Leeds Institute of Health Sciences Trajectories of alcohol use from Year 9 to Year 12 Longitudinal Study of Young People in England (LSYPE): one-day introductory workshop 1 st October 2009 City University, London Gareth Hagger-Johnson, Bridgette Bewick, Robert West, Darren Shickle
Public health perspective Important to identify risky patterns early Early alcohol use associated with –Heavier, more frequent usage –Dependence by age 30 –Worse educational outcomes Partly mediated by impact on brain functioning Very few longitudinal studies considering –Age of onset –Change over time –Educational outcomes
Risk factors for problem drinking Male gender White ethnic minority status Physical or sexual abuse Negative attitudes toward schooling Internalising symptoms (e.g. anxiety, depression) Cigarette smoking Antisocial behaviour Family history of alcohol abuse Poor parenting style Low parental monitoring Low socio-economic status
Introduction Longitudinal Survey of Young People in England (LSYPE) –Next steps Funding –The Department for Children, Schools and Families (DCSF) Designed by –Centre for Longitudinal Studies (CLS) –National Centre for Social Research (NatCen) Cohort study –Year 9 English pupils born 01/09/89 to 31/08/90 –Assessed annually from 2004 (year 9, age 13/14) to 2014 Primary aim –Transition through education into work, evaluate impact of policy
Data linkage Waves 1 to 3 currently available: –15,770 households at Wave One (2004, year 9, age 13-14) –13,539 households at Wave Two (2005, year 10, age 14-15) –12,439 households at Wave Three (2006, year 11, age 15-16) –11,449 households at Wave Four (2007, year 12, age 16-17) National Pupil Database –GCSE subjects and grade –Key Stage 3 results Census data Three data files –Family background –Parental attitudes –Young person
Sampling procedure Two stage probability proportional to size (PPS) –With disproportionate stratification Sampling units –Schools Deprived vs. non-deprived Deprived over-sampled by 1.5 –Pupils within schools Major ethnic minority groups (Indian; Pakistani; Bangladeshi; Black African; Black Caribbean; and Mixed) over-sampled at pupil level (n = 1000) Design efficiency = 78.4% –Accurate assessments of national quantities With simple random sample as large as achieved
Variables with public health relevance Family background Parental socio-economic status Personal characteristics Attitudes, experiences and behaviours –Smoking, alcohol and drug use Attainment in education Parental employment Income and family environment as well as local deprivation The school(s) the young person attends/has attended Health data –Birth weight –Psychological distress (GHQ-12; wave 2 and 4 only) –Self-reported health –Long standing illness or disability
Alcohol items in LSYPE Thinking about the last 12 months, about how often did you usually have an alcoholic drink? Was it... –6. Most days –5. Once or twice a week –4. 2 or 3 times a month –3. Once a month –2. Once every couple of months or –1. Less often? Have you ever had a proper alcoholic drink? That is a whole drink, not just a sip. Please do not count drinks labelled low alcohol. –1. Yes, 2. No
Combining two modelling strategies Mixture modelling, latent class analysis Latent growth curve modelling, trajectories y x c y1y2y3y4 x is
Growth mixture modelling Different trajectories Different means c y1y2y3y4 x is
Growth mixture modelling Two-part growth mixture model (Olsen & Shafer, 2001) –u part (use = 1, non-use = 0) Separates never from ever drinkers –Zeros represent never having drunk alcohol –y part (frequency of drinking) Captures frequency of drinking, for ever drinkers Separately for males and females Choose the number of latent classes Bring covariates into model –Do they change the nature and number of latent classes? Can covariates predict latent class membership? –Early predictions to design public health interventions –Practical usefulness (Muthén, 2008)
Females: 2-class solution
Males: 2-class solution
Discussion Evidence for heterogeneity in alcohol trajectories –Pupils do not follow a normative profile Growth mixture modelling as a useful technique Illustrates the importance of longitudinal data collection Predictors of class membership and change –Demographic –Psychosocial Prioritize interventions Identify at-risk pupils Future analysis plans
Leeds Institute of Health Sciences Academic Unit of Public Health Leeds Institute of Health Sciences Faculty of Medicine and Health University of Leeds Charles Thackrah Building 101 Clarendon Road Leeds, United Kingdom LS2 9LJ