A ddiction (1998) 93(5), 689± 699
A LC O H O L A ND A G G RE SSIO N
Effects on crim inal violence of different
beverage types and private and public
drinking
È M
TH OR NORSTRO
Swedish Institute for Social Research, Stockholm University, Sweden
A bstract
A im . The purpose of the stud y is to analyse the relationship between hom icide and assault rates on one hand,
and various indicators of alcohol consu m ption on the other. The latter includ e private and public drinking
with a further disaggregation into beverage-speci® c drinking. M easurem ents. The da ta com prise aggregate
tim e series for Sweden during the period 1956± 94. The assault rate is m easured as the num ber of police
reported assaults (at all degrees of aggravation) per 100 000 inhabitants (15 1 ). The homicide rate is
m easured as the num ber of hom icides (w here the victim was at least one year old) per 100 000 inhabitants
(15 1 ). Private consu mption is gauged as retail sales of alcohol, and public consu m ption as on -prem ise sales
(litres 100% per inhabitant, 15 1 ). These two measures are disaggregated further into beverage speci® c sales
(beer, spirits and wine). F indings. A ccordin g to the ® ndings, there is a statistically signi® cant relationship
between the assault rate and a combined measure of on-prem ise sales of beer and spirits. The estimated
relationship corresponds to an attributable fraction of about 40% . The hom icide rate is signi® cantly associated
with retail sales of spirits; the attributable fraction is estimated at about 50% . W ine sales are not related to
any of the tw o violence indicators. C onclusions. The ® ndings suggest that the assault rate is related to
consu mption of beer and spirits in bars and restaurants, while the hom icide rate is linked to consu mption of
spirits in private contexts. The ® ndings , notably speci® c to Sweden during a certain tim e period, can be
interpreted as the outco m e of the interplay of a number of factors, including opportunity structur e, social
control and contex t of drinking, drinking patterns associated with the different beverage types and characteristics of the drinkers.
Introdu ction
Several kinds of evidence suggest a link between
alcohol consumption and various form s of
aggression or violence. A m eta-analysis of 30
experimental studies reveals that consumption of
alcohol tends to elicit aggressive behaviour
(Bushman & Cooper, 1990). Situations that
engender frustration seem to increase the likelihood of this response (Gustafson, 1986). Further, crime statistics shows that the majority of
violence offenders are intoxicated when comm itting the crim e (Roslund & Larson, 1979;
M ay® eld, 1976; KuÈ hlhorn et al. , 1984; WikstroÈ m , 1985). In addition, analyses of aggregate
A version of this paper w as presented at the International Conference on Intoxication and Aggressive Behaviour:
U nderstanding and Preventing Alcohol-Related Violence, Toronto, Ontario, Canada, October 1996 .
Correspondence to: Thor NorstroÈ m , Swedish Institute for Social Research, Stockholm U niversity, S 10691
Stockholm , Sweden. T el.: 1 468 16 23 14; Fax: 1 468 15 46 70; e-mail: totto@ so® .su.se
Subm itted 14th March 1997; initial review completed 22nd July 1997, ® nal version accepted 16th Decem ber 1997.
0965 ± 2140 /98/0500689 ± 11 $9.50
Carfax Publishing Limited
Ó
Society for the Study of Addiction to Alcohol and Other Drugs
690
Thor NorstroÈ m
time series data indicate a relationship between
per capita alcohol consumption and criminal
violence (Skog & Bjù rk, 1988; Lenke, 1990).
Considering the great cultural variation in
drunken com portm ent (M acAndrew & Edgerton, 1969) it seem s plausible that this relationship would vary across nations. This notion is
supported by the ® nding that the link between
population drinking and violence is fairly weak in
France, a typically wet culture, while it is of
substantial magnitude in the dry cultures of the
Nordic countries of Finland, Norway and Sweden (Lenke, 1990).
The purpose of this paper is to elaborate
aggregate level evidence on the alcohol± violence
link by decomposing the impact of overall consum ption on violence into effects of private and
public consum ption, and further into beveragespeci® c effects. Tw o forms of criminal violence
are considered, assaults and homicide.
One way of generating hypotheses about how
the various factors are interrelated is to consider
the social contexts of the two form s of violence.
