1 - Alcohol and Alcoholism

Alcohol and Alcoholism Vol. 47, No. 5, pp. 581–590, 2012
Advance Access Publication 4 July 2012
doi: 10.1093/alcalc/ags071
EPIDEMIOLOGY
Drinking Less But Greater Harm: Could Polarized Drinking Habits Explain the Divergence Between
Alcohol Consumption and Harms among Youth?
Mats Hallgren1, Håkan Leifman2, * and Sven Andréasson1
1
Department of Public Health Sciences, Division of Social Medicine, Karolinska Institutet, Stockholm 171 76, Sweden and 2The Swedish Council for
Information on Alcohol and other Drugs (CAN), 107 25 Stockholm, Sweden
*Corresponding author: Tel.: +46-8-12345-501; E-mail: [email protected]
(Received 29 December 2011; first review notified 25 March 2012; in revised form 31 May 2012; accepted 1 June 2012)
Abstract — Aims: This paper describes changes in alcohol consumption among Swedish youth over the past decade with the aim
of exploring the polarization hypothesis, which asserts that while a majority of young drinkers have reduced their alcohol consumption, a subgroup have increased their drinking substantially, resulting in greater harm. Methods: We analysed repeated cross-sectional
self-report data from 45,841 15–16-year olds and 40,889 18–19-year-old high-school students living in the Stockholm municipality
between 2000 and 2010. The questionnaire assessed alcohol and drug use, and risk factors for alcohol misuse. Changes over time at
different levels of consumption are presented by age and gender. Results: We find evidence of a polarization effect in youth drinking, with consumption reducing significantly over the past 10 years among all young people, except the heaviest drinkers, where consumption and binge drinking tended to increase. The dispersion in per capita consumption also increased over time, indicating more
heavy drinkers. The total number of risk factors for alcohol misuse decreased among most survey participants from 2000 to 2010,
but with variability between years. Conclusion: Polarized drinking habits are a likely explanation for the recent divergence between
per capita alcohol consumption, which has decreased, and alcohol-related hospitalizations, which have increased sharply among
Swedish youth in recent years. We suggest that ongoing social changes could be affecting young people in the form of greater disparities, which are associated with a higher incidence of social problems generally, including heavy drinking.
INTRODUCTION
Alcohol consumption is strongly associated with the three
main causes of death and injury among Western youth;
namely violence, suicide and motor vehicle accidents (Cohen
and Potter, 1999). There is also a positive association
between per capita consumption and alcohol-related harm,
both in adults and younger populations (Norström and
Ramstedt, 2005). Consequently, researchers and policymakers focus considerable attention on yearly changes in per
capita consumption. In Sweden, consumption among youth
aged 15–16 years reached a peak in 2000 before reducing
steadily until 2010 (CAN, 2010). These reductions contrast
with sharp increases in hospitalizations due to alcohol intoxication over the same period. For example, between 2000 and
2010, the number of hospital admissions in Stockholm due
to alcohol intoxication or poisoning increased by 89%
among 15–16-year olds and by 182% among 18–19-year
olds—changes that occurred during population increases of
only 15 and 33%, respectively (Ahacic and Thakker, 2010;
Valdatabasen, 2011) (Fig. 1). Increases were also observed
nationally during this period (CAN, 2010). A similar divergence between consumption and alcohol-related harm has recently been reported in the UK (Meier, 2010) and Australia
(Livingston et al., 2008), which raises important questions
about the relationship between per capita consumption,
heavy drinking and alcohol-related problems.
On a population basis, reductions in per capita consumption are associated with less heavy episodic drinking and
fewer alcohol-related harmful effects (Duffy, 1986; Rose and
Day, 1990), and this relationship underpins many prevention
strategies currently implemented around the world. Swedish
alcohol policies, for example, aim to reduce per capita consumption through tight controls over the availability of
alcohol, an approach heavily influenced by Ole-Jörgen
Skog’s theory of the collectivity of drinking, which asserts
that reductions in yearly consumption influence all levels of
drinking concurrently, including heavy consumption (Skog,
1985). Over the past 10 years in Sweden, however, the
opposite trend has occurred among youth, with per capita
consumption decreasing, while alcohol-related hospitalizations have risen. These changes have emerged in the context
of large increases in the number of young people who
abstain from alcohol completely, a scenario that does not
conform to Skog’s theory.
One possible explanation for this divergence is that a subgroup of young people are drinking more alcohol than their
peers over time, or in ways that are causing more alcoholrelated problems. This has been termed a ‘polarization
effect’, where some young people, possibly with an accumulation of risk factors for alcohol misuse, have increased their
drinking over a specified period, while the majority have
reduced their consumption. However, whether or not such a
polarization exists remains unclear. To our knowledge, this is
the first paper to test the polarization hypothesis empirically.
We achieve this by examining changes at different levels of
self-reported alcohol consumption among Stockholm youth
aged 15–19 years between 2000 and 2010. We also explore
one possible explanation for the polarization effect; namely
that emerging social and health inequalities could be increasing the number of risk factors for alcohol misuse among
heavy drinkers over time, while the majority of young
people drink less and are exposed to fewer risk factors.
MATERIALS AND METHODS
Data on the alcohol consumption among youth and risk
factors are obtained from the Stockholm Student Survey, a
© The Author 2012. Medical Council on Alcohol and Oxford University Press. All rights reserved
582
Hallgren et al.
schools influenced the results, but no effects were found
(data not shown).
Fig. 1. Alcohol-related hospital admissions and per capita alcohol
consumption for youths aged 15–16 and 18–19 years living in Stockholm.
