Department of Public Health and Epidemiology, University of Birmingham, Birmingham B15 2TT, UK
Received 3 May 2002; in revised form 14 September 2002; accepted 8 October 2002
![]() |
ABSTRACT |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
![]() |
INTRODUCTION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
![]() |
SUBJECTS AND METHODS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
In 1978, the first two alcohol questions established if respondents drank at all. Those who had drunk even occasionally in the last year were then asked about their consumption of five types of drink: shandy, beer/lager/stout/cider, spirits (e.g. gin/whisky/rum/brandy/vodka) or liqueurs, port/sherry/vermouth/martini and wine. First, for each drink type they were asked How often have you had a drink of (...) during the last 12 months? with the response categories: most days, three to four times a week, once or twice a week, once or twice a month, once or twice in 6 months, once or twice in the year and not at all in the last 12 months. After asking about frequency of consumption for all drink types, they were then asked for those drink types that they had drunk in the past 12 months How much (...) have you usually drunk on any one occasion during the last 12 months? and responses were recorded in usual measures: pints for shandy and beer group, singles for spirits group and glasses for port group and wine group.
These same questions have been used in subsequent surveys, though there have been minor changes over the years. In 1990, the frequency categories were changed to almost every day, 5 or 6 days a week, 3 or 4 days a week, once or twice a week, once or twice a month, once every couple of months, once or twice a year and not at all in the last 12 months. At the same time, the quantity question was changed to How much (...) have you usually drunk on any one day?. The measures in which quantities of the beer group could be reported were expanded to include other measures such as large and small cans. In 1998, two extra categories of drink were added: strong beer and alcopops. The case for believing that these minor changes in question did not have any substantial effect on consumption estimates is argued in the Discussion.
The responses to the quantity questions were coded as units (one unit equals 8 g of alcohol). Up to 1984, the drinking questions were used to derive quantity frequency (QF) ratings, that grouped drinkers into five categories based on the frequency of drinking their main drink and the quantity consumed (occasional, infrequent light, frequent light, modest, heavier). From 1986, drinkers were categorized on the basis of estimated total number of units of alcohol consumed per week into alcohol consumption (AC) ratings. For 1984 and 1986, tables were published showing both QF and AC ratings to allow comparison (Green, 1989). The estimates of weekly alcohol consumption were calculated by converting each frequency category into its median value, multiplying by number of units of that drink consumed on a typical drinking day and then summing consumption of all drink types.
Data on estimated units of alcohol consumed, sex and age for the surveys in even numbered years from 1978 to 1998 were extracted from computer files held by the data archive at the University of Essex (data sources are listed in Appendix 1). So far as the author is aware, no analysis of AC groups for the 19781982 surveys has previously been published. The estimated units consumed per week were grouped into non- or very light drinkers, light, medium and heavy. For males, these groups drank less than 1, 19.95, 1021, and more than 21 units per week, and for females less than 1, 16.95, 714, and more than 14 units per week, respectively.
Birth cohort analysis
The sample was divided into 5-year birth cohorts for males and females (19771981, 19721976, 19671971 and so on). Each cohort was then followed through the series, for example, the 19521956 cohort consisted of the 2226-year-olds in 1978, the 2428-year-olds in 1980, the 2630-year-olds in 1982 and so on. Cohorts were followed up to 1998 or until they reached age 8387 years. Most cohorts could be followed for 11 observations spanning 20 years. This method of analysis was analogous to that used for smoking data from the General Household Survey (Kemm, 2001). Linear regression of prevalence of different drinking states on age was calculated by the least-squares method (Excel regression tool) separately for each cohort.
Most 5-year age sex bands contained 500 or more subjects. Only age bands that included over 80-year-olds contained less than 300. The only male age band to contain less than 100 was that aged 8387 years in 1984, with 88 subjects. The only female age bands to contain less than 200 were that aged 8387 years in 1984, with 172 subjects and that aged 8286 years in 1998, with 183 subjects. The largest band was that aged 2731 years in 1978, with 1034 male and 1064 female subjects.
