcohol volume was decomposed into its constituent parts (quantity and frequency), a similar pattern emerged. Alcohol frequency showed a smaller proportion of state-like variance compared to alcohol volume, with 68.05% of total variance indicating year-to-year fluctuation and 26.32% of total variance indicating year-to-year stability. Alcohol quantity showed an even greater difference than alcohol volume, with only 46.65% of total variance indicating year-to-year fluctuation in use, and 35.56% of total variance indicating year-to-year stability. Nonetheless, all three alcohol measures showed a greater proportion of variance attributable to state than trait. Across all outcomes, components representing cohort differences (cohort) and non-uniformity between cohorts over time (cohort × year) did not differ significantly from zero, indicating negligible effects. There was a small effect of measurement year for alcohol quantity (10.47%), but not for frequency or volume. 3.3. Variance components by cohort To test changes in the proportion of trait-like and state-like variability across adolescence, we analyzed variance components for each cohort separately. The singular developmental trajectory across cohorts (i.e., minimal mean-level differences between cohorts and the nonsignificant cohort-by-year interaction) allows these differences to be

28

L.J. Nealis et al. / Addictive Behaviors 55 (2016) 25–31

Table 3 Variance components and proportion of total variance for alcohol use measures in the total sample. Variance component

Person (trait) Year Cohort Year × cohort Person × year (state) Total

Alcohol volume

Alcohol frequency

Alcohol quantity

S2

Proportion [CI 95%]

S2

Proportion [CI 95%]

S2

Proportion [CI 95%]

10.55 1.69 0.58 0.13 41.64 54.58

19.32 [15.23, 23.50] 3.09 [0.00, 11.92] 1.06 [0.00, 4.31] 0.23 [0.00, 1.06] 76.30 [71.65, 81.13]

3.21 0.48 0.17 0.04 8.30 12.19

26.32 [22.27, 30.93] 3.95 [0.00, 15.20] 1.36 [0.00, 4.88] 0.32 [0.00, 1.14] 68.05 [63.71, 72.27]

4.37 1.29 0.90 0.01 5.73 12.29

35.56 [30.88, 39.85] 10.47 [0.28, 43.37] 7.28 [0.00, 28.29] 0.04 [0.00, 0.38] 46.65 [43.71, 49.78]

S2 = variance estimate; proportion = proportion (%) of total variance within that cohort.

interpreted as developmental. Analyses involved person, year, and person-by-year interaction components, indicating the trait-like component, measurement error, and state-like component, respectively. Variance estimates (as proportions of total variance) are shown in Figs. 1–3. Alcohol volume and frequency showed similar patterns, with state-like variation predominating (74.94–83.76% of total variance) in Cohort 1 (grades 7 to 9) with minimal trait-like components (13.65–19.37% of total variance). By Cohort 3 (grades 9 to 11), traitlike components increased to represent 30.09–33.04% of the total variance and state-like components decreased to 65.07–66.89% of total variance. Across all cohorts, variance due to measurement year was negligible and not significantly different from zero. Alcohol quantity showed a similar pattern, but with greater convergence in state and trait variance. In Cohort 1, state-like and trait-like variance was 57.39% and 29.66%, respectively, showing a smaller discrepancy than for alcohol volume and frequency. By Cohort 3, trait-like variance increased to 49.62%, surpassing state-like variance, which declined to 40.43% in that group. In contrast to alcohol volume and frequency, variance due to measurement year showed a small but significant effect (i.e., 9.95–12.95%) for alcohol quantity.

volume and quantity, males showed similar patterns of change as cohort-level analyses, with relatively low trait-like components that increased steadily across cohorts (i.e., across development). Frequency showed a slightly different pattern where the trait-like component increased the most between Cohort 1 (11.31%) and Cohort 2 (30.33%), and then remained relatively unchanged for Cohort 3 (25.75%). Females showed a similar pattern of change across the three alcohol use measures to that shown by males, although females tended to show the greatest increases in trait-like variance between Cohort 2 and Cohort 3. Although variance components were largely similar across sexes, females showed higher initial trait-like variance in frequency (26.66%) compared to males (11.31%), with females showing much greater convergence between trait-like (43.20%) and state-like variance (54.80%) by Cohort 3. Overall, females show higher trait-like components than males across development, particularly for frequency of alcohol use, although increases in stability occur predominantly later in adolescence for females compared to relatively gradual increases for males.