Hospital admissions are based on ‘acute intoxication’ (ICD-10 code F10.0)
and ‘alcohol poisoning’ (ICD-10 code T51) as the main diagnoses. The
figure shows a sharp increase in youth hospital admissions in Stockholm for
acute alcohol intoxication and poisoning, while per capita alcohol
consumption reduced slightly over the same 10-year period.
repeated cross-sectional self-report questionnaire completed
every second year by high-school students in Year 9
(aged 15–16 years) and Year 11 (aged 18–19 years) in
the Stockholm municipality. The survey, financed by the
Education Department, is conducted during the spring period,
in which the questionnaire is completed anonymously by
students during class time before being returned in a sealed
envelope to teachers. Students absent from school because of
illness were posted a questionnaire to be completed at home
and then returned by regular mail. The survey includes ~350
questions covering demographic information, alcohol and
drug use (frequency, quantity and type) and various risk/
protective factors for alcohol misuse, including criminal behaviour, psychosocial health, truancy, school and parental
support (El-Khouri et al., 2005). The questionnaire is the
largest youth alcohol and drug survey in Stockholm and is
used to monitor important changes in health-related behaviour. Participation in the survey is mandatory for all public
schools, which comprise ~90% of all schools in Stockholm.
Independent (fee paying) schools participate voluntarily.
Participants
As the survey was completed during school hours and supervised by a class room teacher, the response rates were high;
ranging between 75 and 80% across all six survey years
(2000, 2002, 2004, 2008 and 2010). The student participants
represent ~56% of all young people in their age group living
in Stockholm. As the total number of schools in Stockholm
(including independent schools) expanded considerably
between 2000 and 2010, so did the number of students who
completed the survey—from 8915 students and 76 schools
in 2000 to 15,746 students and 182 schools in 2010.
Approximately, equal numbers of males and females in both
school years participated. As the survey was anonymous,
non-responders could not be followed up for comparison
purposes. Separate analyses were conducted to examine
whether changes over time in the number of participating
Alcohol consumption
The frequency and the quantity of alcohol consumption
during the past 12 months were assessed on a 12-item questionnaire. Questions about the quantity of alcohol were
answered on a 9-point scale; for example, ‘When you drink
wine, approximately how much do you normally drink?’
with responses ranging from 1 corresponding to ‘less than a
glass <15 centilitres’ to 9 ‘more than three bottles.’ The frequency scale follows the same format; for example, ‘How
often have you drunk wine during the past 12 months?’ With
responses ranging from 1 corresponding to ‘every day’ to 9
‘I have not drunk wine at all during the past 12 months’. Per
capita consumption (in centilitres) was determined by multiplying the quantity and the frequency of reported consumption from these scales. Changes in binge drinking over time
were assessed with a single question: ‘How often have you
consumed the following amounts of alcohol during a single
occasion?’: at least one bottle of wine, five to six shots of
spirits or four cans of strong beer (or six cans of medium
strength beer). This is an established measure, used in annual
alcohol surveys in Sweden since 1972, and roughly equivalent to ‘five drinks’ (CAN, 2010). Estimates of the yearly
frequency of binge drinking were determined by converting
statement response alternatives into numerical scores; for
example, ‘a few times per year’ became ‘three times per
year’, etc.
Risk factors for alcohol misuse
Thirteen risk factors for alcohol misuse were identified in the
questionnaires on the basis of previous research. Our main
interest was whether or not there was a polarization in the
risk factor total score, with changes over time following a
pattern similar to that of the consumption data. We used a
theory-driven approach to identify the 13 risk factors, focussing both on the international literature (Hawkins et al.,
1992; Petridis et al., 1995; Zufferey et al., 2007; Merline
et al., 2008) and recent Swedish studies of youth alcohol
consumption (El-Khouri et al., 2005; Bränström et al., 2008;
Danielson et al., 2010) to guide our selection. Risk factors
(e.g. parental provision of alcohol, having friends who drink,
truancy from school, etc.) are circumstances or personal
characteristics that are presumed to increase the likelihood of
hazardous or harmful drinking. Our goal was to determine
whether or not the total number of risk factors had changed
significantly over time in the entire population surveyed,
compared with the heaviest drinkers. Here, we defined
‘heavy drinkers’ as young people who consumed 20 l or
more alcohol per year, which is approximately the level at
which the polarization effect in consumption emerges in the
aggregated data. We expected to see a reduction in the total
number of risk factors among the majority of students (those
drinking <20 l of alcohol per year), but an increase among
the heaviest drinkers, between 2000 and 2010, following a
trend consistent with the consumption data changes. The
13 risk factors selected were present in each of the six
survey rounds. Questions with more than two response alternatives were dichotomized as shown in Table 1. One item
‘anti-social behaviour’ was composed of nine separate items.
Polarized youth drinking
583
Table 1. Distribution of young people with risk factors for alcohol misuse
Percentage with risk factor
Males
Risk factor
Truancy
Parents offer alcohol
Not living with both parents
High spending money
Smokes regularly
Previous drug use
Get alcohol from parents
Bullied others
Anti-social behaviour
Age first drunk
Parental monitoring
Poor school connection
Both parents unemployed
Dichotomization
Have you been away from school during the
last term without permission? Twice or
more = 0.1/yes, otherwise 0 = no
Do your parents offer you alcohol? Responses ‘yes
a whole glass’ or ‘regularly offer me alcohol’
were coded 1 = yes, otherwise 0 = no
Do you live with your mother and father?
Responses were coded as 1 = yes, 0 = no
How much spending money do you get each
month? If more than 1000 sek per month, then
1 = yes, 0 = no
Do you smoke cigarettes? Responses coded
sometimes or daily = 1/yes, otherwise 0 = no
Have you ever used the following drugs?
Marijuana, LSD, heroin, cocaine, amphetamine,
ecstasy. 1 = yes, 0 = no
Where do you normally get alcohol? If ‘parents
with permission’ 1 = yes, all other responses
0 = no
Have you bullied another person during the last
school year? Once or more = 1/yes, otherwise
0 = no
How many times have you done each of the
following during the past 12 months? Shoplifted,
vandalism, graffiti, stole motorcycle, stole car,
forced someone to give money or phone,
burglary, serious fight, carried weapon.