With regard to assaults, m ost cases take place in
a public place, particularly restaurants and bars
(WikstroÈ m, 1985). The reason for this m ay in
part be artefactualÐ that is, the higher visibility
of these offences leads to a higher probability of
police reporting. However, the high density of
social interaction in restaurants and bars should
also be of importance in this context. On the
basis of this observation, Lenke (1990) suggests
the hypothesis that the assault rate should be
more strongly associated with public than with
private drinking. This hypothesis is also supported by his analysis of Swedish data. In contrast, the great majority of homicides take place
in a private contextÐ among family mem bers,
friends or acquaintances, with the crime scene
being an apartm ent (according to WikstroÈ m’ s
data (1992), this is the context for about threequarters of the hom icides). Therefore, we should
expect private consumption to be the more
important subcategory of alcohol consum ption
in relation to the homicide rate.
Is there anything to suggest whether beverage
type m atters with respect to violence? A number
of experim ental studies have addressed this issue
(e.g. Boyatzis, 1974; Gustafson, 1985a, 1985b).
The ® ndings are not entirely conclusive although
there are indications that distilled alcohol would
be more likely than brewed alcohol to evoke
aggression (for a review see Smart, 1996). The
question is how these results translate into natural settings. The comm ent of Graham, Schm idt
& G illis (1996) in this context, that beveragespeci® c effects have m ore to do with the social
de® nition of the beverage than with pharmacology, seems plausible. Whether beverage matters
thus seems to be an open question.
Data and m ethod
For the purpose of the present inquiry, overall
consumption of alcohol during the period 1956±
94 is disaggregated, ® rst into private consumption (retail sales) and public consumption
(on-premise sales). Within these two categories a
breakdown is made by beverage type; beer, spirits and wine. The disaggregation is based on
of® cial statistics on sales ® gures which are complete with respect to wine and spirits (source:
Alkohol- och na rkotikautveckling en i Sverige
(1996), and for proportions of sales on premise
before 1985: Alkoholstatistik , yearly issues for the
period 1956± 85). As to beer, only the proportion
of strong beer sold on premise is recorded. This
means that other beer (containing less than 3.5%
alcohol by volum e) is not included in the disaggregated m easures. All series are expressed in
litres of pure alcohol per inhabitant 15 years and
above. The assault rate is the number of police
reported assaults (at all degrees of aggravation)
per 100 000 inhabitants 15 years and above
(source: Brottsutveckling en 1992 och 1993 , 1994)
(Table 1).
The assault rate was probably affected by
changes in the statistical routines in 1965 and
Table 1. Descriptive statistics of indicators of violence and
alcohol consumption, 1956± 1994 . Crime measures are rates
per 100 000 inhabitants 15 years and above; consumption
measures are annua l sales (litres of pure alcohol) per
inhabitant 15 years and above
M ean
Assaults
Hom icide
Total consumption
Private consum ption
Beer
Spirits
W ine
Public consumption
Beer
Spirits
W ine
715.9 2
1.27
6.29
4.43
0.37
2.91
1.16
0.40
0.18
0.14
0.09
S D
354.0 7
0.34
0.79
0.60
0.29
0.59
0.45
0.15
0.16
0.05
0.03
Crim inal violence and different beverage types
1400
1200
1000
800
600
400
200
1956 1961 1966 1971 1976 1981 1986 1991
Year
Figure 1. Number of police reported assaults per 100 000
inhabitan ts 15 years and above, Sweden 1956 ± 94.
2.0
No. of homicides
1.8
1.6
1.4
1.2
1.0
0.8
0.6
0.4
1956 1961 1966 1971 1976 1981 1986 1991
Year
Figure 2. Number of homicides per 100 000 inhabitants
15 years and above, Sweden 1956 ± 94.
1968, and by a change in the law in 1982.
