Childhood and adolescent predictors of alcohol use and problems in adolescence and adulthood in the National Child Development Study
Jennifer Maggs, S-110-D Henderson South, Human Development and Family Studies, Pennsylvania State University, University Park, PA 16802, USA. E-mail: email@example.com
Aims To identify childhood and adolescent predictors of alcohol use and harmful drinking in adolescence and adulthood.
Design Longitudinal data from childhood to mid-life from the National Child Development Study (NCDS) were used, including predictors collected at ages 7, 11, 16 years and alcohol outcomes collected at ages 16, 23, 33 and 42 years.
Setting The NCDS is an ongoing longitudinal study of a cohort of 1 week's births in Britain in 1958.
Participants Childhood and adolescent predictors and alcohol use data from at least one adolescent or adult wave were available from 7883 females and 8126 males.
Measurements Social background, family, academic and behavioural predictors measured at ages 7, 11 and 16 years were entered into hierarchical multiple and logistic regressions to predict quantity of alcohol use at ages 16, 23, and 33 years and harmful drinking [i.e. Cut-down, Annoyed, Guilt, Eye-opener (CAGE) questionnaire score] by age 42 years.
Findings Previous drinking was controlled in final models to predict change. Drinking was heavier among those with greater childhood and adolescent social advantage (especially females), less harmonious family relationships, more social maladjustment, greater academic performance, less internalizing problems, more truancy and earlier school-leaving plans.
Conclusions Alcohol use and problems in adulthood can be predicted by indicators of social background, adjustment and behaviour in childhood and adolescence. Results demonstrate that the early roots of adolescent and adult alcohol use behaviours begin in childhood.
Life course research asks questions about linkages between distinct age periods, examining predictors of continuity and discontinuity as individuals develop and progress through life [1,2]. A number of historical studies have examined predictors of the onset of and desistance from antisocial and criminal behaviour across significant periods of the life course (e.g. [3–7]) as well as predictors of substance use trajectories from adolescence into adulthood (e.g. [8–11]). As Zucker  argues, adult substance use probably develops across time, such that early predictors should be identifiable in childhood. However, compared to antisocial behaviour, much less is known about the extent to which risk factors in childhood and adolescence predict alcohol use in adulthood. In the present paper, we use longitudinal data spanning four decades to examine childhood and adolescent predictors of adolescent and adult alcohol use and harmful drinking. Our general research question focuses upon long-term longitudinal prediction: To what extent do childhood and adolescent characteristics predict alcohol use and harmful drinking in adolescence and early and middle adulthood?
Longitudinal research has identified a substantial set of variables predicting greater adolescent alcohol use (see reviews [13–18]). Many of these predictors are similar to those obtained for antisocial behaviours as well as for other adjustment difficulties. For example, family relationship risk factors for heavier adolescent alcohol use include lower closeness to and communication with parents  and greater family conflict . Similarly, social and behavioural adjustment risk factors associated with increased alcohol use include difficult child temperament , childhood antisocial behaviour and aggressiveness [22–24] and negative affect . Other domains of predictors show less consistent or more complex associations with risk behaviours [26,27]. For example, lower social class predicts smoking and early pregnancy [28–30], but college attendance is associated both with higher family social class and greater drinking in early adulthood [31,32]. In adulthood, abstaining from alcohol use is less frequent among those with higher educational attainment and a non-manual social class . Moreover, academic indicators also show inconsistent relationships with alcohol use. For example, a major review by Hawkins et al.  noted that whereas higher childhood IQ predicted substance use in some research, school failure was a consistent predictor of heavier use. These apparently conflicting results are discussed next.
A growing literature suggests that substance use, in particular, shows associations with a paradoxical set of constructs. Well established are deleterious associations of externalizing behaviour problems, school failure and deviant peer contacts in childhood and adolescence with greater substance use, including drinking alcohol in adolescence  and beyond . However, a number of studies have also demonstrated positive correlates of alcohol use in later adolescence and adulthood, including greater peer acceptance, leadership and psychosocial adjustment, and lower family problems, loneliness and self-derogation [35–38]. In one long-term prospective study from early childhood to late adolescence, Shedler & Block  observed that psychosocial competence in childhood predicted experimentation (but not abuse) of marijuana by age 18 years. Similarly, Siebenbruner et al. found that substance use experimentation, in contrast to abstaining, at-risk use and abuse, was predicted positively by adaptation in childhood. In a recent long-term follow-up of the Woodlawn sample into mid-life, first-grade mathematics achievement and greater shyness predicted a greater likelihood of alcohol use disorders in adulthood , and childhood cognitive ability predicted greater self-reported alcohol abuse assessed at age 53 years in the 1946 British cohort study . Moreover, recent debates in the health arena have drawn attention to potential health ‘benefits’ of moderate alcohol use [43,44]. Research in this area should not be interpreted as discounting either the pathological predictors of alcohol and other substance use and abuse, nor its potential serious deleterious effects, but rather draws attention to the complexity of observed associations. Paradoxically, alcohol confers (or co-occurs with) both risks and benefits in the social and health domains. In interpreting such associations, it is important to distinguish the measure and level of alcohol use. In the present paper we examine childhood and adolescent predictors of alcohol use (quantity) and the life-time incidence of harmful drinking, focusing on family background and relationships, child social and behavioural adjustment and academic ability and performance.
To examine long-term prediction of alcohol use and problems in adolescence to mid-life, we use longitudinal data from the ongoing National Child Development Study (NCDS). This project follows a cohort of individuals born in 1958 in Britain, with waves timed strategically at pivotal points in academic and social development. Data are used from age 7 (childhood, after the start of primary school), age 11 (pre-adolescence, prior to the transition to secondary school), age 16 (adolescence, the minimum school leaving age), age 23 (transition to adulthood), age 33 (adulthood) and age 42 (mid-life). To place the results into context, a brief overview of the educational system and historical changes in drinking prevalence are described next.