Responses coded 1–2 times or more were
re-coded as 1 = yes, 0 = no. A total score
was calculated and dichotomised as follows:
0–3 = 0/no, 4-9 = 1/yes
How old were you the first time you felt drunk
(from alcohol), responses coded 0–7 = missing,
8–12 = 1/yes, 13 plus = 0/no
Do your parents know where you are on the
weekends? Responses ‘rarely know or never’
were coded as 1 = yes, otherwise 0 = no
How do you get on at school? Responses coded
badly or very badly = 1/yes, otherwise 0 = no
What do your parents do? If both mother and father
unemployed, then 1 = yes, 0 = no
2000 (n = 3072)
68
Females
2010 (n = 3230)
33
2000 (n = 3894)
66
2010 (n = 4110)
35
54
34
52
31
30
58
35
55
30
55
28
59
37
35
45
42
23
32
19
23
23
13
32
17
18
24
7
11
35
13
9
3
13
7
10
6
11
12
7
9
7
10
7
10
1
1
0.4
0.6
The percentage of all the young people surveyed with risk factors for alcohol misuse in 2000 compared with 2010 (divided by gender).
A total anti-social score was calculated and then dichotomized so that approximately one-third of the highest
responses were coded as 1 (risk factor present).
Data analysis
The survey data were ‘cleaned’ by removing clearly inappropriate responses from students who indicated the highest possible score on every item in both the alcohol and the drug
questionnaires (<1% of responses). Examination of the distribution of alcohol consumption revealed that few students
drank >100 l of pure alcohol per year. On the basis of this
analysis, we determined that a 100 l cut-off was reasonable
because it excluded students who deliberately exaggerated
their drinking habits. For comparison, however, we also
examined changes in consumption levels over time using a
30 and 50 l cut-off, including and excluding the abstainers,
and found that the trends were similar regardless of which
cut-off was used. As the consumption data were heavily
skewed, and so not to violate parametric test assumptions,
we log-transformed the data before conducting t-tests and
tests of the homogeneity of variance. Data imputation was
considered unnecessary because of the high questionnaire
completion rate (above 95% on most items), and the high
number of study participants.
Changes in mean consumption and binge drinking were
examined for 6 years: 2000, 2002, 2004, 2006, 2008 and
2010. To look more closely at changes at different levels of
consumption over time, we report changes in the centilitres
of alcohol consumed each year by percentile ranks (from the
1st to the 99th). Changes in consumption are reported by
school Years (9 and 11) and gender. As the data were crosssectional and different student populations were surveyed
each year, we used independent sample t-tests (with
Bonferroni adjustment for multiple comparisons) to examine
the significance of changes in per capita consumption
between years, with our main focus on changes that occurred
between 2000 and 2010. Changes in both the shape (skewness and kurtosis) and the dispersion of the data [SD, coefficient of variation (CV)] are reported in Tables 2–5. Changes
584
Hallgren et al.
Table 2. Changes in alcohol consumption (centilitres per year) among Year 9 males
Percentile
2000
(n = 1688)
2002
(n = 1741)
2004
(n = 3652)
2006
(n = 1760)
2008
(n = 1848)
2010
(n = 1733)
2000–2010 %
change
2000–2010
absolute change
1
5
10
25
50
75
90
91
92
93
94
95
96
97
98
99
% abstainers
Mean
Median
SD
CV
Skewness
Std. Error Skew
Kurtosis
Std. Error Kurt
0.53
1.92
4.82
29
205
716
1623
1706
1861
2001
2261
2502
3011
3777
4453
5935
23
618
205
1108
179
3.75
0.06
18.14
0.11
0.54
1.92
6.09
37
206
737
1715
1874
2079
2287
2478
2713
3069
3798
4924
6779
25
659
206
1205
183
3.83
0.05
18.58
0.11
0.48
1.08
2.52
12
97
429
1127
1234
1355
1542
1725
1948
2400
3061
4226
5495
30
452
97
1042
231
4.98
0.04
31.11
0.08
0.48
0.91
2.08
12
105
558
1640
1778
1945
2181
2443
2657
3192
4014
5089
6532
34
576
105
1209
209
3.83
0.05
17.55
0.11
0.48
1.05
2.85
21
136
573
1628
1823
1964
2140
2435
2810
3230
4045
5175
7020
36
601
136
1249
207
3.97
0.05
18.80
0.11
0.26
1.04
2.08
14
109
542
1382
1537
1697
2041
2430
2852
3333
4292
5128
6942
42
569
107
1224
217
4.03
0.05
19.06
0.11
−51
−46
−57
−52
−47
−24
−15
−10
−9
2
7
14
11
14
15
17
−0.27
−0.88
−2.74
−15
−96
−174
−241
−169
−165
40
169
349
321
515
674
1007
19
−49
−98
116
38
0.28
—
0.92
—
−8
−48
—
—
—
—
—
—
Changes over time in alcohol consumption, measured in centilitres per year, at different percentiles for Year 9 males between 2000 and 2010. Importantly, the
table shows that consumption has reduced among all the students, up to about the 93rd percentile, where we observe an increase in consumption among the
heaviest consumers only over time.
