Study participants are from the West of Scotland Twenty-07 Study, a population-sampled cohort designed to investigate the influence of social factors on health. Ethics committee approval for the Twenty-07 study was granted by the General Practice Subcommittee of the Greater Glasgow Health Board Area Medical Committee, and the West of Scotland General Practice Ethical Committee.
The design and sampling have been described in detail elsewhere [26, 27]. In brief, the Twenty-07 study comprises three cohorts recruited at around age 15, 35, and 55 years in 1987/8. Our analyses are based on the oldest age group who, in wave one, took part in two interviews in the study participant's home, one enquiring about social circumstances, and another with a nurse about health-related factors (N [proportion of target sample]: 851 [88%] in women; 700 [88%] in men). In wave 2 (1990/2; age around 59 years), attempts were made to re-interview these men and women about their alcohol drinking habits and any related problems. Again, response was high (681 [83%] in women; 578 [87%] in men).
Assessment of life course socioeconomic position
Early socioeconomic circumstances were based on four indices. Paternal occupational social class was coded into one of six categories according to the Registrar General's schema . Family structure was denoted by the presence of both biological parents at age 15 years. Respondents also reported their number of siblings, and age at leaving school (range: 12–19 years). Assessment of adult socioeconomic circumstances is based on seven indices. Occupational social class was coded as above. To categorise their employment status, study participants identified themselves as: retired, disabled/invalid, caring for the home/housewife, in education, unemployed (no paid work) or employed/worker/self employed. Income was based on total household earnings after tax, including any benefits; respondents were asked to specify an actual amount in pounds sterling per week, month or year, or, if they were unwilling to do so, to identify an appropriate income band on a preprinted card. Housing tenure was categorised as privately owned or other (council, privately rented [unfurnished], privately rented [furnished], tied to job). Household crowding was calculated by dividing the number of people in the household by the number of rooms; respondents were then assigned to a quartile of the distribution, with the highest quartile representing most overcrowding. Study participants also indicated whether or not they owned a car or van. Finally, for marital status, subjects responded to a series of enquiries which allowed them to be categorised into one of three groups: married, no longer married (separated, widowed, divorced), or never married (single).
Assessment of alcohol consumption and problem drinking
Study participants provided a recall of their alcohol consumption over each of the seven days preceding the interview, reporting separately for five categories of alcohol type: beer (including lager and cider), wine, fortified wine, spirits, and 'other' (e.g., 'alcopops'). Responses were expressed in units which represent 8 grams of pure alcohol, equivalent to half a pint of ordinary beer, lager, or cider, a small glass of wine, or a single measure of spirits. For weekly alcohol intake, data were totalled and respondents were dichotomised on the basis of whether or not they exceeded the recommendations for sensible weekly intake (21 units for men, 14 units for women) . For daily intake, the number of days in the preceding seven on which a study participant exceeded 4 units (men) and 3 units (women)  was computed; respondents who had exceeded these guidelines on at least one day in the previous week were classed as 'heavy daily' drinkers. Both alcohol outcomes were categorised into whether or not they surpassed these limits (referred to here as "heavy weekly" or "heavy daily" drinkers).
All participants, with the exception of those who indicated they were lifelong non-drinkers, were asked to complete the CAGE questionnaire [29, 30]. CAGE is an acronym based on the four questions that comprise the inventory: Have you ever felt you ought to cut down on drinking? Have people annoyed you by criticizing your drinking? Have you ever felt bad or guilty about your drinking? Have you ever had a drink first thing in the morning (eye-opener) to steady your hands? These items were used to create a simple drinking problem scale, with each positive response given a score of one; a higher score indicates the presence of an alcohol problem. A total CAGE score of 2 or more is considered clinically significant and this was used in the present analyses. While the CAGE does not provide standard Diagnostic and Statistical Manual diagnosis of alcohol dependence, a positive response on two or more questions indicates a high likelihood of the presence of problematic drinking .
Based on these definitions of alcohol intake and problem drinking, preliminary analyses indicated there were too few women who could be classified as cases in order to facilitate analyses; we therefore focused on data from men only. We examined the relation of both early and later life socioeconomic position with these drinking outcomes using logistic regression modelling. Initially, our analyses were unadjusted, producing bivariate odds ratios. To explore linearity – an assumption inherent in the uses of the relative index of inequality (RII; see below) – we added a quadratic term to the model for each of the socioeconomic exposures variables when examining their relation with the three alcohol outcomes. As it was only possible to test for linearity for those predictor variables with three or more categories, we did not run the analyses for the dichotomously coded family structure, employment status, housing tenure and car ownership. The six remaining variables and three alcohol outcomes therefore resulted in 18 exposure-outcome permutations. Of these, in only one (housing crowding) did the quadratic term attain statistical significance at conventional levels. As this exceptional result is likely to be a chance finding, the linearity assumption can be regarded as not having been violated. Next, as we have done before , we calculated a RII to quantify the association of early and adult life exposures with drinking outcomes. Using the RII facilitates a comparison of effect estimates across a diverse range of indices of socio-economic position. Markers of socioeconomic position were recoded where necessary so that high values reflected disadvantage. The RII was then derived by ranking the participants on each of the socioeconomic measures. For the discrete measures, and in the case of ties for continuous measures, we assigned the mean rank. We then divided these rank scores by the sample size to yield a value between 0 and 1. For the purposes of interpretation, the RII should be regarded as the relative risk of exceeding the stated guidelines or problem drinking in the most disadvantaged group relative to the most advantaged. Its interpretation is the same as a relative risk ratio. Again, logistic regression was used to calculate a RII (odds ratio). The RII is known to elevate effect estimates, especially in variables with only two categories. However, because we used the RII solely as a comparison of the relation of the alcohol outcomes with socioeconomic variables which had different coding structures, our interest did not lie in the absolute size of these relationships.
We then calculated a lifetime composite score for socioeconomic adversity. To do so, we dichotomised all the explanatory variables (0, 1) so that experience of disadvantage on any measure contributed a single point to the score. Explanatory variables were dichotomised at the following demarcation points (for the categories implicitly classified as '0', see Additional file 1): father's and own social class (1 = partly skilled manual, unskilled manual); family structure (1 = circumstance other than having two biological parents in the family at age 15 years); education (1 = left school age 12–14 years); employment status (1 = all other categories but employed [i.e., retired, disabled/ill, caring for the home, in education, unemployed]); income (1 = lowest quartile); housing tenure (1 = all other categories except privately owned house [i.e., council, rent, job-related, other]); household crowding (1 = most overcrowded quartile); car ownership (1 = no); and marital status (1 = single).
Three indices were then created (early life socioeconomic position, range: 0–3; adult socioeconomic position, range: 0–7; life course socioeconomic position, range: 0–10). Again, a higher score indicated greater adversity. Each index was then assigned a RII using the procedure described above. Throughout all these analyses, the analytical sample varies slightly (range: 521–576) depending on missing data for the socioeconomic predictor of interest.