Complaints which were determined by police
investigations to be unfounded were eliminated
from the statistics before 1965 whereas such
cases remained in the ® les from 1965 (Krim inalstatistik, 1965; Kriminalstatistik, 1968 och
1969). In 1968 the principle for counting the
number of crim es was changed from counting
the number of complaints to counting the number of crimes reported in the complaint (Kriminalstatistik, 1968 och 1969). The 1982 change
in the law meant the following: before 1982 it
was possible for the victim to withdraw a complaint if the assault had taken place on private
premises, e.g. an apartment. From 1982 this was
no longer possible (Brottsutvecklingen, 1987). It
appears that the three changes described above
may well have in¯ ated the recorded assault rate,
and they are therefore included in the models as
dummy variables which take the value 0 before
the year of the change, and the value 1 otherwise.
The homicide rate is the number of homicides
(where the victim was at least 1 year old) per
100 000 inhabitants 15 years and above (source:
DoÈ dsorsaker , yearly issues for the period 1956±
94). Since the hom icide rate is based on the
cause of death statistics it is not expected to be
affected by the changes described above.
The time trends during the study period
(1956± 94) in the violence indicators as well as in
total, private and public consumption are
depicted in Figs 1± 5. As can be seen, there is
Total alcohol consumption (litres 100%)
1600
8.0
7.5
7.0
6.5
6.0
5.5
5.0
4.5
1956 1961 1966 1971 1976 1981 1986 1991
Year
Figure 3. Total alcohol consumption (litres 100 %) per
capita 15 years and above, Sweden 1956± 94.
Private alcohol consumption (litres 100%)
No. of police reported assaults
1800
691
6.0
5.5
5.0
4.5
4.0
3.5
3.0
1956 1961 1966 1971 1976 1981 1986 1991
Year
Figure 4. Private alcohol consumption (litres 100%) per
capita 15 years and above, Sweden 1956± 94.
Public alcohol consumption (litres 100%)
692
Thor NorstroÈ m
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
1956 1961 1966 1971 1976 1981 1986 1991
Year
Figure 5. Public alcohol consumption (litres 100%) per
capita 15 years and above, Sweden 1956± 94.
a marked increase in both of the violence indicators, particularly in the assault rate. The developm ent of alcohol consum ption per capita is
more irregular, ® rst dominated by a positive
trend until the end of the 1970s, and thereafter
a substantial decline. Public consumption is stationary until the mid-1970s, and then grows
fairly steadily. However, it would be prem ature
to infer anything about the possible relationship
between alcohol and violence on the basis of the
patterns observed in these ® gures. Skog (1988)
provides several examples of how misleading
inferences from trending time series may be.
The data will be analysed with the technique
for time series analysis that has been suggested
by Box & Jenkins (1976), often referred to as the
ARIMA-model. Since this method requires stationarity and our data exhibit strong time trends,
the analysis will be performed on the differenced
seriesÐ in other words, rather than using raw
series the yearly changes are analysed. This procedure reduces greatly the risk of obtaining spurious correlations since an omitted variable is
more likely to be correlated with alcohol consum ption due to comm on trends than due to
synchronization in the yearly changes. Further,
the noise term, which includes explanatory variables not considered in the model, is allowed to
have a temporal structure that is m odelled and
estimated in terms of autoregressive or m oving
average parameters.
Time series analysis of aggregate data is certainly associated with a number of complications
which may com promise the ® ndings (for a dis-
cussion see Skog, 1988). However, aggregate
® ndings have at least one merit com pared to
individual-level results which is worthy of attention. The major limitation of ® ndings from individual-level data is the possibility of drift, or
selection effects (Lieberson, 1985). In contrast,
an aggregate effect estim ate is expected to
express the exogenous impact of the risk factor,
net of selection effects. The rationale for this
assertion is that the tem poral variation of the risk
factor, e.g. alcohol consum ption (the effect of
which is measured by the regression coef® cient
in a time series analysis) re¯ ects changes in
exposure, but not changes in self-selection. Analytically, this can be shown as follows. Consider
the following micro m odel:
Y it 5
Ci 1
b X it 1
P it
where Y it is the response of individual i on the
outcome variable at tim e t ; C i is the sum total of
those unmeasured tim e-invariant attributes
which are correlated with the outcom e; and P it
is a random error term. The potential bias in the
estimation of b is due to the correlation between
C i and X it that m ay arise because of self-selection. Now consider the model in aggregate form :
Yt 5
C1
b Xt 1
Pt
As we can see the aggregation turns C i into a
constant which by de® nition is uncorrelated with
X t, and thus the source of bias is eliminated (for
a more detailed discussion see NorstroÈ m ,
1988a).