Cohort members moved from primary to secondary school after age 11 (or 12 in Scotland). The mandatory school-leaving age increased from 15 to 16 in 1974, when cohort members were 16 years old . In the post-World War II period the state supported three types of secondary schools, one of which (grammar school) was selective based on academic ability assessed at age 11 . By 1969 or 1970, when cohort members were 12 years old, the non-selective comprehensive school was becoming the predominant state-supported type of secondary school and was attended by more than half the NCDS cohort members . In addition, fee-paying private schools were attended by a small percentage, generally students of higher social class .
Tracking (or streaming) between and within secondary schools is associated with school-leaving age, participation in higher education and attainment of academic qualifications . Thus, in the educational system experienced by NCDS cohort members, academic performance in primary school was particularly predictive of subsequent educational experiences as well as eventual educational and occupational attainment [48,50]. While it is clear that early academic achievement predicts cumulative advantages in academic and occupational arenas, it is less certain whether early academic performance predicts disadvantages in terms of heavier alcohol use and more problems with alcohol. In the present paper, we use assessments of academic ability as rated by teachers and investigator-assessed performance on cognitive tests, both at the start (age 7) and end (age 11) of primary school, test performance at age 16 and intentions to leave school at the minimum age as predictors of alcohol use in adolescence and adulthood. Based on previous research, we examined whether students with higher academic performance and higher educational aspirations reported drinking more or less during adolescence and adulthood, and whether they experienced more or fewer life-time problems with alcohol.
Alcohol use in context
Britain has one of the highest rates of alcohol use and heavy drinking in Europe . Like their American counterparts, the majority of older adolescents in Britain drink alcohol. For example, in their last year of mandatory education 80% of 16-year-olds in the NCDS reported drinking in the past 30 days, compared to 68% of 12th-graders in the national American Monitoring the Future study who were born in the same year . The purchase and consumption of alcohol at the age of 16 was legal in Britain for this cohort, and at present is legal at age 18.
Historically, per capita alcohol consumption increased in Britain during the 1958 birth cohort's adolescent and adult years . Specifically, total recorded consumption doubled in Britain between 1960 and 2002, with corresponding dramatic increases in cirrhosis mortality among adults aged 15–44 and 44–64 years . Increased consumption has been associated with a decreased real price of alcohol, and may reflect a tendency to drink a greater quantity per occasion rather than more frequently, at least among younger drinkers . Binge drinking and associated health, social and criminal problems have become a major topic of public debate in Britain as they are in many other western countries. Therefore, a consideration of childhood and adolescent risk factors for alcohol use and harmful drinking is timely.
Previous work investigating alcohol prevalence and prediction in the NCDS has been reported by Power and colleagues. Overall trends show that the heaviest drinkers at age 16 were most likely to be heavier drinkers at age 23 , and that alcohol use generally decreases between the ages of 23 and 33, except among those who divorced . Disruption to the family (i.e. death of a parent, divorce) during childhood did not predict alcohol use at age 23 , but parental separation during childhood predicted alcohol consumption and problem drinking by age 33 . Also at age 33, U-shaped relations between alcohol use and health have been documented, such that non-drinkers and heavy drinkers have higher rates of psychological distress and ill health than moderate drinkers . The current analyses of these data incorporate longitudinal prediction of alcohol use from childhood (age 7) to mid-life (age 42), based on academic achievement, childhood behaviour and other family background variables.
Plan of analysis
In the present paper, we conduct regression analyses predicting alcohol use and harmful drinking in the NCDS from social background; childhood, pre-adolescent and adolescent family, academic and behavioural variables; and previous alcohol use. Two sets of hierarchical multiple regression analyses were performed. Predictors were social background factors (step 1), age 7 childhood predictors (step 2), age 11 pre-adolescent predictors (step 3) and age 16 adolescent predictors (step 4). The four outcomes were indicators of the weekly quantity of alcohol use consumed at ages 16, 23 and 33 and life-time harmful drinking reported by age 42. Also reported are results for all predictors in models where alcohol use at the previous wave is controlled (at ages 23 and 33). Constructs were selected that were common across the longitudinal studies in this special issue.
The present study contributes to this literature in four primary respects. (i) We use national longitudinal data beginning with an entire week's births in Britain in 1958  and continuing with strong retention into the fifth decade of life . These properties enable us to examine questions about long-term longitudinal precursors of drinking using a broad, population-based sample. Very little long-term longitudinal research has focused upon childhood predictors of adult drinking and problems. (ii) We focus upon family background, academic and behavioural predictors in childhood and adolescence, including broad domains known to predict antisocial behaviour and substance use concurrently. (iii) Outcomes are alcohol use from adolescence to middle adulthood and the life-time experience of harmful drinking. (iv) As explained previously, while heavy alcohol users tend also to use other substances and to engage in other problematic behaviours, for the majority alcohol use is also a normative part of positive social interactions. Therefore, some aspects of alcohol use may have different predictors than illegal substance use and antisocial behaviour, as discussed earlier.
The NCDS is an ongoing, multi-disciplinary longitudinal study of an initial cohort of all children born in Great Britain between 3 and 9 March 1958 [61–63]. Following the initial assessment of more than 17 000 babies in 1958 (98% of births) the entire cohort was followed-up at ages 7, 11, 16, 23, 33 and 42. In the child and adolescent follow-ups, longitudinal tracking used school records and the National Health Service Central Register. Immigrants born during the same week were also added at ages 7, 11 and 16. Half (51.5%) in the present analyses were male, and three-quarters had parents in unskilled to manual jobs.
The NCDS began as the Perinatal Mortality Study with a focus on social and obstetric factors associated with stillbirth and death in early infancy [60,64]. The study continued as the NCDS with an ever-broadening focus on physical, educational and social development. Data collection methods have been matched to participants' developmental stage . At birth, survey information was collected from the mother and from medical records by the midwife. Childhood and adolescent waves were conducted when cohort members were aged 7, 11 and 16, with information from parents (collected in the home by health visitors), head teachers and class teachers (who each completed questionnaires), the school health service (who conducted medical examinations) and the cohort members themselves (who completed tests of ability at all ages and self-report questionnaires beginning at age 16). In adulthood, there have been major waves at ages 23, 33, 42 and 46. Another wave will be undertaken at age 50 in 2008. Participation has remained high, with more than 11 000 taking part in the age 42 survey in middle adulthood. This represents a retention rate of over 75% after correcting for emigration and death . For the present analyses, data on alcohol use from at least one adolescent or adult wave were available from 7883 females and 8126 males. Previous assessments of the representativeness of the achieved sample at age 42 show relatively little bias, although there is a small tendency for more men and those who were educationally disadvantaged to leave the study . In an analysis of patterns of missing data in the NCDS, Hawkes & Plewis  found that although responders and non-responders at age 42 showed systematic differences, these were small. Moreover, the propensity to not respond at age 42 differed by social class but in only a small way: 9% of all non-manual social classes were non-responders, compared to 12% for skilled manual and non-manual social classes and 14% for unskilled workers.