3.4. Sex differences

This research extends previous work (e.g. Mushquash et al., 2014; Thompson et al., 2014) by exploring what proportion of variance in drinking is attributable to state- versus trait-like variance, and how these components change across adolescence.

To test sex differences in variance components across cohorts, we separated cohort-level estimates by sex (see Table 4). For alcohol

4. Discussion

Fig. 1. Variance components (proportion of total variance) of alcohol volume for each cohort. Error bars represent bootstrapped confidence intervals (95%). Components with a confidence interval that includes zero are indicated as non-significant (n.s.). Lower-case letters indicate patterns of significance across cohorts and across components.

L.J. Nealis et al. / Addictive Behaviors 55 (2016) 25–31

29

Fig. 2. Variance components (proportion of total variance) of alcohol frequency for each cohort. Error bars represent bootstrapped confidence intervals (95%). Components with a confidence interval that includes zero are indicated as non-significant (n.s.). Lower-case letters indicate patterns of significance across cohorts and across components.

As hypothesized, alcohol consumption in adolescence is largely influenced by situational factors from year-to-year (i.e., state-like), but also has a dispositional component (i.e., trait-like). The overall predominance of state-like variability was in contrast to the ratio of trait to state variance (57% vs. 43%) in emerging adults reported by Mushquash et al. (2014). This finding supports the view that adolescence is a period of rapid change where patterns of alcohol use are still in flux and largely

influenced by situational factors that vary across time. Interestingly, the dispositional aspect of alcohol use was stronger for quantity and frequency of drinking, rather than the number of weekly drinks (i.e., volume). This may reflect increased total variance resulting from the derivation of volume using two imperfectly correlated components (i.e., quantity and frequency). Overall, the emergence of strong trait-like use may reflect the development of habitual drinking patterns into late

Fig. 3. Variance components (proportion of total variance) of alcohol quantity for each cohort. Error bars represent bootstrapped confidence intervals (95%). Components with a confidence interval that includes zero are indicated as non-significant (n.s.). Lower-case letters indicate patterns of significance across cohorts and across components.

30

L.J. Nealis et al. / Addictive Behaviors 55 (2016) 25–31

Table 4 Variance estimates and proportion of total variance for alcohol use measures by sex and cohort. Sex

Male

Cohort

1

2

3

Female

1

2

3

Variance component

Person (trait) Year Person × year (state) Total Person (trait) Year Person × year (state) Total Person (trait) Year Person × year (state) Total Person (trait) Year Person × year (state) Total Person (trait) Year Person × year (state) Total Person (trait) Year Person × year (state) Total

Alcohol volume

Alcohol frequency

Alcohol quantity

S2

Proportion [CI 95%]

S2

Proportion [CI 95%]

S2

Proportion [CI 95%]

10.95 2.87 89.41 103.24 9.28 3.38 45.99 58.65 17.56 1.97 48.15 67.67 3.83 1.35 15.12 20.29 5.89 1.79 34.66 42.34 15.97 1.44 28.81 46.22

10.61 [1.20, 21.36] 2.78 [0.00, 13.00] 86.61 [73.15, 99.78]

1.35 0.69 9.89 11.93 4.45 0.83 9.40 14.68 4.62 0.35 12.98 17.95 2.69 0.57 6.82 10.07 2.02 0.53 7.02 9.57 4.38 0.20 5.56 10.15

11.31 [1.35, 23.29] 5.79 [0.00, 22.00] 82.89 [71.62, 96.37]

2.73 0.93 5.71 9.37 5.37 2.02 6.76 14.15 7.59 2.27 6.48 16.34 1.86 1.06 3.42 6.34 4.46 1.32 8.13 13.91 4.96 0.66 3.82 9.44

29.15 [19.00, 41.27] 9.94 [0.00, 32.73] 60.90 [51.89, 71.25]

15.83 [6.06, 26.31] 5.76 [0.00, 20.67] 78.41 [66.42, 91.15] 25.95 [14.53, 39.83] 2.90 [0.00, 13.04] 71.15 [60.27, 85.52] 18.86 [9.65, 28.27] 6.64 [0.00, 27.87] 74.50 [64.25, 85.47] 13.91 [4.60, 23.04] 4.22 [0.00, 15.32] 81.87 [70.46, 94.33] 34.55 [22.60, 46.26] 3.12 [0.00, 13.37] 62.32 [53.22, 71.69]