The scheme of analysis is as follows. First the
impact of total consumption is estimated. Next
the impact of private vs. public consumption is
focused upon and ® nally the beverage speci® c
indicators. In addition to the estimated effect
param eter, the alcohol effect is also expressed in
terms of its attributable fraction (AF ). The m easure is usually calculated from individual-level
data (Lilienfeld & Lilienfeld, 1981) but it is also
possible to obtain it from aggregate data by using
the expression:
AF 5
b *X
Y
where b
is the unstandardized regression
coef® cient which is estim ated in a linear model
(the effect of a one unit change in the input
series), X is the input series (alcohol consumption), and Y the output series (violence). (Here
Crim inal violence and different beverage types
693
Table 2. Effect estimate of total alcohol consumption (litres of pure alcohol per
inhabitant 15 years and above) on assault rate. Linear model estimated on
differenced Swedish data, 1956± 1994
Estim ate
Consum ption
Dumm y65
Dumm y68
Dumm y82
Noise
AR(1)
AR(2)
AR(3)
Constant
Diagnostics
Q* (lag 10)
2
SE
p
53.76
58.30
95.99
99.68
25.15
34.94
34.97
33.47
0.04
0.11
0.01
0.006
0.38
0.07
0.45
34.56
0.17
0.21
0.25
22.41
0.03
0.72
0.08
0.13
5.51 (p .
AF
0.47
0.85)
*Test for residual correlation.
Table 3. Effect estimates of private and public alcohol consumption (litres of
pure alcohol per inhabitant 15 years and above) on assault rate. Linear model
estimated on differenced Swedish data, 1956± 199 4
Estimate
Private
Public
Dummy65
Dummy68
Dummy82
Diagnostics
Q* (lag 10)
2 4.85
889.90
64.01
62.15
75.64
10.28 (p .
SE
27.81
190.34
34.62
34.12
34.25
,
p
0.86
0.001
0.07
0.08
0.03
AF
0.51
0.41)
*Test for residual correlation. Correlation between estim ates of private
and public 5 0.02.
the period averages of X and Y are used.) In the
present context, AF is interpreted as the fraction
of violence that according to the estim ates is
attributable to alcohol (for a further discussion of
computing AF from aggregate data, see
NorstroÈ m , 1989).
R esults
Assaults
As can be seen from Table 2, changes in overall
consumption are signi® cantly associated with
changes in the assault rate. If we focus on the
attributable fraction, which is the measure that
has the m ost straightforward interpretation, it
appears that 47% of the assaults would be
attributable to alcohol according to the results. It
is of som e interest to compare this estimate with
® ndings from individual-level data concerning
the proportion of offenders that are intoxicated
at the event of the crim e. It seems sensible to
restrict the com parison to ® ndings pertaining to
Sweden. KuÈ hlhorn et al. (1984) report that 82%
of the offenders who were convicted of crim es of
violence in 1975 were under the in¯ uence of
alcohol. The corresponding ® gure in WikstroÈ m’ s
(1985)
studyÐ comprising
police-reported
crim es of violence in the city of G aÈvle, 1968± 70
and 1973± 75Ð is 75%. The author points out
that quite strict criteria were applied for classifying an offender as intoxicated. The cited ® gures
are markedly higher than the alcohol attributable
fraction that was estimated from the aggregate
data. A reasonable explanation of this difference
is that the ® ndings at the individual level are
affected by selection effects in the way that the
offenders have a higher than average frequency
of intoxication, but that this state is not always a
cause contributing to the commission of the
crim e. This notion accords well with the ® ndings
694
Thor NorstroÈ m
Table 4. Effect estimates of public alcohol consumption of beer, spirits and wine
(litres of pure alcohol per inhabitan t 15 years and above) on assault rate. Linear
model estimated on differenced Swedish data, 1956 ± 1994
Estimate
Beer
Spirits
W ine
Dummy65
Dummy68
Dummy82
Noise
Constant
Diagnostics
Q* (lag 10)
SE
959.4 0
1127.0 2
2 378.7 7
65.45
61.60
84.97
244.7 0
675.6 2
938.0 7
33.87
33.61
35.02
18.13
8.28
13.48 (p .