Table 3. Changes in alcohol consumption (centilitres per year) among Year 9 females
Percentile
2000
(n = 1834)
2002
(n = 1831)
2004
(n = 1786)
2006
(n = 1918)
2008
(n = 2031)
2010
(n = 1762)
2000–2010 %
change
2000–2010
absolute change
1
5
10
25
50
75
90
91
92
93
94
95
96
97
98
99
% abstainers
Mean
Median
SD
CV
Skewness
Std. Error Skew
Kurtosis
Std. Error Kurt
0.53
1.20
3.46
20
120
412
940
1004
1089
1160
1331
1460
1623
1992
2535
3519
19
360
120
671
186
4.57
0.05
30.56
0.11
0.54
1.52
3.68
24
124
400
906
982
1060
1164
1269
1474
1750
2103
2533
3218
19
357
124
642
180
4.03
0.05
21.74
0.11
0.27
0.63
2.16
12
85
336
826
891
979
1078
1185
1415
1665
1927
2480
3614
27
334
84
764
229
6.08
0.05
50.84
0.11
0.26
0.64
1.92
12
94
405
954
1021
1104
1252
1393
1583
1743
2138
2705
3492
32
366
94
750
205
5.08
0.05
37.07
0.11
0.26
0.65
2.58
19
124
500
1079
1151
1254
1436
1586
1849
2100
2538
3256
4967
27
438
124
874
200
4.61
0.05
27.86
0.10
0.26
0.66
2.08
12
83
351
847
914
963
1020
1155
1291
1499
1807
2365
3356
35
316
82
627
198
4.43
0.05
26.07
0.11
−51
−45
−40
−38
−31
−15
−10
−9
−12
−12
−13
−12
−8
−9
−7
−5
−0.27
−0.55
−1.38
−8
−37
−61
−93
−90
−126
−140
−176
−169
−124
−184
−170
−162
16
−44
−38
−44
12
−0.14
—
−4.49
—
−12
−32
—
—
—
—
—
—
Changes over time in alcohol consumption, measured in centilitres per year, at different percentiles for Year 9 females between 2000 and 2010. Importantly,
the table shows that consumption has reduced among all the students, but the reductions are smaller among the heaviest consumers (i.e. the top 5%) compared
with the remaining 95% of drinkers.
Polarized youth drinking
585
Table 4. Changes in alcohol consumption (centilitres per year) among Year 11 males
Percentile
2000
(n = 1546)
2002
(n = 1604)
2004
(n = 1555)
2006
(n = 1863)
2008
(n = 1947)
2010
(n = 2154)
2000–2010 %
change
2000–2010
absolute change
1
5
10
25
50
75
90
91
92
93
94
95
96
97
98
99
% abstainers
Mean
Median
SD
CV
Skewness
Std. Error Skew
Kurtosis
Std. Error Kurt
1.75
11.85
38.78
149
484
991
1919
2038
2146
2323
2531
2729
3033
3404
4028
5299
10
799
484
1043
131
3.18
0.06
14.18
0.12
1.92
12.70
36.77
182
512
1138
2158
2301
2399
2541
2732
2995
3335
3816
4920
5880
10
899
512
1166
130
3.01
0.06
12.66
0.12
1.19
7.67
25.38
115
390
899
1803
1884
2023
2222
2409
2668
3032
3548
4538
5776
14
738
390
1077
146
3.43
0.06
16.14
0.12
1.06
7.49
23.24
123
391
921
1946
2089
2255
2393
2537
2790
3112
3552
4310
5694
16
765
391
1097
143
3.38
0.05
15.89
0.11
0.65
6.86
26.76
111
417
984
2068
2245
2415
2602
2803
3071
3491
4022
4992
6281
17
819
417
1184
145
3.12
0.05
12.66
0.11
0.88
6.57
20.59
109
379
912
1907
2023
2221
2417
2604
2885
3288
3648
4570
6022
18
768
379
1137
148
3.40
0.05
15.81
0.10
−50
−45
−47
−27
−22
−8
−1
−1
4
4
3
6
8
7
13
14
−0.87
−5.28
−18.19
−40
−105
−80
−12
−15
75
95
74
156
255
244
542
723
8
−31
−105
94
18
0.22
—
1.63
—
−4
−22
—
—
—
—
—
—
Changes over time in alcohol consumption, measured in centilitres per year, at different percentiles for Year 11 males between 2000 and 2010. Importantly,
the table shows that consumption has reduced among all the students, except the top 8% of drinkers, where consumption has risen over the past 10 years.
Table 5. Changes in alcohol consumption (centilitres per year) among Year 11 females
Percentile
2000
(n = 1593)
2002
(1710)
2004
(n = 1610)
2006
(n = 1883)
2008
(n = 2125)
2010
(n = 2339)
2000–2010 %
change
2000–2010
absolute change
1
5
10
25
50
75
90
91
92
93
94
95
96
97
98
99
% abstainers
Mean
Median
SD
CV
Skewness
Std. Error Skew
Kurtosis
Std. Error Kurt
0.60
6.04
14.27
71
236
503
1000
1046
1140
1199
1261
1392
1494
1631
1899
3007
10
404
236
551
136
3.58
0.06
19.34
0.12
1.03
4.92
16.28
87
281
632
1152
1267
1363
1496
1585
1675
1844
2358
2706
3621
10
500
281
714
143
4.01
0.05
25.58
0.11
0.63
3.33
11.89
62
217
531
1062
1131
1221
1335
1418
1584
1728
2027
2553
3299
12
430
217
622
145
3.27
0.06
14.38
0.12
0.52
3.24
11.13
63
235
573
1166
1277
1355
1436
1569
1770
1901
2148
2466
3251
15
462
235
701
152
4.16
0.05
28.89
0.11
0.52
4.39
13.71
79
300
679
1386
1449
1533
1617
1751
1877
2040
2304
2636
3576
15
527
300
714
135
3.11
0.05
14.26
0.10
0.57
4.81
13.23
69
250
607
1243
1338
1407
1492
1605
1716
1885
2229
2673
3659
16
484
250
724
150
3.99
0.05
25.06
0.10
−4
−20
−7
−3
6
21
24
28
23
24
27
23
26
37
41
22
−0.03
−1.24
−1.05
−2
14
104
243
292
267
293
344
324
391
598
773
652
6
80
14
173
13
0.41
—
5.72
—
20
6
—
—
—
—
—
—
Changes over time in alcohol consumption, measured in centilitres per year, at different percentiles for Year 11 females between 2000 and 2010. Importantly,
the table shows that consumption has reduced among the light to moderate drinkers (i.e. up to the 50th percentile), but then increases exponentially among the
heavier consumers.