30.33 [19.15, 42.67] 5.66 [0.00, 20.34] 64.01 [53.96, 73.22] 25.75 [14.08, 39.23] 1.94 [0.00, 8.30] 72.31 [60.79, 85.42] 26.66 [17.71, 37.85] 5.66 [0.00, 22.07] 67.68 [58.54, 78.24] 21.15 [13.02, 32.26] 5.54 [0.00, 21.04] 73.31 [63.34, 84.15] 43.20 [31.58, 55.88] 2.00 [0.00, 7.89] 54.80 [46.48, 64.57]

37.92 [27.10, 49.59] 14.30 [0.00, 56.17] 47.78 [40.26, 55.26] 46.43 [34.39, 60.75] 13.91 [0.00, 50.11] 39.66 [33.52, 46.04] 29.31 [20.13, 39.10] 16.72 [0.15, 65.52] 53.97 [46.53, 61.66] 32.04 [22.63, 41.77] 7.03 [0.00, 32.43] 58.48 [50.88, 66.91] 52.50 [40.35, 64.97] 7.03 [0.00, 27.29] 40.47 [34.58, 46.07]

S2 = variance estimate; % = proportion of total variance within that cohort.

adolescence, such that alcohol use (particularly the quantity of alcohol consumed per occasion) involves fewer explicit choices and becomes more implicit, reflexive, and automatic over time. Hypotheses regarding change in trait and state variance components from grades 7 to 11 were also supported, with the three alcohol use measures showing an increasing dispositional component across adolescence. Volume showed a steady shift from predominantly state-like in Cohort 1 to approximately 30% trait-like in Cohort 3. After separating volume into its constituent parts, frequency showed a similar pattern as volume (albeit with slightly higher trait-components across development), and alcohol quantity showed higher initial trait-like variability that eventually surpassed the state-like component to closely resemble the trait-state shown by Mushquash et al. (2014) in university students. These results suggest frequency of drinking occasions are still largely in flux during adolescence, and does not shift to dispositional influences to the same degree as other facets of alcohol consumption during the same period. Interestingly, both alcohol frequency and volume are highly related to problematic drinking in adolescence (Chung et al., 2012; Thompson et al., 2014). In contrast, the amount adolescents drink on each drinking occasion does seem to stabilize across development to become much more dispositional, which is consistent with the pattern shown with university students. Hypotheses regarding sex differences were supported in some cases, but not all. Results showed the expected pattern of higher dispositional influence on drinking frequency for females relative to males initially (Cohort 1) and throughout the developmental period. This is noteworthy given alcohol frequency is an indicator of problematic alcohol use in adolescents (Chung et al., 2012). One possible explanation for this trend is that self-regulation skills develop sooner for females than males (Raffaelli, Crockett, & Shen, 2005), which makes then better able to resist situational or contextual factors that would encourage more frequent alcohol use. Contrary to hypotheses, trait-like aspects of weekly alcohol volume and typical quantity per drinking occasion actually increased later for females relative to males, with this increased dispositional influence not becoming apparent until later in adolescence. This may coincide with previous research showing males tend to initiate alcohol use earlier than females (Buu et al., 2014), which tends to escalate into more stable but problematic patterns of heavy use compared to the patterns of light drinking more typical of females

(Van Der Vorst et al., 2009). These hypotheses remain speculative, however, and require further study. 4.1. Limitations and future directions We used a large sample of students measured annually in a cohortsequential design. Although this allows estimation of year-to-year stability in drinking, this measurement spacing is not adequate to capture trait- and state-like variability across weeks and months. Future research should use more frequent measurement occasions. We used students in grades 7 to 11, which includes the transition from middle school to high school but neglects other potentially important transitions (e.g., to university or the workplace). Bridging this gap will be important to understanding the development of trait-states observed by Mushquash et al. (2014) in emerging adults. We used a descriptive approach to identify normative trends, which is not ideally suited to testing predictors of change. For example, contextual factors (e.g., peer influence, transition to high school) may influence trait and state aspects of drinking differentially across development. This remains an important area for future research. Future research should also test state and trait components of alcohol use in different subgroups of adolescents (e.g., heavy chronic users versus low stable users), which could yield useful information regarding ways to improve prevention programs for those most at risk of problematic drinking. 4.2. Conclusions Research on normative development in alcohol use has largely focused on average rates of change over time. This focus, however, neglects the difference between situational and dispositional aspects of behavior. Both aspects have potent implications for adolescent alcohol use, with situational influences being potentially more malleable and susceptible to intervention. Our research suggests alcohol use in early adolescence involves predominantly situational influences with dispositional influences becoming stronger as adolescents mature. These findings highlight the importance of early intervention while problematic behaviors are more malleable. It may also be more effective to target aspects of use that are more conducive to situational influence (e.g., frequency), rather than aspects of use that are more dispositional