AF
p
,
0.001
0.11
0.69
0.06
0.08
0.02
0.23
(0.31)
0.04
0.19)
*Test for residual correlation. Correlation betw een estimates of beer and
spirits 5 2 0.30; correlation between estim ates of beer and wine 5 0.10;
correlation between estim ates of wine and spirits 5 2 0.39.
Table 5. Effect estimates of public alcohol consumption of beer plus spirits and wine
(litres of pure alcohol per inhabitant 15 years and above) on assault rate. Linear
model estimated on differenced Swedish data, 1956 ± 1994
Estimate
Beer 1 spirits
W ine
Dumm y65
Dumm y68
Dumm y82
Noise
Constant
Diagnostics
Q* (lag 10)
2
SE
989.33
303.80
65.95
61.54
86.19
197.52
856.74
33.28
33.11
34.03
16.95
6.08
13.53 (p .
*Test for residual correlation.
beer 1 spirits and wine 5 2 0.15.
from a prospective, longitudinal study (White,
Brick & Hansell, 1993) which focuses the
dynamics between alcohol use and aggression.
They ® nd that early aggressive behaviour is
linked to later alcohol use and alcohol-related
aggression; `It appears, therefore, that individuals who engage in alcohol-related aggression are
aggressive from early adolescence and behave
aggressively whether or not they use alcohol.’
(White et al. , 1993, p. 74.) The overall conclusion would be that the large proportion of
intoxicated offenders in violent crimes tends to
exaggerate the role of the alcohol factor because
these individuals also have an elevated violence
risk in the absence of intoxication. As described
in the previous section, estim ates obtained from
aggregate data are expected not to be contaminated by this kind of selection effect.
Finally, it is noted that the estimated effects of
,
p
AF
0.001
0.73
0.06
0.07
0.02
0.43
0.01
0.19)
Correlation
between
estim ates
of
the dummy variables representing changes in
statistical routines and in the law are in the
expected direction.
In the next m odel (Table 3) the indicators of
private and public drinking are included, and it
appears that only the latter has a statistically
signi® cant association with the assault rate. The
outcome accords with the expectations, but the
estimation is potentially complicated by the
possible strong correlation between the two consumption indicators, which would give rise to
co-linearity and dif® culties in disentangling their
separate effects. As a matter of fact the correlation is close to zero (0.01), which is also m irrored in a similarly m inuscule correlation
between the estim ates of private and public
drinking.
In the ® nal model for assaults (Table 4) the
aim is to assess the beverage-speci® c in¯ uences
Crim inal violence and different beverage types
695
Table 6. Effect estimate of total alcohol consumption (litres of pure alcohol per
inhabitant 15 years and above) on homicide rate. Linear model estimated on
differenced Swedish data, 1956± 1994
Estimate
Consum ption
Noise
M A(1)
Constant
Diagnostics
Q* (lag 10)
SE
0.14
0.0 6
0.67
0.02
0.14
0.01
8.22 (p .
p
0.03
,
AF
0.69
0.001
0.10
0.60)
*Test for residual correlation.
Table 7. Effect estimates of private and public consumption (litres of pure
alcohol per inhabitant 15 years and above) on homicide rate. Linear model
estimated on differenced Swedish data, 1956± 199 4
Estim ate
Private
0.19
Public
0.35
Noise
AR(1)
2 0.68
AR(2)
2 0.25
Diagnostics
Q* (lag 10) 7.40 (p . 0.77)
SE
0.10
0.58
P
0.17
0.16
,
0.07
0.55
AF
0.6 6
0.001
0.13
*T est for residual correlation. Correlation betw een estim ates of private
and public 5 0.19.