586
Hallgren et al.
in the dispersion of the consumption data and the risk factor
total score were also assessed with Levene’s test, which
determines the statistical significance of changes in the
spread or homogeneity of the data between 2000 and 2010.
Increased dispersion means more heterogeneous data, and
vice versa. We also explored changes in the total number
and dispersion of risk factors over time, on the basis of the
risk factor total score, including the percentage of respondents with different numbers of risk factors for alcohol
misuse. As the consumption data were heavily skewed, we
used Spearman’s non-parametric bivariate correlation to test
the relationship between alcohol consumption and the risk
factor total score. All the analyses were performed using
SPSS version 20.0.
RESULTS
Alcohol abstention rates
Between 2000 and 2010, there has been a steady increase in
the number of young people in Stockholm who abstain from
alcohol completely (Fig. 2). The largest increase in abstention was among Year 9 females and males (up 83 and 82%,
respectively).
Per capita alcohol consumption
Changes in per capita consumption are shown in Fig. 3. Per
capita consumption reduced between 2000 and 2010, mainly
because of the increase in non-drinkers over the same period.
However, even when abstainers are excluded, significant
reductions were found for Year 9 males [t = 5.86(3409),
P < 0.000], Year 9 females [t = 4.27(3559), P < 0.000] and
Year 11 males [t = 3.91(3507), P < 0.000]. Year 11 females
were the only group to increase their consumption between
2000 and 2010 (by 20%); however, the increase was not statistically significant.
Tables 2–5 illustrate important changes at different levels
of drinking over the past decade. The data show consistent
reductions in per capita consumption between 2000 and
2010 across most levels of drinking, except among the very
Fig. 2. Percentage change in alcohol abstention rates. The figure shows
changes over time in the percentage of young people who abstain from
alcohol, by school year and gender.
heaviest drinkers where, in most cases, the trend reverses and
consumption begins to increase over time. This pattern of
consumption is most evident among Year 9 males (Table 2),
where reductions in drinking occur up to the 93rd percentile,
and then consumption increases over time among the heaviest drinkers. A similar pattern was found among males and
females in both school years, but the trend is stronger for
Year 9 males, compared with Year 9 females. The switch to
increasing consumption occurs at different levels of drinking,
depending on school year and gender: for example, at the
93rd percentile for Year 9 males; at the 92nd percentile for
Year 11 males and at the 40th percentile for Year 11
females.
Dispersion of alcohol consumption
Tables 2–5 show changes in both the shape (skewness and
kurtosis) and the dispersion of the consumption data
between 2000 and 2010. The high SD in comparison with
the mean indicates that the data are heavily skewed. Both the
SD and the CV increased between 2000 and 2010 across all
the four groups, except Year 9 females. The skewness of the
data also increased over time (except for Year 9 females), indicating that more young people drink at very high levels,
while the majority continue to drink less.
Using log-transformed data, we examined changes in the
homogeneity of variance between 2000 and 2010. The
results indicate significant increases in the dispersion or
spread of the alcohol consumption data between 2000 and
2010 for Year 9 males (F = 15.84, P < 0.000), Year 9
females (F = 7.99, P < 0.005), year 11 males (F = 13.89, P <
0.000) and Year 11 females (F = 8.57, P < 0.003).
Binge drinking
Changes in heavy episodic drinking (binge drinking) are particularly relevant because this pattern of consumption is
often associated with serious acute harmful effects among
young people (Babor et al., 2010). In the Stockholm Student
Survey, binge drinking was defined as the consumption of at
least one bottle of wine, five to six shots of spirits or four
Fig. 3. Changes in per capita alcohol consumption. The figure shows
changes over time in per capita alcohol consumption (drinkers only), by
school year and gender.
Polarized youth drinking
cans of full-strength beer (or six cans of medium strength
beer), consumed during a single occasion. Changes in the
estimated yearly frequency of binge drinking are shown in
Fig. 4.
As might be expected, males reported binge drinking
more frequently than females, and older adolescents binge
drank more than younger adolescents. Year 11 females were
the only group to significantly increase their binge drinking,
from about 10 times per year in 2000 to 13 “times per
year in 2010 [t = − 6.51(3501), P < 0.000]. By contrast,
binge drinking decreased significantly among Year 9 males
[t = − 3.67(3157), P < 0.000], while Year 9 females and Year
11 males remained fairly constant. The dispersion of binge
drinking frequency scores, assessed by Levene’s homogeneity of variance test, reduced significantly between 2000 and
2010 in Year 9 males (F = 26.05, P < 0.000) and increased
significantly among Year 11 females (F = 20.24, P < 0.000).
To determine whether binge drinking increased over time
among the heaviest alcohol consumers, we examined
changes in the frequency of yearly binge drinking among the
heaviest 5% of drinkers, compared with the remaining 95%.
Binge drinking ‘once per year or more’ increased between
2000 and 2010 among the top 5% of young people by
14.5%, but decreased among the majority of lighter drinking
students by 15.2%. This result might be expected because
most of the consumption occurring in the top 5% of the distribution is heavy episodic drinking.
Risk factors for alcohol misuse
There was a large and statistically significant correlation
between the per capita alcohol consumption and the risk
factor total score (r = 0.468, P < 0.000), indicating that as
consumption increased, so did the number of risk factors.
