Research Article: Changes in sobriety in the Swedish population over three decades: age, period or cohort effects?

Date Published: April , 2012

Publisher: Blackwell Publishing Ltd

Author(s): Kozma Ahacic, Robert F Kennison, Ingemar Kåreholt.


This study aimed to examine age, cohort and period trends in alcohol abstinence.

Two surveys, the Level of Living Survey collected in 1968, 1974, 1981, 1990 and 2000, and the Swedish Panel Study of the Oldest Old (SWEOLD) collected in 1992 and 2002, were studied with graphical depictions of cross-sectional and longitudinal data presented over time and over age. Cross-sectional 10-year age group differences, time-lag differences between waves and within-cohort differences between waves for 10-year birth cohorts were examined. Logistic regression models were applied to confirm the observed patterns.

The samples were representative of the Swedish population.

Participants ranged in age from 18 to 75 (n = 5000 per wave), and 77+ at later waves (n = 500).

Alcohol abstinence was determined by asking ‘Do you ever drink wine, beer, or spirits?’, where a ‘no’ response indicated abstinence.

Decreases in abstinence rates were observed from 1968 to 2000/02. While cross-sectional analysis indicated increased abstinence with advancing age, the longitudinal analysis suggested otherwise. Inspection of cohort differences revealed little change within cohorts and large differences between cohorts; abstinence rates declined in later-born cohorts up to the 1940s birth cohorts; stability was observed in cohorts born since the 1940s. Logistic regression models indicated that neither age nor period were significant (P > 0.05) predictors of abstinence when cohort (P < 0.001) was included. Decreasing proportions of total alcohol abstainers in Sweden from 1968 to 2000 appear to be attributable primarily to decreases in successive cohorts rather than drinkers becoming abstainers.

Partial Text

The purpose of this study is to examine age, cohort and period effects in alcohol abstinence rates for the Swedish population from 1968 to 2002. Whereas age effects are due to biological, psychological and behavioural processes connected with maturation and ageing [1], cohort effects reflect historical differences in the social or physical environment that occur during development in childhood and young adulthood [2]. Period effects refer to the events and influences during a specific time-period, such as the period from 1968 to 2002. Both cohort and period effects may result from the same kind of sudden events and changes in the physical and social environment (e.g. policy changes), but they may also reflect the influences from more subtle changes such as shifting attitudes. Despite their possible common origin, it is important to differentiate between cohort and period effects in order to place events and influences into a proper time-frame.

The data used in this study were from two Swedish studies. The Level of Living Surveys (LNU) were conducted in 1968, 1974, 1981, 1991 and 2000 [14]. All waves were nationally representative cross-sectional samples (n = 5000 per wave) of the Swedish adult population and included people aged 15–75 years. In the 1991 wave the lower age limit was raised from 15 to 18. Between 1968 and 2000 the response rate fell from 90.6% (n = 5654) to 76.6% (n = 5126). In 1968 the sample consisted of approximately 6000 people between the ages of 15 and 75. When the survey was repeated in 1974, participants from the initial sampling in 1968 who were still living in Sweden and under the age of 76 were recruited to participate. In 1974 and in successive waves, samples from later-born birth cohorts and recent immigrants were added to maintain the sample’s cross-sectional representativeness.

Figure 1 shows the abstinence rates, where age plots appear in the top row; period plots appear in the middle row; and cohort plots appear in the bottom row. The two confounded effects, whether age, cohort or period, are identified in the heading of each graph. Figure 1a,e displays the cross-sectional differences; Fig. 1c,f shows the time-lag differences; and Fig. 1b,d displays the longitudinal differences.

As indicated by the different graphical displays, abstinence rates were dominated by cohort effects. Rates of abstinence decreased in a curvilinear manner for later-born cohorts. While approximately half the 1895–1904 cohort was abstinent, the rate of abstinence decreased successively to 1/10th of the 1945–1954 cohort. Cohorts born after the 1945–1954 cohort had approximately the same rate of abstinence.