L.J. Nealis et al. / Addictive Behaviors 55 (2016) 25–31

(e.g., quantity). These findings also suggest early prevention is particularly important for males, who show greater sensitivity to contextual influences that encourage more frequent alcohol use compared to females. Consideration of state and trait aspects of behavior could lead to more effective public health initiatives to help mitigate the harms associated with adolescent substance use. Role of funding sources The Project on Adolescent Trajectories and Health (PATH) was funded through a Society Culture and Health grant from the Social Sciences and Humanities Research Council (SSHRC) of Canada and the Canadian Institutes of Health Research (CIHR; M. D. Krank). L. J. Nealis was supported by a SSHRC Canada Graduate Scholarship, and K. Thompson's postdoctoral fellowship was funded through an operating grant from the Movember Foundation. S. H. Stewart's work on this project funded through an operating grant from SSHRC. Funding sources had no role in the study design, collection, analysis or interpretation of the data, writing the manuscript, or the decision to submit the paper for publication. Contributors Marvin Krank designed the study and oversaw data collection. Logan Nealis conducted the literature searches, conducted the statistical analyses and wrote the first draft of the manuscript with supervision from K. Thompson and S. Stewart. All authors contributed to and have approved the final manuscript. Conflict of interest All authors declare that they have no conflicts of interest. Acknowledgments We gratefully acknowledge participants in the longitudinal data collection that provided the data for this article: Anne-Marie Wall and Daniel Lai, who were co-investigators; Peter Molloy, and Christine Wekerle, who were collaborators on the PATH study; and Tricia Johnson and Aarin Frigon, who served as project coordinators for the study. We thank Abby Goldstein, Jana Atkins, Jessica van Exan, Jonathan Brown, Tara Schoenfeld, Rob Callaway, Tabatha Freimuth, Adrienne Girling, and Pam Collins for their valued research assistance.

References Bates, D., Maechler, M., Bolker, B., & Walker, S. (2014). lme4: Linear mixed-effects models using Eigen and S4. R package (version 1.1-7). http://CRAN.R-project.org/package= lme4 Bloomfield, K., Hope, A., & Kraus, L. (2013). Alcohol survey measures for Europe: A literature review. Drugs: Education, Prevention & Policy, 20, 348–360. Buu, A., Dabrowska, A., Mygrants, M., Puttler, L.I., Jester, J.M., & Zucker, R.A. (2014). Gender differences in the developmental risk of onset of alcohol, nicotine, and marijuana use and the effects of nicotine and marijuana use on alcohol outcomes. Journal of Studies on Alcohol and Drugs, 75, 850–858. Chen, P., & Jacobson, K.C. (2012). Developmental trajectories of substance use from early adolescence to young adulthood: Gender and racial/ethnic differences. Journal of Adolescent Health, 50, 154–163. http://dx.doi.org/10.1016/j.jadohealth.2011.05.013. Chung, T., Smith, G.T., Donovan, J.E., Windle, M., Faden, V.B., Chen, C.M., & Martin, C.S. (2012). Drinking frequency as a brief screen for adolescent alcohol problems. Pediatrics, 129, 205–212. http://dx.doi.org/10.1542/peds.2011-1828.

