Effects of Excessive Alcohol Use on Antisocial Behavior Across Adolescence and Early Adulthood

Objective . Antisocial behavior (ASB) declines with age in the majority of the population; however, excessive alcohol use may inhibit the desistance process. We hypothesize that excessive early drinking will slow a young person’s overall pattern of crime desistance in comparison to that of others (“between-person effects”), and that short-term increases in alcohol consumption will result in short-term increases in ASB (“within-person effects”). Method. Frequency of ASB and typical alcohol consumption were assessed repeatedly with young people from ages 15 to 21 years in a population-based birth cohort (Avon Longitudinal Study of Parents and Chilren). Longitudinal trajectories showed ASB decreasing and alcohol use increasing across adolescence, with both stabilising in adulthood. The parallel growth model was re-parameterized to simultaneously estimate the person-specific (or “between-person”) and time-specific (or “within-person”) influences of alcohol on ASB. Results. Typical alcohol consumption by young people aged 15 years was positively associated with ASB both cross-sectionally and into young adulthood (i.e. there were “between-person” effects of initial levels of alcohol consumption on both initial (b[SE]=1.64[0.21]; p <.001) and final levels of ASB (b[SE]=0.53[0.14]; p <.001). Within-person effects were also identified in early adulthood (b[SE]=0.06[0.02]; p =.001), showing that when a young person reported consuming more alcohol than normal across the past year, they also reported engaging in higher than their usual levels of ASB. Conclusion. The results are consistent with both between- and within-person effects of excessive alcohol use on ASB desistence. Future research should further investigate this relationship by investigating pathways into excessive alcohol use and ASB in adolescence.


INTRODUCTION
Antisocial behavior (ASB) is a major public policy and health concern 1 ; it not only places a large financial burden on society, 1,2 but is also associated with increased risk of negative outcomes including criminal behavior 3,4 and mental health disorders. 4 In addition to prevention strategies prior to ASB onset, key targets for intervention may also arise later in development. The age-crime curve, for example, consistently shows that ASB peaks in midadolescence and then declines throughout late adolescence and early adulthood. 3,4 However, there is evidence for individual differences in the course of ASB across this time period, 4 and identifying factors associated with desistance is important to guide post-onset interventions. 5 There are ways in which an exposure such as excessive alcohol use might promote ASB, limiting the reduction typically seen through late adolescence. 4,6 First, excessive alcohol use may slow a young person's overall pattern of crime desistance in comparison to that of others (known as "between-person effects"). Between-person effects provide evidence for who is at risk and can be tested with covariates that are present either before or when ASB begins to decrease. Support for this hypothesis has mainly come from studies examining between-person differences in the long-term course of ASB predicted by baseline levels of alcohol use. [6][7][8] Results generally show that baseline alcohol use is associated with ASB crosssectionally; however, findings regarding the effect of alcohol on ASB desistance have been more mixed. [6][7][8][9] Second, alcohol use may affect desistance from ASB through a series of short-term, time-specific influences [6][7][8]10,11 (known as "within-person effects"). In contrast to between-person effects, within-person effects focus on when a person is at risk. 6 In the current study, we aim to contribute to the literature in two ways. First, few studies have examined these person-specific (or between-person) and time-specific (or within-person) influences in the same model. Recent work by Curran et al. has detailed advances in analytical approaches that allow the between-and within-person effects to be M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 3 disaggregated. [12][13][14] Additionally, prior studies that have examined within-person effects have generally treated alcohol use as a time-varying covariate, rather than modelling a longitudinal trajectory; therefore, the time-varying measures of alcohol confound variance due to the adolescents' typical trajectory with that due to time-specific deflations in that trajectory.
Modelling the characteristic (and differing) trajectories for both ASB and alcohol use across adolescence is not only important to obtain reliable estimates for the within-and betweenperson effects, 12 but also to allow the effect of drinking more alcohol than usual on short-term increases in ASB to be examined. Second, findings to date are mainly based on a selection of small, specific samples, such as male offenders, 7 those in treatment for substance use, 11 and single sex samples, 6,8,10,15 or focus on specific outcomes such as dating aggression 16 and psychopathic features 17 rather than ASB more generally.
The present investigation expands on the extant literature in its use of a large, prospective population cohort to examine whether excessive alcohol use acts as a snare, and reduces the rate of decline in ASB across young adulthood. The specific aims are to: 1) examine changing patterns of typical alcohol consumption and ASB in tandem across adolescence and early adulthood, 2) examine the between-person effects of typical alcohol consumption in mid-adolescence on the course of ASB into young adulthood, and 3) investigate the within-person, time-specific effects of alcohol consumption on ASB. It is hypothesized that there will be both between-and within-person effects of alcohol consumption on the desistence of ASB.