![]() |
RESULTS |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
|
|
|
|
Females were not only less likely to drink, but also less likely to drink at higher levels, than males. At age 25 years, about 25% of females were non- or very light drinkers, about 40% were light (1<7 units per week) drinkers, about 20% moderate (714 units per week) drinkers and about 15% heavy (more than 14 units per week) drinkers. With increasing age, women tended to shift into lower drinking categories. By age 80 years, the percentage of non- and very light drinkers has risen to about 65%, the percentage of light drinkers has fallen slightly to about 25% and the percentage of moderate and heavy drinkers has fallen sharply (moderate drinkers to about 5%, heavy drinkers to about 2%).
In summary, women at all ages were more likely to be non- or very light drinkers and less likely to be heavy drinkers than men. Above the age of 20 years in both men and women, the level of drinking tended to decrease with age.
Cohort effects
The plots for each cohort in Figs 1 and 3
tend to overlap, indicating that, at any given age, the cohorts were behaving fairly similarly, but closer inspection and regression analysis revealed some differences. In earlier (older) cohorts for men (cohorts born before 19171921) and less clearly earlier cohorts for women (cohorts born before 19421946), the percentages of non- or very light drinkers (Figs 1
and 2
) in each cohort showed a tendency to separate. Cohort differences were also discernible in the percentages of light drinkers (data not shown). For men born before 19271932 and for women born after 19271932, those in earlier cohorts were more likely to be light drinkers at any age, than those in later cohorts. In females, earlier cohorts were less likely than later cohorts to be heavy drinkers at any age (Figs 3
and 4
). This effect was particularly marked in the most recent cohorts (19771981, 19721976, 19671971). For men, the same trend was apparent only for cohorts born before 19371941.
In summary, earlier (older) cohorts of both men and women drank less at any age than cohorts born later.
![]() |
DISCUSSION |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
This study also reveals cohort effects, but these are strikingly less obvious than those found when the same data source was analysed for cohort effects in smoking using a similar method (Kemm, 2001). In this latter study, the plots of prevalence of never smoking and current smoking against age were clearly separate for different cohorts. A general tendency can be discerned for members of each cohort to be less likely to be non- or very light drinkers, less likely to be light drinkers and more likely to be heavy drinkers at any age, than members of earlier cohorts. However, for men this trend seems to have ended with those born from about 1927 to 1931, with no clear cohort effect thereafter. For women, the trend has continued and possibly becomes stronger in cohorts born after 19271931.
Concern that younger women are drinking more is reinforced by the finding in this paper that the cohort of females born from 1977 to 1981 stands out as having a much higher percentage of heavy drinkers (Fig. 3) than earlier cohorts. Other reports show that the percentage of heavy drinkers among women aged 1824 years has risen from 18% in 1996 to 33% in 2000 (Walker et al., 2001
). However, the data set and the analytical approach used in the present paper are not well suited to examine the lower extreme of the age range.
The wide confidence intervals around some points, particularly those for older ages, could have masked some cohort differences. The data covered a period when alcohol consumption, as estimated from excise data, has been relatively constant. Mean consumption in litres of absolute alcohol per year per head aged over 15 years was 9.4 in 1978 and 9.7 in 1998 (British Beer and Pub Association, 2001). Cohort effects might have been more apparent if overall consumption by the population had been changing rapidly, as smoking levels have.
As there have been changes in the sampling procedure, the precise questions asked and the precise method of estimating alcohol consumption used in the General Household Survey over the period described, one must question the comparability of the samples. However, the sampling procedure has always been robust and any differences arising from changes in the sampling procedure are likely to be small. Similarly, the changes in questions and estimation process, described in the Methods, have been minor. They have not produced obvious discontinuities in the trends published [for example, see tables 9.14 and 9.15 on AC groups in the report of the General Household Survey 1998 (Bridgwood et al., 2000) or table 11.1 on QF groups in General Household Survey 1984 (OPCS, 1986
)].
Surveys of alcohol consumption are liable to numerous errors, including incomplete sampling frames, differential non-response, under- or over-reporting of frequency, under- or over-reporting of quantity, recall bias and social acceptability bias (DeVries et al., 1999; Feunekes et al., 1999). There is particular concern that heavy drinkers may be disproportionately lost to the sample and, when included, may under-report their consumption. The General Household Survey is not immune to these problems, but they are minimized by detailed interviewing protocols, trained interviewers and quality control systems. It may thus be expected to produce high quality data.