Table 1 presents a summary of all predictor measures.
Table 1. Childhood and adolescent precursors of adult alcohol use and problems.
| Social class, fathers'||Contrast of manual jobs (unskilled, partly skilled, and skilled manual, coded 0, 68% of sample) with non-manual jobs (skilled non-manual, managerial and professional, coded 1, 32%)|
| Parental education||Years of education completed past mandatory schooling age attained by each parent by child's birth. 84% of mothers and 78% of fathers left school before or at the minimum leaving age; 6% of mothers and 11% of fathers completed three or more additional years of education|
| Reading with child||Mean of 2 items assessing frequency of mothers' and fathers' reading with their child at age 7, α = 0.67|
| Relations with parents||Adolescent reports, age 16, of how well they ‘got on’ with mother and father; mean of 2 items, α = 0.60|
| Social maladjustment||Teacher reports, Bristol Social Adjustment Guide , 12 syndromes (e.g. withdrawal, hostility, restlessness). Means of standard scores at ages 7 (α = 0.89) and 11 (α = 0.90); higher scores indicate more maladjustment|
| Externalizing/ internalizing behaviours||Parent reports of children's emotional and behavioural difficulties, using short forms of Rutter et al.  Health and Behaviour Checklists. Two scales reflecting externalizing problems (six items, e.g. destroys own or others belongings; α = 0.64, 0.66, for ages 7 and 11, respectively) and internalizing problems (four items, e.g. worries about many things; α = 0.54, 0.54, for ages 7 and 11, respectively) were created|
| Teacher ratings||Teacher reports of competence in oral ability, awareness of the world, reading, creativity, and numbers (age 7, α = 0.89) and oral ability, general knowledge, numbers and reading comprised a scale (age 11, α = 0.90)|
| Academic test scores||Composite scores of study-administered test scores. Age 7 measures were the Southgate Reading Test  and an arithmetic test. Age 11 measures were verbal and non-verbal scores on the General Ability Test , reading comprehension , and arithmetic–mathematics ability [64,79]|
| Weekly quantity||Self-reports of the number of pints of beer, glasses of wine, and so on consumed in the previous 7 days at ages 16, 23 and 33 were combined to yield the total number of standard units of alcohol consumed in that week|
| Harmful drinking||The CAGE  measured life-time incidence problems due to alcohol use [i.e. felt you should cut down on your drinking, people annoyed you by criticizing your drinking, felt guilty about your drinking, had a drink (eye-opener) first thing in the morning]. A cut-point of 1 or more affirmative responses to the four items at age 33 or 42 was used as a marker of harmful drinking|
Social class. Fathers' social class was assessed by the Registrar General's social class measure (RGSC), which codes status of current or most recent job along with associated education, prestige and life-style [67,68]. Categories of I, professional; II, managerial/technical; IIINM, skilled non-manual; IIIM, skilled manual; IV, partly skilled; and V, unskilled were dichotomized to indicate whether the occupation of the child's father was manual (including IIIM, IV and V, coded 0, 68% of sample) or non-manual (including I, II and IIINM, coded 1, 32% of sample). Thus, as coded here a higher score indicates higher social class.
Parental education. The level of education attained by each parent by the child's birth was measured by whether the parent had stayed in school past the mandatory age, which was age 14 or 15, depending on the year parents were born. Eighty-four per cent of the mothers and 78% of the fathers left school before or at the minimum leaving age; 6% of mothers and 11% of fathers completed 3 or more years of education beyond the minimum age.
Reading with child. Frequency of mothers' and fathers' reading with their child at age 7 was measured (e.g. [69,70]). In these analyses a mean of the two items (maternal and paternal reading, α = 0.67) was used.
Relations with parents. At age 16, adolescents were asked about their relationships with their parents. Specifically, the questionnaire asked how well they ‘got on’ with their mother and with their father. A mean of the two items (α = 0.60) was used.
Social maladjustment, teacher reports. At ages 7 and 11, teachers completed the Bristol Social Adjustment Guide (BSAG), which is used to measure social maladjustment . Classroom teachers reported whether each child in the study showed any of 12 syndromes (e.g. withdrawal, hostility, restlessness). The means of standardized scores on the scale were calculated for children at ages 7 and 11 (α = 0.89 and 0.90, respectively), with higher scores indicating more maladjustment.
Externalizing and internalizing behaviours, parent reports. Parents rated children's emotional and behavioural difficulties at ages 7 and 11 using short forms of Rutter, Tizard & Whitmore's  Health and Behaviour Checklists (see also [73,74]). Elander & Rutter  reported generally good psychometric properties across a variety of studies, especially for antisocial problems and for teacher ratings. Based on theory and psychometric analyses, two scales reflecting externalizing problems (six items, e.g. destroys own or others belongings; α = 0.64 and 0.66, for ages 7 and 11, respectively) and internalizing problems (four items, e.g. worries about many things; α = 0.54 and 0.54, for ages 7 and 11, respectively) were created.
Teachers were asked to rate the child's academic ability in a variety of domains. At age 7, a scale of academic ability was based on the mean of teacher-rated competence in oral ability, awareness of the world, reading, creativity and numbers (α = 0.89). At age 11, the mean of teacher-rated oral ability, general knowledge, numbers and reading comprised a scale (α = 0.90).