Sample
Data were utilised from a large UK birth cohort, the Avon Longitudinal Study of Parents and Children (ALSPAC), which was set up to examine genetic and environmental determinants of health and development. 18 The "core" enrolled sample consisted of 14,541 women residing in the former county of Avon, UK, who had an expected date of   delivery between 1 st April 1991 and 31 st December 1992. Of the 13,988 offspring alive at age one year, a small number of participants withdrew consent (n = 24). The sample was also restricted to singletons or first-born twins, leaving a starting sample of 13,775. Parents and children have been followed up regularly since recruitment via questionnaire and clinic assessments. Further details on the sample characteristics and methodology have been described previously. 18,19 Detailed information about ALSPAC and a data dictionary can be found on the study website (http://www.bristol.ac.uk/alspac) and at http://www.bris.ac.uk/alspac/researchers/data-access/data-dictionary. Ethical approval for the study was obtained from the ALSPAC Ethics and Law committee and the local research ethics committees.

Measures
A timeline for data collection is shown in Figure S1, available online.
Antisocial Behavior (ASB). A self-report questionnaire asking about antisocial acts committed in the past year 20 was completed by the young person at four time points between ages 15 and 21 years. At age ~15 years (mean=15 years 6 months, SD=4 months) and ~18 years (mean=17 years 10 months, SD=5 months) data were collected during a computerbased session at a focus clinic and at ages ~19 years (mean=18 years 8 months, SD=6 months) and ~21 years (mean=20 years 11 months, SD=6 months), data were collected via online or postal questionnaire. Eight ASB items were consistent across all time points (stole from shops, broke into a vehicle or building, stole from person, damaged property, assault, carried a weapon, rowdy in a public place, hurt animals). For each item, respondents were asked: "how often in the last year have you…" with responses classified into three categories: "not at all," "just once," and "2 or more times." All items were then combined to create a sum score representing frequency of antisocial acts committed in past year at each time point (range 0-16). External validity for this self-report questionnaire has been examined previously using cross-checks with agency records and teachers' questionnaires. 21 Alcohol Consumption. At each time-point (15,18,19, and 21 years) respondents were asked to report the number of units of alcohol they consume on a day when they have a drink over the past year, with responses classified into five categories (scored on a scale of 0-5): "none," "1 or 2," "3 or 4," "5 or 6," "7 to 9," and "10 or more." Sensitivity analyses were performed using a measure of alcohol frequency. Respondents were asked to report how often in the past year they had a drink containing alcohol, with responses classified into four cateogories: "never," "occasional," "weekly," and "daily or almost daily." Covariates. Maternal questionnaires completed during pregnancy were used to assess housing tenure (owned/mortgaged; privately rented; subsidised housing rented from council/ housing association), maternal level of education (no high school qualifications; high school only; beyond high school), parity (study child 1 st , 2 nd , 3 rd or subsequent child born in family) and crowding (up to one person per room in house; more than one person per room). These sociodemographic variables were included in all analyses primarily to aid in addressing potential bias from missing data; however, they may also be confounders of the betweenperson effect of alcohol consumption on ASB.
Developmental trajectories of conduct problems between age 4 and 13 years, exposure to antisocial peers at ~age 11 years, and parental crime and problematic alcohol use from the child's birth to 11 years were also included as potential confounders in secondary analyses. Details of the assessment of these confounders is provided in Supplement 1, available online.