31

Duncan, S.C., Alpert, A., Duncan, T.E., & Hops, H. (1997). Adolescent alcohol use development and young adult outcomes. Drug and Alcohol Dependence, 49, 39–48. http://dx. doi.org/10.1016/S0376-8716(97)00137-3. Duncan, S.C., Duncan, T.E., & Strycker, L.A. (2006). Alcohol use from ages 9 to 16: A cohortsequential latent growth model. Drug and Alcohol Dependence, 81, 71–81. http://dx. doi.org/10.1016/j.drugalcdep.2005.06.001. Fulton, H.G., Krank, M.D., & Stewart, S.H. (2012). Outcome expectancy liking: A self-generated, self-coded measure predicts adolescent substance use trajectories. Psychology of Addictive Behaviors, 26, 870–879. http://dx.doi.org/10.1037/a0030354. Hanson, K.L., Medina, K.L., Padula, C.B., Tapert, S.F., & Brown, S.A. (2011). Impact of adolescent alcohol and drug use on neuropsychological functioning in young adulthood: 10-year outcomes. Journal of Child & Adolescent Substance Abuse, 20(2), 135–154. http://dx.doi.org/10.1080/1067828X.2011.555272. Heron, J., Macleod, J., Munafò, M.R., Melotti, R., Lewis, G., Tilling, K., & Hickman, M. (2012). Patterns of alcohol use in early adolescence predict problem use at age 16. Alcohol and Alcoholism, 47, 169–177. http://dx.doi.org/10.1093/alcalc/agr156. Irons, D.E., Iacono, W.G., & McGue, M. (2015). Tests of the effects of adolescent early alcohol exposures on adult outcomes. Addiction, 110, 269–278. http://dx.doi.org/10.1111/ add.12747. Johnston, L.D., O'Malley, P.M., Bachman, J.G., & Schulenberg, J.E. (2011). Monitoring the future: National survey results on drug use, 1975–2010: Volume I, secondary school students. Ann Arbor, MI: Institute for Social Research, University of Michigan. Koskinen, S.M., Ahveninen, J., Kujala, T., Kaprio, J., O'Donnell, B.F., Osipova, D., & Rose, R.J. (2011). A longitudinal twin study of effects of adolescent alcohol abuse on the neurophysiology of attention and orienting. Alcoholism: Clinical and Experimental Research, 35, 1339–1350. http://dx.doi.org/10.1111/j.1530-0277.2011.01470.x. Krank, M., Stewart, S.H., O'Connor, R., Woicik, P.B., Wall, A., & Conrod, P.J. (2011). Structural, concurrent, and predictive validity of the Substance Use Risk Profile Scale in early adolescence. Addictive Behaviors, 36, 37–46. http://dx.doi.org/10.1016/j.addbeh.2010. 08.010. Leatherdale, S.T., & Burkhalter, R. (2012). The substance use profile of Canadian youth: Exploring the prevalence of alcohol, drug and tobacco use by gender and grade. Addictive Behaviors, 37, 318–322. http://dx.doi.org/10.1016/j.addbeh.2011.10.007. Liang, W., & Chikritzhs, T. (2015). Age at first use of alcohol predicts the risk of heavy alcohol use in early adulthood: A longitudinal study in the United States. The International Journal on Drug Policy, 26, 131–134. http://dx.doi.org/10.1016/j.drugpo. 2014.07.001. Miyazaki, Y., & Raudenbush, S.W. (2000). Tests for linkage of multiple cohorts in an accelerated longitudinal design. Psychological Methods, 5, 44–63. http://dx.doi.org/10. 1037/1082-989X.5.1.44. Mushquash, A.R., Sherry, S.B., Mackinnon, S.P., Mushquash, C.J., & Stewart, S.H. (2014). Heavy episodic drinking is a trait-state: A cautionary note. Substance Abuse, 35, 222–225. http://dx.doi.org/10.1080/08897077.2013.876486. Raffaelli, M., Crockett, L.J., & Shen, Y.L. (2005). Developmental stability and change in selfregulation from childhood to adolescence. Journal of Genetic Psychology, 166, 54–75. Stueve, A., & O'Donnell, L.N. (2005). Early alcohol initiation and subsequent sexual and alcohol risk behaviors among urban youths. American Journal of Public Health, 95, 887–893. http://dx.doi.org/10.2105/AJPH.2003.026567. Thompson, K., Stockwell, T., Leadbeater, B., & Homel, J. (2014). Association among different measures of alcohol use across adolescence and emerging adulthood. Addiction, 109, 894–903. http://dx.doi.org/10.1111/add.12499. Van Der Vorst, H., Vermulst, A.A., Meeus, W.J., Deković, M., & Engels, R.E. (2009). Identification and prediction of drinking trajectories in early and mid-adolescence. Journal of Clinical Child and Adolescent Psychology, 38, 329–341. http://dx.doi.org/10.1080/ 15374410902851648. Young, M.M., Saewyc, E., Boak, A., Jahrig, J., Anderson, B., Doiron, Y., ... Clark, H. (2011). Cross-Canada report on student alcohol and drug use: Technical report. Ottawa, ON: Canadian Centre on Substance Abuse.