There are appreciable regional differences in drinking behaviour in Great Britain, which is the area covered by the General Household Survey. The analytical approach used in this paper cannot explore these. Breaking the data into a large number of cells produced some small numbers especially for the extreme groups (non-drinkers and heavy drinkers) and the oldest age bands. In consequence, the confidence intervals are quite wide for some groups. For example, the 95% confidence intervals around an estimated prevalence of 30% would be 25.534.5% with a sample size of 400 and 21.039.0% with a sample size of 100, while around an estimated prevalence of 10% the corresponding intervals would be 7.112.9% and 4.115.9%, respectively.
There are limited longitudinal studies and sequential cross-sectional studies with which these findings may be compared. The General Household Survey data analysed in this paper covers a longer period, with more consistent data collection methods, more frequent observations, more cohorts and a wider age range, than most reported studies.
In a complex meta-analysis of several longitudinal studies, Johnstone et al. (1996) found evidence of age, cohort (cultural) and period effects. A study in The Netherlands took advantage of a series of four cross-sectional surveys to compare six cohorts of men and women from 1958 to 1989 (Neve et al., 1993
). These latter authors concluded that, in men and women, the percentage of abstainers increased with age, but considered differences between cohorts were due to period, rather than cohort, effects. The normative ageing study in the USA, which collected data on three cohorts in 1973, 1982 and 1991 (Levenson et al., 1998
), found that consumption declined with age in some, but not all, cohorts and that there were significant differences between cohorts. A study of two cohorts of women in Gothenburg first seen in 1968 and followed after 12 and 24 years (Bengtsson et al., 1998
) showed increases in the percentages consuming wine and spirit, but no clear differences between the cohorts.
Several studies, which followed single cohorts, gave information on age effects, but, of course, allowed no comparison between cohorts. Examples of such studies are a study of 40-year-old Danish men and women from 1976 to 1987 (Saelan et al., 1992) and a study of Czech men from 1983 to 1993 (Kubicka et al., 1998
), neither of which showed a simple decrease of consumption with age.
It should be noted that, although analysed as a cohort, this study is a multiple cross-sectional one based on repeated samples of each birth cohort, rather than the whole cohort. Studies such as this indicate changes in percentages in different categories, but give no information on the percentages moving between categories. For example, the situation in which no abstainer starts to drink cannot be distinguished from the situation in which a percentage of abstainers start to drink but are replaced by a corresponding number of drinkers who stop. The data on drinking are less satisfactory in this respect, than data on smoking, which included a category of ex-smoker.
Only longitudinal studies appropriately analysed can give information on percentages changing drinking categories. Two studies from Finland over 5 years (Seppa et al., 1999) and 10 years (Kivela et al., 1988
) showed appreciable numbers moving between drinking categories. A study of French electricity workers (Zins et al., 1999
) suggested that, even over a period as short as 2 years, 30% of heavy drinkers reduced their consumption but were partly replaced by a smaller number who increased their consumption. A study of a cohort of Dutch men and women showed relatively few becoming frequent drinkers, but many ceasing to be frequent drinkers over a 10-year period (Hajema et al., 1997
).
In conclusion, cohort analysis of data from Great Britain confirms that, in this population, at all ages, men are more likely to be drinkers and more likely to consume alcohol heavily, than women. In both men and women after the age of 20 years, younger people are more likely to drink and more likely to drink heavily, than older ones. Cohort effects are also apparent with members of earlier (older) cohorts likely to have drunk less at any age than members of later cohorts.
![]() |
DISCLAIMER |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
![]() |
APPENDIX 1: DATA SOURCES |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
GHS, 1980 [computer file]. The Data Archive [distributor]. 1982, SN 1766.
GHS, 1982 [computer file]. The Data Archive [distributor]. 1985, SN 2034.
GHS, 1984 [computer file]. The Data Archive [distributor]. 2 December 1986, SN 2154.
GHS, 1986 [computer file]. The Data Archive [distributor]. 20 February 1989, SN 2569.
GHS, 19881989 [computer file]. The Data Archive [distributor]. 20 November 1990, SN 2724.
GHS, 19901991 [computer file]. The Data Archive [distributor]. 6 October 1992, SN 2937.
GHS, 19921993 [computer file]. The Data Archive [distributor]. 19 January 1994, SN 3166.