Study-administered test scores were factor-analysed to create composite scores at ages 7 and 11. The measures consisted of the following. At age 7, cohort members were administered the Southgate reading test  (reliability = 0.94) and an arithmetic test (scored 0–10). Principal components analysis (PCA) revealed a single factor capturing 77% of the variance, with component loadings of 0.88 and 0.88. At age 11, cohort members were given a general ability test , which gives a combined verbal and non-verbal score. In addition, tests were developed specifically for the NCDS in collaboration with the National Foundation for Educational Research to assess reading comprehension  (split-half reliability = 0.82) and arithmetic–mathematics ability  (reliability = 0.92) . A single factor again emerged in PCAs, explaining 85% of the variance. The principal component loadings were 0.93 (general ability test), 0.90 (reading test) and 0.93 (mathematics). Factor scores at each age were computed using the regression method for subsequent analyses.
Weekly quantity: units of alcohol consumed. From age 16 to mid-life, participants reported the number of pints of beer, glasses of wine and so on consumed in the previous 7 days. For example, at age 16 the survey asked: ‘How long is it since you had an alcoholic drink (beer, wine, spirits, etc.)?’ with response options ‘less than 1 week’, ‘2–4 weeks’, ‘5–8 weeks’, ‘9–12 weeks’, ‘over 12 weeks’, ‘uncertain’ and ‘never had one’. Participants were then asked: ‘If it is less than 1 week since your last drink, please write down below the number of drinks you have had in the past week, and what they were (e.g. one whisky and one pint of beer)’. Responses to these questions were combined into a measure of the total number of standard units of alcohol consumed in that week, where one unit is equal to a half-pint of beer, a small glass of wine, a standard pub measure of spirits (25 ml) or a small glass of vermouth or sherry (50 ml). In 1992, the Department of Health in Britain issued recommendations for sensible drinking as 22 or fewer units of alcohol per week for men and 14 or fewer for women.
Harmful drinking. Was assessed using the Cut-down, Annoyed, Guilt, Eye-opener (CAGE) questionnaire [80,81] for life-time incidence of four types of problems due to alcohol use (i.e. felt you should cut down on your drinking, people annoyed you by criticizing your drinking, felt guilty about your drinking, had a drink (eye-opener) first thing in the morning). The CAGE is a simple and widely used pre-diagnostic tool for screening for alcohol dependence  with a median internal consistency across 22 samples of 0.74 . We used a cut-point of 1 or more of the four CAGE items as a marker of potential problems with alcohol. An affirmative response to two of four items on the CAGE is considered traditionally clinically significant . However, predictive power for identifying harmful drinking is increased by employing a cut-point of 1 rather than 2 [84,85], particularly for women . In the present study, we examine life-time prevalence of harmful drinking by age 42. Respondents who replied affirmatively at ages 33 or 42 to one or more of four items were coded as having a life-time incidence of harmful drinking.
Description of the variables
Table 2 presents means and standard deviations of the alcohol use variables for males and females. Consistent with previous research, males reported greater alcohol use quantity than females on all three occasions from ages 16 to 33, all P < 0.001, and reported a greater life-time incidence of problems with alcohol by age 42, P < 0.001. Effect sizes of these gender differences were large, as noted in the table. The pattern of results shows the expected skew as well as commonly observed increases into young adulthood and subsequent small declines.
Table 2. Means and standard deviations for outcome variables by gender.
|Weekly alcohol quantity*|
| Age 16||2.67 (3.47)||1.20 (2.01)†||0.51|
| Age 23||20.96 (21.61)||4.50 (7.37)||1.02|
| Age 33||16.38 (18.01)||4.93 (7.52)||0.84|
|CAGE life-time, age 42‡||0.53 (0.50)||0.33 (0.47)||0.41|
Table 3 presents intercorrelations among the predictor variables for males and females. Indicators of similar domains were related positively within and across time, including social background (social class, parent education), academic performance (teaching ratings, test scores, plans to leave school at 16) and behavioural adjustment (social maladjustment, externalizing, internalizing, truancy). Cross-domain and cross-variable correlations were generally in the low to moderate range (i.e. under 0.50). Indicators of academic performance were highly correlated within ages 7 and 11, respectively, and these and externalizing behaviours were highly stable between these ages. These intercorrelations should be considered in the interpretation of unique associations in the subsequent regressions. Table 4 presents correlations of background variables with alcohol use and alcohol problem outcome variables.
Table 4. Correlations of background predictor variables with alcohol use outcomes for males and females.
|Parents read with child||−0.02||0.00||−0.07***||0.03*||−0.04**||0.00||−0.03*||0.01|
|Social maladjustment (age 7)†||−0.06***||−0.02||−0.01||−0.05***||−0.01||−0.07***||0.01||−0.01|
|Academic ability (age 7)||0.12***||0.10***||0.02||0.13***||0.03||0.11***||0.03*||0.08***|
|Academic test scores (age 7)||0.12***||0.10***||0.03*||0.12***||0.06***||0.11***||0.05***||0.09***|
|Externalizing behaviour (age 7)||0.01||0.02||0.02||−0.02||0.03||−0.01||0.05***||0.03|
|Internalizing behaviour (age 7)||−0.04**||−0.01||−0.06***||−0.03*||−0.06***||−0.04**||−0.00||0.01|
|Social maladjustment (age 11)||−0.03*||−0.03*||0.00||−0.05**||0.03||−0.01||0.04**||0.00|
|Academic ability (age 11)||0.11***||0.10***||0.01||0.13***||0.01||0.10***||0.03*||0.09***|
|Academic test scores (age 11)||0.11***||0.11***||0.01||0.15***||0.02||0.10***||0.05**||0.10***|
|Externalizing behaviour (age 11)||−0.01||0.01||0.03**||−0.02||0.03||−0.02||0.04**||0.04**|
|Internalizing behaviour (age 11)||−0.08***||−0.03||−0.06***||−0.03||−0.07***||−0.06***||−0.02||0.00|
|Relations with parents (age 16)||−0.11***||−0.13***||−0.01||−0.03*||−0.05**||−0.04||−0.09***||−0.07***|
|Truancy (age 16)||0.24***||0.17***||0.10***||0.02||0.09***||0.03*||0.08***||0.06***|
|School leaving plans (age 16)||0.06***||0.05***||0.07***||−0.07***||0.05***||−0.05**||−0.01||−0.06***|
Predicting alcohol use and problems
Hierarchical multiple regressions were conducted separately for males and females. The analysis proceeded in a series of steps, distinguishing social background predictors (step 1), childhood predictors from age 7 (step 2), pre-adolescent predictors from age 11 (step 3), adolescent predictors from age 16 (step 4) and previous drinking variables (step 5). Outcomes were weekly drinking quantity at ages 16, 23 and 33 and life-time incidence of harmful drinking by age 42. Results are presented in Tables 5–8. Each table contains four sets of columns of results: the coefficients for each predictor on the step entered (column set 1) and after previous drinking was controlled on step 5 (column set 2), for males and females, respectively. It should be noted that the regressions predicting age 16 alcohol use do not control for previous drinking, as this was the first wave in which alcohol use was assessed. Due to the large number of childhood and adolescent predictors, mean substitution of values missing on predictors was used. Due to the large number of tested associations, results are presented in summary format. Readers are encouraged to examine the tables for more detail.