This relationship was present for both males (r = 0.463, P <
0.000) and females (r = 0.472, P < 0.000) when analysed
separately. However, the association weakened considerably
when only the heaviest drinkers were examined (i.e. >20 l
per year) (r = 0.123, P < 0.000). Although the association
remained significant among the heavy drinkers, possibly
because of the high number of participants, the weak
587
correlation coefficient indicates that only a small proportion
of the total variance in scores could be explained by this relationship. Table 6 shows changes over time in the mean
number of risk factors for alcohol misuse when all the participants were included in the analysis. The data indicate a
trend towards fewer risk factors over time, but also some
variability in scores between years. Unlike the consumption
data, which increased in dispersion over time, indicating
more heavy drinkers and a likely polarization effect, the dispersion of the risk factor total score reduced significantly
between 2000 and 2010 (F = 20.32, P < 0.001) (Levine’s
test). This trend was observed for both school years and
genders, and when drinkers only and all youth (including
abstainers) were examined separately. Other measures of dispersion, including the coefficient of variation, indicated no
significant changes in either direction.
Table 7 shows the mean number of risk factors by year for
the heavy drinkers only (i.e. young people consuming 20 l or
more alcohol). The heaviest drinkers report more risk factors
for alcohol misuse compared with the majority of young
people who consumed less alcohol, but with large variability
between years. There were no significant changes in the
number or dispersion of risk factors over time in this category.
DISCUSSION
Our findings indicate that a polarization in youth drinking is
a likely explanation for the recent divergence between per
capita alcohol consumption, which has decreased, and
alcohol-related harmful effects, which have increased sharply
among Stockholm youth over the past decade. For most
young people, consumption reduced between 2000 and
2010, while the heaviest consumers (those in the top 5–10%
of the drinking distribution) mostly increased their consumption over the same period. There was only one exception to
Table 6. Risk factor total score by year (all the participants)
Year
2000
2002
2004
2006
2008
2010
(n = 6302) (n = 6929) (n = 6330) (n = 7008) (n = 7583) (n = 8092)
Mean 3.41
Median 3
SD
2.08
2.91
3
2.02
2.72
2
1.92
3.11
3
2.05
3.17
3
2.06
3.08
3
2
The mean, median and SD for the risk factor total score by year when all
the participants are included in the analysis. The data indicate a trend
towards fewer risk factors over time. However, the relatively large SD and
the variability in scores between years suggest that the data are inconclusive.
Table 7. Risk factor total score by year (participants drinking 20 l of alcohol
or more per year)
Fig. 4. Estimated yearly frequency of binge drinking. The figure shows
changes over time in the estimated frequency of yearly binge drinking
(number of times per year) by age and school year.
Year
2000
(n = 78)
2002
(n = 449)
2004
(n = 737)
2006
(n = 621)
2008
(n = 587)
2010
(n = 448)
Mean
Median
SD
6
6
2.08
5.21
5
2.4
2.97
3
1.88
3.33
3
2
3.63
3
2.03
5.49
5
2.26
The mean, median and SD for the risk factor total score by year for the
heaviest drinkers only. The high variability in scores between years indicates
that the data are inconclusive.
588
Hallgren et al.
this trend, with the younger females in Year 9 reducing their
per capita consumption across all levels of drinking. Yet
even here, the reductions over time were smaller among the
heaviest drinkers compared with the moderate drinkers—a
trend consistent with the polarization hypothesis. Across all
the four groups, the dispersion of per capita consumption
also increased significantly over time, indicating an increasing number of heavy drinkers in the tail end of the distribution. Together, these changes indicate that more young
people drink at extremely high levels over time, while the
majority continues to drink less.
Of particular concern are females aged 18–19 years. These
young women report the largest increases in both the total
volume of alcohol consumed and the frequency of binge drinking. Although we were not able to connect the anonymous selfreport data with the hospitalization data, the dramatic increase
in alcohol-related hospitalizations among females aged 15–24
years in Stockholm suggests that these young women are overrepresented in serious alcohol-related harmful effects and represent a high-risk group for alcohol misuse.
The divergence between per capita consumption and
alcohol-related harm has been observed elsewhere, so the polarization hypothesis may well apply to other populations;
not only those in Sweden. In the UK, there has been a
marked reduction in per capita consumption since ~2000
among 16–24-year olds (Meier, 2010). At the same time,
alcohol-attributable hospital admissions have been increasing
at a rate of 11% year on year, with an estimated 945,000
alcohol-related admissions in England in 2008/2009 (around
7% of all admission). Similarly, in Victoria, Australia, seven
key measures of alcohol-related harm increased in the adult
population between 1999 and 2008, while total consumption
remained relatively stable (Livingston, 2008). Possible explanations for the divergence suggested by the authors included
a rise in heavy episodic drinking (especially among females),
an increased preference for higher alcohol content beverages
and a polarization in drinking—although the latter possibility
was not tested empirically.
The overall increase in alcohol-related hospitalizations in
Stockholm could be driven by changes in the drinking behaviour of a relatively small but high-risk group, such as the
heavy-drinking 18–19-year-old females identified in this
study. This cohort may not be large enough to have much
impact on per capita consumption, but large enough to influence hospital admission data, resulting in the present divergence between consumption and harm. A similar idea has
been suggested by Mäkela and Österberg (2009) to explain
the relatively large impact of a reduction in alcohol taxes in
Finland on alcohol-related problems compared with per
capita consumption.
On the aggregate level, the mean number of risk factors
for alcohol misuse decreased between 2000 and 2010.
However, the reduction was not consistent over time and
there was intervening variability in scores. We predicted that
the total number of risk factors reported by the heaviest drinkers would increase over time, mirroring the pattern seen in
the consumption data, but this did not occur. Instead, we
found a non-significant reduction in the total risk factor
score for the heaviest drinkers, but with large variability in
scores. The strength of the association between the risk
factor total score and alcohol consumption was large when
all the participants were examined (r = 0.468), but reduced to
a small association among the heaviest drinkers only (r =
0.123). This weakened relationship might explain the
increased variability in scores among the heavy drinkers, as
the selected risk factors were responsible only for a small
proportion of the total variance in alcohol consumption.
Given these findings, the evidence for a polarization effect in
the risk factor data is inconclusive.