of publicly consum ed alcohol. The outcom e is
somewhat ambiguous, because in this case the
alcohol indicators exhibit stronger intercorrelations which makes it harder to separate their
associations with the assault rate. The only indicator that is clearly signi® cant is that of beer;
spirits are below conventional limits of
signi® cance, but the correlation between the estimates of the beer and spirits param eters indicates som e indeterm inacy. (The computation of
AF , particularly for spirits, becomes questionable
in this case.) The estimate of the wine effect is
far below signi® cance but correlated with the
estimate of the spirits effect, which warrants
some caution in the interpretation. Since we
cannot accurately separate the effects of beer and
spirits it seems sensible to com bine these two
indicators into one single measure. The outcome
of such a model is displayed in Table 5. As can
be seen, the combined measure is highly
signi® cant. M oreover, the estim ate of the wine
effect rem ains insigni® cant but it is now only
weakly correlated with the estim ate of the
combined indicator, which makes the outcome
more conclusive.
Hom icide
In view of the small number of homicides, averaging 83 per year during the study period, it is
far from obvious that the model estimation has
power enough to yield any signi® cant alcohol
effect even if it actually is present. However, as
can be seen in Table 6, the estimated relationship between total consumption and the hom icide rate is statistically signi® cant. The
associated attributable fraction (0.69) is larger
than the one observed for assaults, although the
difference should not be stressed too much considering the margins of error. When the alcohol
effect is decomposed into effects originating
from private and public drinking it appears that
only the private one is statistically signi® cant
(Table 7). Finally, of the privately consum ed
alcohol only spirits seem to be of any importance
(Table 8). In none of the models for homicide is
the problem of co-linearity present.
696
Thor NorstroÈ m
Table 8. Effect estimates of private consumption of beer, spirits and wine (litres
of pure alcohol per inhabitant 15 years and above) on homicide rate. Linear
model estimated on differenced Swedish data, 1956 ± 1994
Estim ate
Beer
Spirits
W ine
Noise
AR(1)
AR(2)
Constant
Diagnostics
Q* (lag 10) 4.97 (p .
2
2
2
2
SE
0.19
0.23
0.10
0.33
0.10
0.49
0.72
0.29
0.03
0.17
0.17
0.02
P
,
0.58
0.04
0.84
AF
0.5 3
0.001
0.09
0.17
0.89)
*T est for residual correlation. Correlation between estimates of beer
and spirits 5 2 0.02; correlation between estim ates of beer and
wine 5 0.07 ; correlation between estim ates of wine and spirits 5 2 0.08.
D iscussion
In this paper the total impact of alcohol consum ption per capita on assaults and homicide
has been decom posed ® rst into effects of private
and public consumption, and then into beverage-speci® c effects (private or public). In this
concluding section ® rst the statistical and then
the substantive plausibility of the results will be
discussed.
The outcome was fairly clear-cut: the variation
in the assault rate seem s to be related to changes
in public consumption, whereas the variation in
the homicide rate appears to be linked to
changes in private consumption. The further
decomposition suggested that beer and spirits
(consumed on premise) were the more important
beverages with respect to assaults. The corresponding ® nding with respect to homicide was
that it is (private) spirits that matters primarily.
As already pointed out, beer other than strong
beer is not included in the decomposed beverage
indicators. The most likely outcome of this
seems to be that if the weaker beer actually has
any in¯ uence on violence, it is absorbed by estimated effect of the strong beer indicator that was
used in the analyses.
One way of checking the consistency of the
results is to inspect how successful the decomposition of the estimated total alcohol effect is. It
can be seen that in the case of assaults all the
total alcohol effect is attributed, ® rst to public
consumption, and next to public consumption of
beer and spirits. (The discrepancy between the
AF s reported in Tables 2, 3 and 5 is well within
the margins of error.) With respect to homicide
the ® rst decom position is successfulÐ all the
alcohol effect is attributed to private consumptionÐ but in the further beverage-speci® c
decom position there is some leakage since the
only effect that is signi® cant (spirits) does not
wholly absorb the previously estim ated effects.
Statistical associations of the kind that have
been presented here are of course no proof of
causality. However, the ® ndings are based on a
method that minim izes the risk of spurious relationships, they are internally consistent and
accord with results from similar studies for other
countries and tim e periods (Skog & Bjù rk, 1988;
Lenke, 1990). Further, the ® ndings are com patible with results that have been obtained through
quite other methods, notably laboratory experiments and descriptive data on offenders. Therefore, if the ® ndings seem fairly robust and
reliable when they are scrutinized on statistical
and methodological grounds, how do they fare
from a more substantive and theoretical perspective? That the number of assaults would be
associated prim arily with public consumption
and the number of homicides to private intake
was anticipated on the grounds stated in the
introduction: assault is an offence that typically
takes place in a public setting, and hom icide in a
private context. It may be objected that this is
merely to beg the question, and provides little
understanding of what mechanisms are at work.