Table 5. Hierarchical multiple regressions predicting weekly alcohol quantity, age 16.
| Social class||0.46 (0.12)||0.06***||0.47 (0.11)||0.06***||0.26 (0.07)||0.06***||0.24 (0.07)||0.05***|
| Father years of education||−0.06 (0.06)||−0.02||−0.04 (0.05)||−0.01||0.04 (0.03)||0.02||0.04 (0.03)||0.02|
| Mother years of education||−0.03 (0.06)||−0.01||0.00 (0.06)||0.00||−0.03 (0.04)||−0.01||−0.01 (0.04)||−0.01|
|Step R2Δ = 0.003**|| || ||Step R2Δ = 0.004***|| || |
|Childhood (age 7)|
| Parents read with child||−0.18 (0.07)||−0.03*||−0.06 (0.07)||−0.01||−0.00 (0.04)||−0.00||0.07 (0.04)||0.02|
| Social maladjustment||0.01 (0.09)||0.00||0.02 (0.09)||0.00||0.13 (0.07)||0.03||0.12 (0.07)||0.03|
| Academic ability, teacher||0.31 (0.10)||0.06**||0.32 (0.10)||0.07**||0.21 (0.06)||0.07***||0.17 (0.06)||0.06**|
| Academic test scores||0.29 (0.07)||0.08***||0.14 (0.07)||0.04*||0.13 (0.04)||0.06**||0.05 (0.04)||0.02|
| Externalizing behaviour||0.33 (0.14)||0.03*||0.15 (0.16)||0.01||0.28 (0.09)||0.05**||0.20 (0.10)||0.03*|
| Internalizing behaviour||−0.36 (0.13)||−0.04**||−0.06 (0.13)||−0.01||−0.15 (0.08)||−0.03||−0.08 (0.08)||−0.02|
|Step R2Δ = 0.017***|| || ||Step R2Δ = 0.011***|| || |
|Early adolescence (age 11)|
| Social maladjustment||0.14 (0.09)||0.02||0.08 (0.09)||0.01||0.03 (0.07)||0.01||−0.02 (0.07)||−0.01|
| Academic ability, teacher||0.22 (0.11)||0.04||0.32 (0.11)||0.07**||0.05 (0.07)||0.02||0.09 (0.07)||0.03|
| Academic test scores||0.10 (0.09)||0.03||0.24 (0.08)||0.07**||0.13 (0.05)||0.06*||0.19 (0.05)||0.09***|
| Externalizing behaviour||0.15 (0.15)||0.02||0.10 (0.14)||0.01||0.10 (0.09)||0.02||0.02 (0.09)||0.00|
| Internalizing behaviour||−0.64 (0.14)||−0.07***||−0.57 (0.13)||−0.06***||−0.08 (0.08)||−0.02||−0.06 (0.08)||−0.01|
|Step R2Δ = 0.006***|| || ||Step R2Δ = 0.003**|| || |
|Adolescence (age 16)|
| Relations with parents||–|| ||−0.35 (0.06)||−0.08***||–|| ||−0.24 (0.03)||−0.10***|
| Truancy||–|| ||1.67 (0.09)||0.24***||–|| ||0.68 (0.05)||0.17***|
| School leaving plans||–|| ||0.69 (0.11)||0.10***||–|| ||0.37 (0.06)||0.09***|
|Step R2Δ = 0.079***||Total R2 = 0.10***||Step R2Δ = 0.052***||Total R2 = 0.07***|
Table 6. Hierarchical multiple regressions predicting weekly alcohol quantity, age 23.