GHS, 19941995 [computer file]. The Data Archive [distributor]. 13 May 1996, SN 3538.
Office for National Statistics, Social Survey Division, General Household Survey, 19967 [computer file], 2nd edn. Colchester, Essex; The Data Archive [distributor]. 5 October 1998, SN 3804.
GHS 19989 [computer file]. The Data Archive [distributor]. 3 August 2000, SN 4134.
![]() |
REFERENCES |
---|
![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() ![]() |
---|
Bridgwood, A., Lilley, R., Thomas, M., Bacon, J., Sykes, W. and Morris, S. (2000) Living in Britain: Results for the 1998 General Household Survey. The Stationery Office, London.
British Beer and Pub Association (2001) Statistical Handbook. Brewing Publications, London.
De Vries, J. H. M., Lemmens, P. H. H. M., Pietinen, P. and Kok, F. J. (1999) Assessment of alcohol consumption. Chapter 2. In Health Issues Related to Alcohol Consumption, 2nd edn, MacDonald, I., ed., pp. 2762. ILSI Europe and Blackwell Science, Oxford.
Fuenekes, G. L., Vant Veer, P., VanStaveren, W. A. and Kok, F. J. (1999) Alcohol intake assessment: the sober facts. American Journal of Epidemiology 150, 105112.[Abstract]
Green, H. (1989) General Household Survey 1986. Drinking Report. HMSO, London.
Hajema, K.-J., Knibbe, R. A. and Drop, M. J. (1997) Changes in alcohol consumption in a general population in the Netherlands: a 9-year follow up study. Addiction 92, 4960.[CrossRef][ISI][Medline]
Johnstone, B. M., Leino, E. V., Ager, C. R., Ferrer, H. and Fillmore, K. M. (1996) Determinants of life course variation in the frequency of alcohol consumption: meta-analysis of studies from the collaborative alcohol related longitudinal project. Journal of Studies on Alcohol 57, 494506.[ISI][Medline]
Kemm, J. R. (2001) A birth cohort analysis of smoking by adults in Great Britain 19741998. Journal of Public Health Medicine 23, 306311.
Kivela, S., Nissinen, A., Ketola, A., Punsar, S., Puska, P. and Karvonen, M. (1988) Changes in alcohol consumption during a ten year follow up among Finnish men aged 55 to 74 years. Functional Neurology 3, 167178.[Medline]
Kubicka, L., Csemy, L, Duplinsky, J. and Kozeny, J. (1998) Czech mens drinking in changing political climates 198393: a three wave longitudinal study. Addiction 93, 12191230.[CrossRef][ISI][Medline]
Levenson, M. R., Aldwin, C. M. and Spiro, A. (1998) Age, cohort and period effects on alcohol consumption and problem drinking: findings from the Normative Ageing Study. Journal of Studies on Alcohol 59, 712722.[ISI][Medline]
Neve, R. J. M., Diederiks, J. P. M., Knibbe, R. A. and Drop, M. J. (1993) Developments in drinking behaviour in the Netherlands from 19581989, a cohort analysis. Addiction 88, 611621.[ISI][Medline]
OPCS (Office of Population Censuses and Surveys) Social Survey Division (1980) Drinking. Chapter 8. In General Household Survey 1978, pp. 134139. HMSO, London.
OPCS (Office of Population Censuses and Surveys) Social Service Division (1986) General Household Survey 1984. HMSO, London.
Saelan, H., Moleer, L. and Koster, A. (1992) Alcohol consumption in a Danish cohort during 11 years. Scandinavian Journal of Social Medicine 20, 8793.[ISI][Medline]
Seppa, K., Pitkajarvi, T. and Sillanaukee, P. (1999) Alcohol consumption profile by time in middle aged men: a longitudinal study based on three different diagnostic instruments. Alcohol and Alcoholism 34, 6570.[Abstract]
Walker, A., Maher, J., Coulthard, M., Goddard, E. and Thomas, M. (2001) Living in Britain: Results from the 2000/01 General Household Survey. The Stationery Office, London.
Zins, M., Carle, F., Bugel, I., LeClerc, A., DiOrio, F. and Goldberg, M. (1999) Predictors of change in alcohol consumption among Frenchmen of the GAZEL study cohort. Addiction 94, 385395.[CrossRef][ISI][Medline]
|