The keys to understanding what is occurring can
probably be found by considering the following
factors: opportunity structure, social control and
context of drinking, drinking patterns associated
with the different beverage types, and character-
Crim inal violence and different beverage types
Public consumption of beer
plus spirits (litres 100%)
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
1956 1961 1966 1971 1976 1981 1986 1991
Year
Private consumption of spirits (litres 100%)
Figure 6. Public consumption of beer plus spirits (litres
100%) per capita 15 years and above, Sweden 1956± 94.
4.5
4.0
3.5
3.0
2.5
2.0
1.5
1956 1961 1966 1971 1976 1981 1986 1991
Year
Figure 7. Private consumption of spirits (litres 100 %) per
capita 15 years and above, Sweden 1956± 94.
istics of the drinkers. However, given the lack of
detailed docum entation of these aspects the following interpretations are somewhat speculative.
It is clear that public places such as restaurants
and bars are characterized by a concentration of
social interaction which, aside from the positive
aspects, inevitably contain elem ents of friction
and frustration, e.g. competition about seats and
partners, incorrect bills, drinks being refused, etc.
Following laboratory ® ndings (e.g. Gustafson,
1986) there is an elevated risk that alcohol triggers overt aggression when frustration is present.
At the sam e time the public context provides a
high degree of social control which should prevent the aggression, once it occurs, from taking
697
such a serious form that the outcom e becomes
fatal. The beverage-speci® c effect of (public) beer
and spirits consumption that was suggested m ay
be interpreted as resulting from the circumstance
that beer is the preferred beverage of young
people (Alkohol- och narkotikautvecklingen i
Sverige , 1996), an age group that is supposedly
more violence-prone than average, and that spirits are typical intoxication beverages.
The suggested link between private spirits consumption and hom icide m ay also be viewed in
terms of characteristics of context and drinkers.
Alcohol-related homicides occur typically at private parties among heavy drinkers who often have
a criminal record; in the news items such occasions are commonly labelled `spirits parties’ . The
context is characterized by a fairly low degree of
social control, enhancing the risk that the outcome of possible violence becomes more severe.
A study by Gruenewald & Ponicki (1995) is of
interest in this context. They expect cirrhosis
mortality to be more closely associated with
spirits use. The reason for this would be that the
people in the primary risk group for cirrhosis,
alcohol-dependent and alcoholic drinkers, tend
to prefer spirits to beer and wine (Selzer, Vinokur
& W ilson, 1977). Spirits sales are thus regarded
as a m arker for the size of the risk group. Pooled
cross-section tim e series analyses of US data lent
support to this expectation; there was a
signi® cant effect on cirrhosis mortality of spirits
sales, but not of beer or wine sales. This line of
reasoning also seems feasible in the interpretation
of the suggested association between homicide
and spirits sales, since chronic alcohol dependence and alcohol abuse is particularly prevalent
in homicide offenders (for a review see Lenke,
1990).
The fact that we did not ® nd any signi® cant
association between wine consumption and violence m ay also tell us som ething about the social
context and drinkers associated with this beverage type. Wine is thus the preferred alcohol
beverage among wom en ( Alkohol- och narkotikautvecklingen i Sverige , 1996), and in comparison with other alcohol beverages it is m ore
likely to be consumed in connection with meals.