| Social class||−0.40 (0.71)||−0.01||−0.22 (0.71)||−0.00||0.68 (0.24)||0.04**||0.16 (0.24)||0.01|
| Father years of education||−0.78 (0.35)||−0.03*||−0.54 (0.35)||−0.02||0.44 (0.12)||0.06***||0.30 (0.12)||0.04**|
| Mother years of education||0.14 (0.38)||0.01||0.37 (0.38)||0.01||0.41 (0.13)||0.04**||0.24 (0.13)||0.03|
|Step R2Δ = 0.001|| || ||Step R2Δ = 0.012***|| || |
|Childhood (age 7)|
| Parents read with child||−2.07 (0.44)||−0.06***||−1.72 (0.44)||−0.05***||0.15 (0.15)||0.01||0.13 (0.15)||0.01|
| Social maladjustment||−0.15 (0.56)||−0.00||−0.20 (0.56)||−0.01||0.17 (0.23)||0.01||0.10 (0.23)||0.01|
| Academic ability, teacher||0.30 (0.59)||0.01||0.28 (0.62)||0.01||0.78 (0.20)||0.07***||0.43 (0.22)||0.04*|
| Academic test scores||0.84 (0.41)||0.04*||0.64 (0.43)||0.03||0.37 (0.15)||0.04*||0.10 (0.15)||0.01|
| Externalizing behaviour||2.17 (0.84)||0.04*||0.48 (0.98)||0.01||0.27 (0.30)||0.01||0.21 (0.35)||0.01|
| Internalizing behaviour||−3.66 (0.76)||−0.06***||−2.54 (0.80)||−0.04**||−0.59 (0.26)||−0.03*||−0.54 (0.28)||−0.03*|
|Step R2Δ = 0.009***|| || ||Step R2Δ = 0.010***|| || |
|Early adolescence (age 11)|
| Social maladjustment||0.16 (0.56)||0.00||0.11 (0.55)||0.00||0.23 (0.23)||0.01||0.21 (0.23)||0.01|
| Academic ability, teacher||0.37 (0.68)||0.01||0.29 (0.67)||0.01||0.07 (0.24)||0.01||0.04 (0.24)||0.00|
| Academic test scores||−0.27 (0.52)||−0.01||0.09 (0.52)||0.00||0.70 (0.19)||0.08***||0.66 (0.19)||0.08***|
| Externalizing behaviour||2.72 (0.91)||0.05**||2.36 (0.90)||0.04**||0.09 (0.33)||0.00||0.05 (0.33)||0.00|
| Internalizing behaviour||−2.54 (0.82)||−0.04**||−1.88 (0.81)||−0.03*||−0.01 (0.28)||−0.00||0.02 (0.28)||0.00|
|Step R2Δ = 0.002**|| || ||Step R2Δ = 0.004***|| || |
|Adolescence (age 16)|
| Relations with parents||0.19 (0.43)||0.01||0.57 (0.42)||0.02||−0.24 (0.14)||−0.02||−0.11 (0.14)||−0.01|
| Truancy||3.58 (0.65)||0.07***||1.73 (0.66)||0.04**||0.63 (0.22)||0.04**||0.29 (0.22)||0.02|
| School leaving plans||2.86 (0.74)||0.06***||2.47 (0.74)||0.05**||0.08 (0.25)||0.01||−0.08 (0.25)||−0.01|
|Step R2Δ = 0.009***|| || ||Step R2Δ = 0.004***|| || |
|Previous alcohol use|
| Units of alcohol, age 16||–|| ||1.09 (0.09)||0.15***||–|| ||0.53 (0.05)||0.13***|
|Step R2Δ = 0.021***||Total R2 = 0.04***||Step R2Δ = 0.015***||Total R2 = 0.04***|
Table 7. Hierarchical multiple regressions predicting weekly alcohol quantity, age 33.
| Social class||−0.30 (0.62)||−0.01||0.04 (0.58)||0.00||0.65 (0.25)||0.04*||0.36 (0.25)||0.02|
| Father years of education||−0.58 (0.30)||−0.03||−0.32 (0.28)||−0.02||0.35 (0.12)||0.04**||0.15 (0.12)||0.02|
| Mother years of education||0.19 (0.34)||0.01||0.07 (0.31)||0.00||0.39 (0.14)||0.04**||0.26 (0.13)||0.03*|
|Step R2Δ = 0.001|| || ||Step R2Δ = 0.009***|| || |
|Childhood (age 7)|
| Parents read with child||−0.91 (0.39)||−0.03*||−0.20 (0.36)||−0.01||−0.17 (0.16)||−0.01||−0.13 (0.15)||−0.01|
| Social maladjustment||0.00 (0.49)||0.00||−0.22 (0.47)||−0.01||−0.40 (0.24)||−0.03||−0.58 (0.23)||−0.04*|
| Academic ability, teacher||−0.45 (0.52)||−0.02||−0.42 (0.51)||−0.02||0.55 (0.22)||0.05*||0.35 (0.22)||0.03|
| Academic test scores||1.65 (0.36)||0.09***||1.31 (0.36)||0.07***||0.36 (0.15)||0.04*||0.23 (0.16)||0.03|
| Externalizing behaviour||2.41 (0.75)||0.05**||1.24 (0.81)||0.02||0.70 (0.32)||0.03*||0.65 (0.36)||0.03|
| Internalizing behaviour||−3.08 (0.67)||−0.06***||−1.11 (0.66)||−0.02||−0.83 (0.28)||−0.04**||−0.34 (0.28)||−0.02|
|Step R2Δ = 0.011***|| || ||Step R2Δ = 0.010***|| || |
|Early adolescence (age 11)|
| Social maladjustment||1.21 (0.49)||0.04*||1.01 (0.45)||0.03*||0.84 (0.25)||0.05**||0.71 (0.24)||0.04**|
| Academic ability, teacher||−0.04 (0.60)||−0.00||−0.16 (0.55)||−0.01||0.36 (0.26)||0.03||0.29 (0.25)||0.03|
| Academic test scores||0.14 (0.46)||0.01||0.44 (0.43)||0.02||0.09 (0.20)||0.01||−0.04 (0.19)||−0.00|
| Externalizing behaviour||1.49 (0.80)||0.03||0.59 (0.74)||0.01||−0.10 (0.35)||−0.01||−0.22 (0.34)||−0.01|
| Internalizing behaviour||−2.77 (0.72)||−0.06***||−1.89 (0.66)||−0.04**||−0.76 (0.30)||−0.04*||−0.72 (0.29)||−0.04*|
|Step R2Δ = 0.004**|| || ||Step R2Δ = 0.003**|| || |
|Adolescence (age 16)|
| Relations with parents||−0.75 (0.38)||−0.03*||−0.76 (0.35)||−0.03*||−0.27 (0.14)||−0.03||−0.23 (0.14)||−0.02|
| Truancy||2.51 (0.57)||0.06***||1.42 (0.53)||0.03**||0.72 (0.23)||0.04**||0.55 (0.22)||0.03*|
| School leaving plans||1.74 (0.66)||0.04**||0.93 (0.61)||0.02||−0.03 (0.26)||−0.00||0.01 (0.25)||0.00|
|Step R2Δ = 0.007***|| || ||Step R2Δ = 0.003**|| || |
|Previous alcohol use|
| Units of alcohol, age 23||−|| ||0.35 (0.01)||0.38***||−|| ||0.31 (0.01)||0.28***|
|Step R2Δ = 0.139***||Total R2 = 0.16***|| ||Step R2Δ = 0.077***||Total R2 = 0.10***|| |
Table 8. Logistic multiple regressions predicting life-time harmful drinking [Cut-down, Annoyed, Guilt, Eye-opener (CAGE) questionnaire], age 42.
