The material for the study is drawn from a Scottish longitudinal community health and lifestyle survey of young people, administered first in-school via questionnaire (ages 11, 13, and 15) and then in the post-education period by nurse interview at age 18/19. The focus here is on data collected between 1999 (aged 15) and 2003 (age 18/19) within the framework of the 'West of Scotland 11 to 16 Study/16+' . The study received approval from Glasgow University's Ethics Committee, participating Education Authorities and schools, and informed consent was obtained from the parents of all participants via 'opt-out' consent forms at ages 11, 13 and 15, verbal consent from participants at each wave and written consent at age 18/19.
Due to the school-based nature of the sample the sampling scheme involved several elements to ensure a representative sample at both the primary and secondary school stages and sufficient school units to investigate or control for school-level clustering . Briefly, the survey used a reverse sampling procedure which randomly selected 43 secondary schools stratified by religious denomination and deprivation, with a separate stratum for independent vs. local authority run schools. These 43 secondary schools were used to select a random sample of 135 primary schools, comprising 'feeder schools', together with those making a high number of placing requests. From these primary schools, classes were randomly selected with all pupils in the classes eligible to participate. Of the 2793 pupils who attended the 43 targeted secondary schools, 2586 (93%) participated in the baseline (age 11) survey. At age 13, the number of participants reduced to 2371 (85%), and by 15 to 2196 (79%), as expected losses in the post-school period substantially reducing the sample size at age 18/19 to 1256 (45%). Full details of the sampling strategy are available elsewhere .
At age 11 the sample was representative (in terms of sex and social class composition) of 11 year olds in the study area . Differential attrition made later waves less representative, with attrition greater among lower social class groups, school truants, pupils of lower ability and with greater emotional and behavioural problems. To compensate for these biases, a weighting scheme was derived . Use of these weights did not substantively alter any of the results presented here. The data used in this paper refer to 2196 pupils in their final year of compulsory education in 43 mainstream secondary schools in the Glasgow area, 1256 of whom provided information when aged 18/19. Parents provided information on pupils' religious background and family socioeconomic status via a supplementary questionnaire in the first wave (age 11) of the study.
In 1999 (aged 15) pupils were asked to rate on a 5-point Likert scale (strongly agree to strongly disagree) how much they endorsed 32 questions relating to values and social attitudes, derived from a number of well-established studies of young people's values [25–28] (see additional file 2). Principal components analysis (varimax) reduced the 32 items to eight factors accounting for 45% of the variance (see additional file 1). The eight factors are broadly comparable with the generic values found by other researchers and each can be located within Schwartz's circumplex model (see additional file 1), and are labeled as follows; Traditional sex-roles (Schwartz's Tradition), e.g. 'Some equality in marriage is a good thing but by and large the husband ought to have the main say'; Work Ethic (Schwartz's Conformity: self-discipline), e.g. 'Even if I didn't like the work, I would still want to do it as well as I could'; Equity (Schwartz's Universalism: equality), e.g. 'The government should tax the rich more in order to help the poor'; Citizenship and sense of belonging (Schwartz's Security: sense of belonging), e.g. 'It is a privilege to be Scottish'; Anti-authority (Schwartz's Anti-tradition: no respect for tradition/obedience), e.g. 'Young people today don't have enough respect for traditional values' (reverse scored); Anti-traditional - apolitical or environmentalist - politics (Schwartz's Universalism/Anti-Power: protect environment), e.g. "There should be restrictions on car drivers in the city to cut down on pollution'; Materialism (Schwartz's Power/Achievement: wealth, successful), e.g. 'There's nothing wrong with having a big house or an expensive car'; Individualism (Schwartz's Power/Anti-Universalism: social power, social justice), e.g. 'The idea that society owes you a living is out of date'. The vast majority of these items have been validated in past studies, but they do not originate from established values scales. However, given the universal nature of values and the high face validity of many items it is highly likely our items are strongly correlated with equivalent items drawn from an established values scale e.g. "Parents can tell you what to do" (our item) Vs "Honoring parents" (equivalent item Schwartz values scale; see figure 1 and additional file 1).
Several background factors related to either values or substance-use at age 15 were recorded. An area deprivation score, range 1 (least) to 7 (most deprived), was derived from pupils' postal codes using the 'Carstairs'  index, a standard measure based upon census data. Social class of the head of household was derived from parental questionnaires completed at wave one (age 11), coded using the standard UK classification system  and categorized as non-manual, manual, or missing. Religious affiliation was obtained from parents and categorized, Church of Scotland (Protestant), Catholic, Muslim, other (Jewish, Methodist, Baptist, etc) and 'none, atheist/agnostic'. At age 15, pupils family structure was coded as 2-parent, 1-parent, reconstituted (one 'birth' parent and new partner) or other (relative, foster parent, or other carer). Principal component (varimax) analysis of the (age 15) 6-item Brief Parental Bonding Instrument , produced two scales representing (low) parental care, e.g. 'My parents help me as much as I need' (reversed) and (high) control, e.g. 'My parents treat me like a baby'. At age 15, pupils reported parental smoking status which was used as a proxy for parental substance-use. At age 15, a generic 4-point Likert (very true to very untrue) scale asked pupils if they identified as a 'risk-taker' and is arguably a good proxy for stimulation and hedonistic (desiring an exciting/varied life) values. Due to low cell frequencies very untrue and untrue categories were collapsed for analysis.
With respect to measures of substance-use at ages 15 and 18/19, smoking was defined as regular smoking, derived from a 5-point frequency scale (never smoked, tried, used to, occasionally or regularly smoke) and is similar to that used by the UK Office for National Statistics . Alcohol-use was assessed on a 7-point frequency scale ('every day' to 'I never had an alcoholic drink') and dichotomized into frequent (drink most days) vs. less frequent use. Pupils reported using a variety of illegal (primarily marijuana) drugs, dichotomized into weekly drug-use vs. less frequent use. In order to assess if our results also applied to moderate levels of substance-use, an additional set of indicators with a lower threshold were chosen for smoking (regular or occasional), alcohol (drank weekly) and drug-use (ever used). Since the results of both sets of analyses were very similar, we elected to present the more severe outcomes because of their greater relevance to public health. Those relating to more moderate outcomes are available upon request.
The analysis used logistic regression to determine the association between values at age 15 and both concurrent and later (age 18/19) substance-use. Analysis was conducted first unadjusted and then adjusted for background factors and included past substance-use for age 18/19 outcomes. We constructed weights to compensate for differential attrition (21), but use of these weights did not alter results, nor did adjusting for school clustering (either via multilevel modeling or adjusting estimates for clustering). The influence of missing data was further explored by comparing results for models using three different methods for dealing with missing data; complete data only; including an additional missing data category for variables with more than modest amounts (50+ cases) of missing data and multiple imputation methods. Multiple imputation was implemented using the STATA 'ice' procedure and included all variables from the relevant model. Categorical variables were imputed using logit or multiple logit, continuous variables using regression and deprivation using ordinal logit commands. Ten imputed datasets were used to calculate the final combined estimates. Although the results for each method were not substantively different, we report results based on multiple imputation. Results from all other alternative models are available upon request.