Data Analysis
Parallel Growth Model for Typical Alcohol Consumption and ASB. Longitudinal trajectories for typical alcohol consumption and ASB were derived using a parallel M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 6 exponential growth model. These growth curves were specifically selected based on a combination of exploring the shape of the population mean change for both constructs along with selecting a theoretically justifiable functional form (see Supplement 2, available online).
In the traditional exponential growth model, three growth factors are estimated: the intercept, rate, and asymptote. The intercept (when fixed at baseline) is the average predicted starting point or initial level, and the asymptote is a line that the curve approaches as it heads towards infinity, or the average predicted final level. The rate represents the manner in which the asymptote is approached. In the current study, the model was re-parameterized to estimate the "half-life" instead of the rate. The half-life, measured in years, is the time by which 50% of a person's total change has been observed. It is therefore not only more interpretable than the rate, but can also easily be compared across measures with different scales. Additionally, the half-life is of greater interest for examining desistance from ASB, given that it provides an indication of the time taken for a person to desist. In an exponential growth model, the factor loadings are a function of the estimable parameters, and the loading for the final repeated observed measure on the asymptote indicates the total change to the asymptotic level that is gained by the end of the study. 22 This factor loading is important to consider given that extrapolation far beyond the period of observation may result in an asymptote that is poorly estimated. Figure 1 shows example estimated exponential decay trajectories that differ on the intercept ( Figure 1A), the half-life ( Figure 1B), and the asymptote ( Figure 1C). Further information on the exponential model is given in Supplement 2, available online.
Addressing Age Variability. Age variability within wave is common in longitudinal studies, and there is a range of methods for incorporating this age variation into the growth model. In the current study, we opted to preserve some but not all age variation by dividing respondents at each wave into younger and older age groups and treating these groups as two Between-Person Effects. First, an unconditional parallel exponential growth model for ASB and typical alcohol consumption was estimated. There are a number of different between-person effects that can be estimated with a parallel growth model; however, the focus here was the effect of baseline alcohol use on the ASB trajectory. Therefore, the ASB growth factors (intercept, half-life, and asymptote) were regressed on the latent intercept for alcohol consumption. The model was then re-parameterized in order to examine the effect of the alcohol intercept on the ASB growth factors, after accounting for the effect of the ASB intercept on the ASB half-life and asymptote. This was necessary given that the relationship between initial alcohol levels and the rate of decrease in ASB is likely to be dependent on a person's initial ASB level. Additionally, although initial alcohol levels affect final ASB levels both directly and indirectly, via the effect on initial ASB levels, it is the direct effect that is most relevant for identifying factors related to ASB desistance.
Within-Person Effects. Within-person effects of typical alcohol consumption on ASB were examined by regressing the observed repeated ASB measure (net the underlying ASB trajectory) on the time-specific residual for repeated alcohol consumption. The time-specific residual represents the deviation between the observed repeated alcohol measure and the underlying alcohol trajectory (see Figure S2, available online). If the repeated ASB measures are regressed directly on the observed alcohol measures, instead of on the alcohol residuals, the observed alcohol measures serve as mediators for the influence of the alcohol growth factors on the repeated ASB measures, meaning that the between-person effects will be altered with the inclusion of the within-person effects. By specifying the time-specific M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 8 residuals for alcohol use and using these residuals in the time-specific regressions, this mediated pathway is interrupted, therefore allowing the between-and within-person effects to be estimated simultaneously. 14 The repeated ASB measure was regressed on the time-specific residual for alcohol at the same assessment, as it was hypothesized that alcohol consumption would have a proximal rather than a lagged effect on ASB as has been found previously. 6 The final model is shown in Figure 2.
Model Fit. In the parallel growth model, an additional parameter was included to allow the trajectory functions to absorb artefactual differences between clinic and questionnaire data collection that may be due to the respondents' tendency to more readily report antisocial acts or alcohol use with the privacy of a questionnaire assessment completed at home. We note that this is equivalent to including an assessment-technique (clinic/questionnaire) dummy variable as a fixed effect in the multilevel modelling formulation of a latent growth model. The time-specific residual (co)variances for repeated measures (and therefore, the within-person effects) were also permitted to be heteroscedastic between but not within assessment-technique. Subsequently, fit for each trajectory was evaluated by examining residuals for the mean and covariance structure, and the final model variables, related to missing data. Young people that had complete data for both ASB and typical alcohol consumption from at least one of four time points were included in the trajectory analyses (n = 6,699). The inclusion of socio-demographic confounders resulted in a sample size of N = 6,112 (2,772 males, 3,340 females). Figure S3 (available online) shows a flow chart of retention in ALSPAC. Sensitivity analyses were performed using inverse probability weighting (IPW). 24 Further information on the IPW analyses is given in Supplement 3, available online.