|Background||Step χ2 = 3.13|| || ||Step χ2 = 29.69***|| || |
| Social class||0.85|| ||0.79*||(0.63, 0.99)||1.24|| ||1.19|| |
| Father years of education||1.01|| ||0.99|| ||1.02|| ||0.98|| |
| Mother years of education||0.98|| ||0.95|| ||1.21**||(1.08, 1.35)||1.18**||(1.05, 1.32)|
|Childhood (age 7)||Step χ2 = 19.77**|| || ||Step χ2 = 19.52**|| || |
| Parents read with child||0.97|| ||1.00|| ||1.05|| ||1.08|| |
| Social maladjustment||1.00|| ||0.95|| ||1.14|| ||1.09|| |
| Academic ability, teacher||0.97|| ||0.92|| ||1.17|| ||1.01|| |
| Academic test scores||1.26**||(1.10, 1.44)||1.17*||(1.01, 1.36)||1.16|| ||1.05|| |
| Externalizing behaviour||1.10|| ||0.91|| ||1.54**||(1.14, 2.09)||1.20|| |
| Internalizing behaviour||0.91|| ||1.04|| ||0.89|| ||0.88|| |
|Early adolescence (age 11)||Step χ2 = 13.88*|| || ||Step χ2 = 16.11**|| || |
| Social maladjustment||1.25*||(1.05, 1.49)||1.24*||(1.04, 1.48)||1.21|| ||1.16|| |
| Academic ability, teacher||1.14|| ||1.17|| ||1.16|| ||1.16|| |
| Academic test scores||1.11|| ||1.09|| ||1.21|| ||1.21|| |
| Externalizing behaviour||1.30|| ||1.29|| ||1.59*||(1.07, 2.36)||1.52|| |
| Internalizing behaviour||0.78|| ||0.75|| ||1.04|| ||1.05|| |
|Adolescence (age 16)||Step χ2 = 32.39***|| || ||Step χ2 = 22.20***|| || |
| Relations with parents||–|| ||0.82**||(0.72, 0.93)||–|| ||0.83**||(0.73, 0.94)|
| Truancy||–|| ||1.53***||(1.26, 1.84)||–|| ||1.42**||(1.15, 1.75)|
| School leaving plans||–|| ||0.91|| ||–|| ||0.90|| |
Female cohort members who came from more advantaged social backgrounds (as indicated by social class and parents' educational level) reported greater alcohol quantity per week at ages 16, 23 and 33 and more harmful drinking at age 42, as indicated by significant associations on the step entered. For males, this was true only at age 16 for weekly quantity. In the final model, when the childhood and adolescent variables were added, social background was a significant predictor of age 16 quantity (both genders), ages 23 and 33 quantity (females only) and age 42 harmful drinking (females only). Thus, where relationships were observed, they were in the direction of greater social advantage predicting more alcohol use.
Predictors in the family domain included parents reading with the child (at age 7) and relations with parents (at age 16). Males who had had parents who read with them reported less weekly alcohol quantity at ages 16, 23 and 33 and a lower incidence of harmful drinking by age 42. At age 23 only, this unique relationship remained after controlling for ages 11 and 16 variables and previous drinking. For females, no significant predictive associations were observed with parental reading. For relations with parents at age 16, all observed associations were in the direction that individuals with more positive relations drank less. Unique associations were observed after all predictor variables were entered for age 16 quantity (both genders), age 33 (males only) and harmful drinking by age 42 (both genders).
The results for academic ability should be interpreted in light of the large (noted previously) intercorrelations of these predictors within and across time. As a result, the raw correlations are given some weight in the interpretation. The four predictors were teacher-rated ability and study-assessed test performance at ages 7 and 11, respectively. For females at all ages and males at age 16 the four predictors were related consistently and positively to weekly alcohol quantity. When the other childhood and adolescent predictors were entered (and previous drinking, where applicable) a majority of tested associations were significant, predicting alcohol quantity at age 16 (both genders) and age 23 (females only). At ages 23 (males) and 33 (both genders), the positive bivariate associations were generally not uniquely predictive of drinking quantity. Predicting the CAGE, female cohort members were 22–32% more likely to have engaged in harmful drinking by age 42 for each standard deviation increment in academic ability (not shown in table). The corresponding figures for males were 9–10% more likely. When the other predictors and previous weekly alcohol quantity were controlled, only higher academic test scores at age 7 for males predicted an increased likelihood of harmful drinking in adulthood.
Social maladjustment showed different associations with alcohol at different ages. Greater maladjustment in childhood and adolescence did not predict weekly alcohol quantity at ages 16 or 23 in the regressions. At age 33, however, cohort members who had shown more maladjustment at age 11 tended to drink more, and by 42 they were 24% more likely to have engaged in harmful drinking, as shown by models containing the other predictors including previous drinking.
Externalizing behaviour (EB) at ages 7 and 11 was associated positively but inconsistently with weekly alcohol quantity and problems across genders and ages. For example, age 7 EB was a significant predictor on the step entered in six of 10 analyses (and age 11 EB in three of 10 analyses). Where significant, greater EB predicted greater quantity and problems. Across all 20 final models, however, in only three cases was EB a significant unique predictor when all the other predictors and previous drinking were controlled.
Internalizing behaviour (IB) at ages 7 and 11 predicted lower weekly drinking quantity but did not predict harmful drinking. Specifically, those who exhibited more IB at age 7 drank less quantity at ages 23 and 33. Greater IB at age 11 predicted greater drinking quantity at age 33 (both genders) and ages 16 and 23 (males only). IB did not predict harmful drinking on the step entered or in the final models, nor was it related to harmful drinking in bivariate logistic regressions (not shown).
Truancy at age 16 and plans to leave school at the minimum age predicted weekly drinking quantity at ages 16, 23 and 33 for males. For females, associations were in the same direction but less consistent. These associations were maintained generally in the presence of the other predictors and after controlling for previous drinking. Truancy predicted a greater likelihood of harmful drinking by age 42 for both genders, but school-leaving plans showed no relationship.
The NCDS follows a nationally representative British cohort of individuals from birth to mid-life. Major strengths of this study include the unique characteristics of the national longitudinal data set spanning childhood through to mid-life, a socio-economically diverse sample, and the relative lack of previous longitudinal research linking child and adolescent risk factors with adolescent and adult alcohol use and problems. Consistent with a life-span perspective on human development, linkages between distinct age periods and predictors of continuity and discontinuity were assessed [1,2]. The question of whether precursors of alcohol behaviours can be identified earlier in the life-span is relevant for understanding the aetiology of substance use and abuse, as well as for the identification of potentially modifiable risk factors that could be the target of universal or indicated interventions.