It is likely that young people who routinely drink to
excess are qualitatively different from their ‘moderate’ drinking peers. Adolescents who regularly drink to extremely
high levels typically report more serious and on-going social
and psychological problems, making themselves a unique
subgroup in this respect (Petraitis et al., 1995; Zufferey
et al., 2007). Consequently, the risk factors that apply to the
moderate drinkers in the present study may not necessarily
influence the behaviour of the heavy drinkers the same way
(and visa versa). Risk factors for alcohol misuse differ
according to gender (Danielsson et al., 2010) and substance
type (Becker and Grilo, 2005), so it is conceivable that they
could also vary according to the level of drinking, especially
when consumption is exceptionally high. This possibility is
supported by two recent studies that found that the risk
factors for alcohol misuse in a sample of alcohol-dependent
adolescents were different from the risk factors commonly
reported in community samples where drinking levels tend to
be much lower (Becker and Grilo, 2005; Nation and
Helfinger, 2006).
Wider economic and social changes in Sweden have influenced recent drinking trends. Alcohol became more available
and affordable as Sweden joined the EU in 1995 and associated trade restrictions were eroded. This general increase in
the availability of alcohol may have disproportionately
affected some people more than others. A recent Finnish
study found that large reductions in the price of alcohol have
led to substantial increases in alcohol-related mortality,
mainly among individuals from lower socio-economic backgrounds (Herttua et al., 2008). On a societal level, there have
been shifts in the distribution of wealth in Sweden, which
have resulted in increased social and economic inequalities,
especially since the 1990s (Klevmarken, 2006). A study by
Fritzell et al. (2007a,b) examining changes in the living conditions of young people in Sweden between 1994 and 2005
(the period immediately following the last economic recession) also found a polarization tendency on three central
dimensions of welfare: work and employment, economic
resources and health, especially mental health. The full
impact of the current global economic crisis remains to be
seen. However, it is conceivable that these social changes
now affect young people in the form of greater disparities,
which are associated with a higher incidence of social problems generally, including heavy drinking.
Our findings have implications for alcohol policy, both
in Sweden and elsewhere. First, our data are an important
reminder that changes in per capita consumption can hide
significant shifts in the drinking habits of heavy drinkers.
To see the complete picture, changes in the dispersion of
drinking relative to per capita consumption should also be
examined. Policy decisions based on changes in per capita
consumption alone are insufficient because this can hide the
emergence of heavy drinking subgroups. For the moment,
we need to know more about the social backgrounds and the
risk factors these young people incur, so that targeted
Polarized youth drinking
interventions can be developed to reduce their current levels
of drinking and associated harmful effects.
Although we have not set out to empirically test Skog’s
theory of the collectivity of drinking cultures—and some
have questioned whether this can be done (e.g. Gmel and
Rehm, 2000), our findings have implications for this highly
influential theory. Given the reductions in per capita consumption seen here, Skog’s theory would predict roughly
parallel reductions in the heaviest drinkers, but this did not
occur. Skog has noted that smaller, high-risk groups might
be less influenced by the drinking behaviour of those around
them because of social isolation, exclusion or other individual traits. If this is true, then we need to understand why
alcohol policies appear to influence the drinking behaviour
of some individuals more than others. Previous researchers
have observed diverging patterns of alcohol consumption in
subgroups (e.g. Gustafsson, 2010), and other exceptions to
Skog’s general rule of collectivity (Stockwell et al., 1997).
What the present study adds is an empirical confirmation
that the polarization phenomenon occurs among youth in the
Swedish context. Future research should examine the alcohol
and risk factor polarization hypotheses in different contexts,
and explore associations between changes in consumption
and both the number and the type of risk factors that young
people are exposed to. What may also be needed is analysis
of risk factors over time at community and societal levels, including measures of income and social inequality, which
have widened in many parts of the world. Such analyses
should be combined with a theory of how societal increases
in inequality are linked to individual-level risks for alcohol
misuse, and the various mechanisms that are involved.
Strengths and limitations
The Stockholm Student Survey provides a unique opportunity to closely examine changes in the drinking habits of
young people. Both the number of participants and the response rates are consistently high, helping to ensure that the
sample is representative. The quality of the data has enabled
us to empirically test the polarization hypothesis using both
alcohol consumption and risk factor data. The results have
implications for alcohol prevention in Sweden and possibly
in other countries where similar patterns have been observed.
Our findings are based on self-report surveys and the
inherent limitations of such surveys are well known.
Respondents tend to under-report the amount of alcohol they
consume, particularly at high levels (Northcote and
Livingston, 2011). Previous Swedish studies suggest that
adolescent non-responders are more likely to be heavy consumers than those who do respond (Romelsjö and Branting,
2000). It is also possible that some of the heaviest drinkers
were excluded from the survey because they were not attending school when the survey was conducted. Again, this
could lead to an underestimate of consumption. However,
our reliance on self-report data does not invalidate our findings; anonymous self-reports are generally valid, provided
confidentiality is stressed, which it was in this survey
(Campanelli et al., 1987). After the year 2000, the survey
was expanded to include additional risk and protective
factors. Some of these new factors (e.g. number of heavy
drinking friends, social support, etc.) are relevant, but were
not included in the analyses because they were absent from
589
the 2000 survey, and therefore, could not be cross-matched
with the 2010 data. Using a theory-driven approach to select
the risk factors (as opposed to a statistical approach) enabled
us to see whether there had been a change over time in the
same 13 risk factors, both in the total sample and among the
heaviest drinkers. This approach may result in a different
number and/or collection of risk factors, compared with a
statistical approach driven by logistic regression modelling.
Finally, as the questionnaires were anonymous, it was not
possible to follow-up non-responders and compare them
with the survey participants. We also acknowledge that many
factors other than alcohol use per se can influence alcoholrelated hospitalization data, including administrative changes
to patient admissions.
Funding — This study was supported with funding from the Karolinska Institute (KID
funding for doctoral education, Department of Public Health Sciences).