Means and variances for observed repeated measures of typical alcohol consumption
and ASB for males and females are shown in Table S1, available online.

Parallel Growth Model for Typical Alcohol Consumption and ASB
Estimated and observed means for the parallel growth model are shown in Figure 3 with means, variances, and correlations between growth factors given in Supplement 4 and Table S2, available online. The alcohol trajectory started at an average of 1.2 (standard error [SE] = 0.04) at age 15 years 3 months (a score of 1 is equivalent to drinking "1 or 2" units of alcohol on a typical day when drinking) and increased to an average of 2.9 (SE = 0.03) (a score of 3 is equivalent to drinking "5 or 6" units of alcohol on a typical day when drinking).
The factor loading for the final repeated alcohol measure on the asymptote suggested that  Unadjusted results were very similar and therefore not shown.

Within-Person Effects
There was evidence that within-person increases in alcohol consumption were  Table 1. Model fit statistics indicated a good fit to the data (RMSEA = 0.02; CFI = 0.92).

Sensitivity Analyses
The final model (as shown in Figure 2) was rerun in order to perform a number of sensitivity checks. First, analyses were adjusted for childhood conduct problems, exposure to antisocial peers, and parental crime and problematic alcohol use in addition to sociodemographic factors (Table 1; Sensitivity test 1). Second, the impact of missing data was further assessed by using inverse probability weighting (  Table 1 shows that conclusions were similar across all sensitivity analyses, with the exception of weaker evidence for within-person effects of frequency of alcohol consumption on ASB. Finally, although the theoretical justification and a priori hypothesis for this study was that excessive alcohol consumption affects ASB, it is possible that effects are also present in the opposite direction. Therefore, the model was re-parameterized in order to examine the effect of the ASB intercept on the alcohol growth factors, after accounting for the effect of the alcohol intercept on the alcohol half-life and asymptote ( Figure S4, available online).
Those with higher initial ASB levels also had higher initial levels