Paradoxical associations: disadvantages of social advantage and academic ability
Alcohol use was predicted by a paradoxical set of variables. There was some evidence that positive parent–child relationships, as indicated by reading together in childhood (for males) and getting along well together in adolescence (for both genders), were protective against greater drinking quantity and harmful drinking. The predictive effect of reading together, however, appeared to be explained by other variables in the model. Adjustment difficulties in childhood, similarly, had small predictive relationships with drinking, where social maladjustment in pre-adolescence predicted heavier drinking in adulthood and more harmful drinking, externalizing behaviours predicted greater drinking and internalizing problems predicted less drinking, and truancy predicted more drinking (especially for males). These relationships, although small, somewhat inconsistent and often reduced in predictive power in the presence of other predictors, were in the expected directions based on dominant models about childhood risk and protective factors for a variety of problematic behaviours (e.g. [14,17]).
Perhaps more puzzling were the predictive relationships involving family social background and the child's academic abilities. Why might higher social class and greater parental education predict more alcohol use and problems in females generally and in males in adolescence? A simple explanation may be access to financial resources with which to purchase alcohol. At the aggregate level, Britain did experience significant increases in overall consumption of alcohol as economic prosperity increased historically [53,55]. An alternative possibility is that the social acceptability of drinking by girls and women, particularly in public settings such as pubs and bars, may have increased earlier and faster among the middle class than the working class. Early academic abilities propelled individuals into subsequent educational experiences for this cohort [48,50], such that those with greater ability may have been more likely to be exposed to non-manual social classes with a greater acceptance of drinking, in general, and drinking by women, in particular. That is, higher educational achievement is likely to lead to managerial and professional jobs associated with the middle class, in contrast to skilled and unskilled employment in the working class. Thirdly, the cultures surrounding alcohol use may differ by social class in ways that facilitate consumption more for the socially advantaged. For example, traditionally people in the middle and upper social classes have been more likely to consume alcohol at home and to eat in restaurants , entertain others in their homes  and be members of social and community organizations . Each of these leisure patterns may increase opportunities for social drinking. Similar results for childhood cognitive ability predicting greater alcohol problems indicated by elevated CAGE scores were noted recently in the 1946 British cohort study, a group born 12 years before the NCDS cohort analysed here . Finally, continuing schooling past age 16 is strongly associated with social background  and extended education (including attendance of university with its historical tradition of drinking) and postponement of adult roles are conducive to increased socializing, which may include alcohol use .
Similar questions and arguments could be raised about the positive associations of teacher-rated academic abilities and investigator-assessed academic performance in childhood with alcohol use in adulthood. It should be noted that similar associations were observed with harmful drinking, although these were explained by the other predictors including the weekly quantity of previous alcohol use. In addition to the explanations noted above for social background, it is possible that children who are more intellectually able, or at least who perform better academically, are also more socially competent and engaged. In the present sample, for example, children and pre-adolescents who had higher academic performance showed lower social maladjustment, externalizing behaviour and internalizing behaviour. The present results for academic achievement are consistent with long-term longitudinal predictions from the Woodlawn study in several ways . First, the prediction interval was similar, from first grade in Woodlawn and age 7 (as well as 11) in NCDS to the early 40s in both studies. Secondly, in both studies, greater achievement and less shyness were associated with more drinking later in life. Thirdly, the proportion of variance explained was small in both sets of analyses.
Strengths and limitations
A number of measurement limitations affect the interpretation of the data. The range and quality of parent–child relationship variables is relatively narrow. Future long-term longitudinal research on this topic would benefit from incorporating indicators of parental alcohol use, as well as parental warmth, consistent discipline and monitoring that have been shown to be powerful (and somewhat malleable) predictors of substance use during adolescence. The long-term impact of such high-quality parent–child and parent–adolescent relationships on substance use into adulthood could be assessed. A second measurement limitation is that no indicator of clinically significant alcohol problems during adulthood was available. Rather, the CAGE is a commonly used pre-diagnostic tool for alcohol problems and dependence among adults [81–83,91–93]. Affirmative responses to CAGE items in the present study are not used as clinical diagnoses, but rather as indicators of having experienced problematic levels of alcohol consumption during adulthood. Another important limitation is the relatively small effect sizes and the reliance on statistical significance to determine effects worthy of discussion. In support of the strategy employed, we argue that any prediction across such lengthy intervals makes a noteworthy contribution to the literature.
An important strength of the NCDS sample is its diversity with respect to social class and social disadvantage in childhood. It contains equal numbers of males and females, and represents all areas of Britain. It is also important to note that the birth cohort is homogeneous with respect to ethnicity, reflecting Britain's population at the time. More than 95% of the sample was white, although immigrants born in the same week were added to the sample to age 16.
Our findings illustrate the importance of background characteristics as risk factors for alcohol use. Although there were measurement limitations, predictive models spanning up to four decades showed that precursors of adolescent and adult alcohol use behaviours could be identified early in the life-span. At the same time the variance accounted for in the models, particularly that accounted for by the early and distal variables, was small. This may indicate that we have not accounted for potentially important childhood or adolescent predictors of substance use. Clearly, considerable between-person and within-person variability in substance use behaviour is also determined by more proximal contexts in adolescence and adulthood that is not predictable based on childhood and adolescent factors. The present paper, in the context of the accompanying papers in this issue, demonstrates that the early roots of adolescent and adult alcohol use behaviours begin in childhood. Understanding the personal, family and peer characteristics that are precursors to later substance use is probably necessary to understand completely the alcohol use of adults.
Preparation of this manuscript was supported by a grant from the National Institute on Alcohol Abuse and Alcoholism to J. Maggs (AA-015535). Collaboration on the supplemental issue was supported by NSF Grant no. 0322356 to the Center for the Analysis of Pathways from Childhood to Adulthood.
Conflicts of interest
The authors have declared no conflicts of interest.