REFERENCES
Ahacic K, Thakker D. (2010) Alkohol och narkotikarelaterad
vårdkonsumtion och dödlighet I olika åldersgrupper i
Stockholms län 1998–2008, rapport 2009. (Alcohol and drug
related hospitalisations and deaths in different age groups in
Stockholm Municipality 1998–2008, Report 2009). Karolinska
Institute, Public Health Academy, 2010:7.
Babor T, Caetano R, Casswell S et al. (2010) Alcohol: No Ordinary
Commodity, 2nd edn. New York: Oxford University Press.
Becker DF, Grilo CM. (2005) Prediction of drug and alcohol abuse
in hospitalized adolescents: comparisons by gender and substance type. Behav Res Ther 44:1431–49.
Bränström R, Sjöström E, Andréasson S. (2008) Individual, group
and community risk and protective factors for alcohol and drug
use among Swedish adolescents. Eur J Public Health 18:12–18.
Campanelli PC, Dielman TE, Shope JT. (1987) Validity of adolescents self-reports of alcohol use and misuse using a bogus pipeline procedure. Adolescence 22:7–22.
CAN. (2010) Drug trends in Sweden 2010. Central agency for
studies on alcohol and narcotic use.
Cohen LR, Potter LB. (1999) Injuries and violence: risk factors and
opportunities for prevention during adolescence. Adolesc Med
10:125–35.
Danielsson A-K, Romelsjö A, Tengström A. (2010) Heavy episodic
drinking in early adolescence: gender-specific risk and protective
factors. Subst Use Misuse 46:633–43.
Duffy JC. (1986) The distribution of alcohol consumption—30
years on. Br J Addict 81:735–41.
El-Khouri B, Sundell K, Strandberg A. (2005) Riskfactorer for
normbrytande beteenden (Risk factors for normbreaking behaviour). Rapport 2005:17. Stockholm Fou-enheten, Stockholms Stad.
Fritzell J, Gähler M, Nermo M. (2007a) Vad hände med 1990-talets
stora förlorargrupper? Välfärd och ofärd under 2000-talet. (What
happened to the disadvantaged groups of the 1990s? Welfare
and disadvantage during the first years of the new millennium).
Socialvetenskaplig tidskrift 2–3:110–33.
Fritzell J, Lennartsson C, Lundberg O. (2007b) Health and inequalities in Sweden: long and short term perspectives. In Fritzell IJ,
Lundberg O. (eds). Health Inequalities and Welfare Resources:
Continuity and Change in Sweden. Bristol: Policy Press.
Gmel G, Rehm J. (2000) The empirical testability of Skog’s theory
of collective drinking behaviour. Drug Alcohol Rev 19:391–9.
Gustafsson NK. (2010) Changes in alcohol availability, price and
alcohol-related problems and the collectivity of drinking cultures: what happened in southern and northern Sweden? Alcohol
Alcohol 45:456–67.
Hawkins JD, Catalano RF, Miller RI. (1992) Risk and protective
factors for alcohol and other drug problems in adolescents and
early adulthood: implications for substance use prevention.
Psychol Bull 112:64–105.
590
Hallgren et al.
Herttua K, Mäkela P, Martikainen P. (2008) Changes in alcohol
related mortality and its socioeconomic differences after a large
reduction in alcohol prices: a natural experiment based on register data. Am J Epidemiol 168:1126–31.
Klevmarken A. (2006) The distribution of wealth in Sweden: trends
and driving factors. Working Paper 2006:4. Department of
Economic, Uppsala University.
Livingston M. (2008) Recent trends in risky alcohol consumption
and related harm among young people in Victoria. Aust NZJ
Public Health 32:266–71.
Mäkela P, Osterberg E. (2009) Weakening of one more alcohol
control pillar: a review of the effects of the alcohol tax cuts in
Finland 2004. Addiction 104:554–63.
Meier P. (2010) Polarized drinking patterns and alcohol deregulation. Trends in alcohol consumption, harms and policy,
United Kingdom 1990–2010. Nordisk Alkohol Nark 27:
383–405.
Merline A, Jager J, Schulenberg JE. (2008) Adolescent risk factors
for adult alcohol use and abuse: stability and change of predictive value across early and middle adulthood. Addiction
83:84–99.
Nation M, Heflinger CA. (2006) Risk factors for serious alcohol
and drug use: the role of psychosocial variables in predicting the
frequency of substance use among adolescents. Am J Drug
Alcohol Abuse 32:415–33.
Norström T, Ramstedt M. (2005) Mortality and population drinking:
a review of the literature. Drug Alcohol Rev 24:537–47.
Northcote J, Livingston M. (2011) Accuracy of self-reported drinking: observational verification of ‘last occasion’ drink estimates
of young adults. Alcohol Alcohol 46:709–13.
Petraitis J, Play BR, Miller TQ. (1995) Reviewing theories of adolescent substance use: organizing pieces in the puzzle. Psychol
Bull 117:67–8.
Romelsjö A, Branting M. (2000) Consumption of illegal alcohol
among adolescents in Stockholm county. Contemp Drug Probl
27:315–33.
Rose G, Day S. (1990) The population mean predicts the number of
deviant individuals. Br Med J 391:1031–4.
Skog OJ. (1985) The collectivity of drinking cultures: a theory of
the distribution of alcohol consumption. Br J Addict 80:83–99.
Stockwell T, Single E, Hawks D et al. (1997) Sharpening the focus
of alcohol policy from aggregate consumption to harm and risk
reduction. Addict Res 5:1–9.
Valdatabasen. (2010) A public hospital registry containing information about alcohol related hospital admissions in Sweden,
recorded by ICD-10 diagnosis codes. Stockholm, Sweden.
Zufferey A, Pierre-André M, Jeannin A et al. (2007) Cumulative
risk factors for adolescent alcohol misuse and its perceived consequences among 16 to 20 year old adolescents in Switzerland.
Prev Med 45:233–9.