DISCUSSION
In this UK population-based sample, there was evidence for both between-and within-person effects of alcohol consumption acting against desistance from ASB. That is, those who reported higher alcohol use compared to their peers in mid-adolescence also reported higher levels of ASB both cross-sectionally and in early adulthood, although M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 12 excessive alcohol use at the start of the study did not appear to affect the rate of decrease in ASB across adolescence. Additionally, within those time periods when young people reported consuming more alcohol than normal, they also reported engaging in more antisocial activities than would be expected given their overall pattern of ASB throughout adolescence and young adulthood. These time-specific effects were present in young adulthood, but not adolescence.
The results need to be interpreted in the context of several limitations. First, as with most cohort studies, there was selective attrition over time. Comparatively few cohort members provided data on all measures across adolescence and early adulthood. However, all analyses were performed using FIML estimation, which allowed over 6,000 participants to be included, and the inclusion of auxiliary variables related to missing data or using IPW made little difference to the results. Second, typical alcohol consumption and ASB were assessed using both questionnaires completed by respondents at home and computer-based sessions during focus clinics. To address any artefactual differences across assessment techniques, we incorporated this information into the derivation of the trajectories; however, the possibility remains that the lack of within-person effect of alcohol on ASB during adolescence is related to the data during this period being collected via a computer-based session rather than postal survey.
Third, it is possible that the between-person association between alcohol use and ASB is spurious with common risk factors not examined here (such as shared genetic risk) causing both. 25 However, accounting for earlier behavioural problems, exposure to antisocial peers, and parental crime and problematic alcohol use had little impact on the between-person effects. Fourth, the effect size found for the within-person effects in early adulthood was small. However, this is because each individual serves as their own control; therefore, all time-stable confounds that impact analyses examining between-person effects are M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 13 eliminated 6 ; additionally, within-person effects are the associations that remain even after accounting for the underlying growth trajectories for alcohol and ASB. Finally, given the complexity of the models, there was insufficient power to examine differences in effect between men and women separately. However, previous research using the same sample has shown little evidence that associations between alcohol and ASB differ by gender. 26,27 Additionally, the ASB items assessed in this study are rather "male-centric," and, in general, will only capture quite overt or confrontational behaviour, although the prevalence of shoplifting and being rowdy was similar in males and females.
In the current study, we simultaneously estimate the between-and within-person effects of alcohol on ASB across adolescence and early adulthood using a population-based sample. Recent studies have begun to use similar techniques to tease apart the effects of heavy alcohol (or substance) use on dating aggression, 16 psychopathic features, 17 and crime levels in young adults in substance use treatment. 11 The importance of disaggregating these effects has been highlighted in order to provide a more comprehensive investigation of the hypothesized developmental processes underlying the relationship between behaviors that change together over time. 14 Between-person effects tell us who is at risk, whereas withinperson effects tell us when a person is at risk; therefore, both provide unique information about the aetiology of the association between alcohol and ASB and have important but different implications for theory and clinical practice. [12][13][14] Consistent with previous research, [6][7][8]11,17 between-person effects of baseline alcohol use on the course of ASB were identified, with higher typical alcohol consumption in midadolescence increasing both initial and final levels of ASB. There was also a counterintuitive negative association between the alcohol intercept and the ASB half-life, indicating that higher alcohol consumption is associated with a faster decrease in ASB; however, this association disappeared after accounting for initial ASB levels. The negative association M A N U S C R I P T A C C E P T E D ACCEPTED MANUSCRIPT 14 between baseline alcohol use and change in ASB over time has been reported previously, when initial levels of ASB are not taken into account. 6,8 These findings highlight the importance of exploring this association in greater detail by considering the effect of initial levels of ASB 9 and investigating the effects of alcohol use on final levels of ASB, in addition to the rate of decrease. 6 The ability to estimate the effect of alcohol use in mid-adolescence on the long-term course of ASB into adulthood is a key strength of this study. Using an exponential growth model not only allowed more complex and theoretically relevant longitudinal change to be examined, but also enabled estimation of growth parameters that were of greater interest for research questions focused on ASB desistance (for further detail see Supplement 2, available online). Although typical alcohol consumption in midadolescence had no effect on the rate of decrease of ASB, it was associated with final ASB levels in adulthood, even after accounting for ASB in mid-adolescence, indicating that excessive early alcohol use does have a long-term effect on ASB desistance.
The within-person effects identified in the current study were specific to young adulthood, and not present during adolescence. This finding supports previous research that has identified within-person effects across young adulthood. [6][7][8]17 Previous findings regarding within-person effects in adolescence have been inconsistent, with one study finding stronger effects of alcohol on dating aggression in adolescence compared with early adulthood, 16    Note: Unstandardized coefficient (standard error) shown; all results adjusted for sex, housing tenure, maternal education, parity, and household crowding. "Original model" shows estimates presented in manuscript from final analysis model (model shown in Figure 1); "Sensitivity test 1" shows estimates after additionally adjusting for childhood conduct problems, antisocial peers, and parental crime and alcoholism (n = 4,465 for these analyses); "Sensitivity test 2" shows weighted estimates (from inverse probability weighting analyses); "Sensitivity test 3" shows estimates using the total number of types of crime committed in last year instead of frequency of crimes committed; "Sensitivity test 4" shows estimates using alcohol frequency instead of typical consumption. a Parameter for clinic measures constrained to equality. b Parameter for questionnaire measures constrained to equality.  Note: a = paths testing between-person effects; Asymp = asymptote; b = paths testing withinperson effects; Int = intercept.