Another issue is to what degree the trends in
assaults and hom icide can be accounted for by
the trends in beverage-speci® c consumption
(public beer and spirits consum ption, and private
spirits consum ption respectively). By com paring
Figs 1 and 6, it is seen that assaults and public
698
Thor NorstroÈ m
intake of beer and spirits coincide fairly well
during the latter part of the period, but diverge
during the early part. Sim ilarly, there is a reasonable m atch between hom icide and private spirits
consumption until the m id 1970s; after that point
there is a m arked and steady decrease in consum ption, and a slight increase in the homicide
rate (Figs 2 and 7). The diverging trends that are
observed here do not imply that the underlying
correlation between alcohol and violence would
be negative during these sub-periods, only that
there are trends in other causes of violence that
mask the alcohol effect. For instance, some other
factor behind homicide must have grown in
importance after the mid-1970s, and this has
more than offset the alleged in¯ uence of the
decreasing intake of spirits. It is not exceptional
that the correlation between the trends of two
series does not re¯ ect the underlying relationship.
For instance, Norwegian data for the period
1865± 1939 show a m arked downward trend in
per capita consumption of alcohol, and a steady
increase in real incom e; that is, there is strong
negative correlation between the two series. However, when these data are analysed with the same
technique that we have applied here (ARIMAmodelling of differenced data), the expected positive income elasticity emerges (Skog, 1986a). A
plausible explanation of this seeming contradiction is that the positive relationship between
incom e and alcohol consumption is concealed in
the raw data by ideological in¯ uences of the
temperance movement. The latter exerted a lowering effect on consumption, and thereby
camou¯ aged the effect of increasing real income
(see NorstroÈ m, 1989b and Skog, 1986b, 1988 for
further exam ples of mis-match between trends
and underlying relationships).
Finally we should remind ourselves of the great
cultural variation in the association between
alcohol and violence, which was pointed out in
the introduction. This means that the ® ndings
reported above are speci® c to Sweden for a
certain period of time. In m any other countries,
e.g. on-premise drinking is certainly different with
regard to social context and com position of
patrons. This and other differences in alcohol
culture are likely to result in other patterns of
interrelationships than have been found in this
study.
R eferen ces
Alkohol- och narkotikautvecklingen i Sverige, Report
no. 96 (1996 ) [Trends in alcohol- and drug abuse in
Sweden] (Stockholm , FolkhaÈ lsoinstitutet and CentralfoÈ rbundet foÈ r alkohol- och narkotikaupplysning).
Alkoholstatistik [Alcohol Statistics] (Stockholm , Statistics Sw eden).
B OYATZI S , R. E. (1974 ) The effect of alcohol consum ption on the aggressive behavior of m en, Quarterly
Journal of Studies on Alcohol, 35, 959± 972 .
B OX , G. E. P. & J ENKINS , G. M. (1976 ) Time Series
Analysis: forecasting and control (London, H oldenDay).
Brottsutvecklingen 1987, Report no. 1987:5 (1987 )
[Trends in Crim e 1987] (Stockholm, National
Council of Crim e Prevention).
Brottsutvecklingen 1992 och 1993, Report no. 1994:3
(1994 ) [Trends in Crim e 199 2 and 1993] (Stockholm, National Council of Crim e Prevention).
B USHM AN , B. J. & C OOPER , H . M. (1990 ) Effect of
alcohol on human aggression: an integrated research
review, Psychological Bulletin, 107, 341± 354.
DoÈ dsorsaker [Causes of Death] (Stockholm , Statistics
Sweden).
G RAHAM , K., S CH M IDT , G . & G ILLIS , K. (1996 ) Circumstances when drinking leads to aggression: an
overview of research ® ndings, Contemporary Drug
Problems, 23, 493 ± 557.
G RU ENEWAL D , P. J. & P ONICKI , W . R. (1995 ) The
relationship of alcohol sales to cirrhosis m ortality,
Journal of Studies on Alcohol, 56, 635± 641 .
G U STAFSON , R. (1985a ) Alcohol and aggression: a validation study of the Taylor aggession paradigm , Psychological Reports, 57, 667± 676.
G U STAFSON , R. (1985b ) Alcohol-related aggression: a
further study of the importance of frustration, Psychological Reports, 57, 683± 697.
G U STAFSON , R. (1986) Alcohol and Human Physical
Aggression: the mediating role of frustration, Dissertation (Uppsala, U niversity of U ppsala).
Kriminalstatistik 1965 [Crim inal Statistics 1965]
(Stockholm, Statistics Sw eden).
Kriminalstatistik 1968 och 196 9 [Crim inal Statistics
1968 and 1969] (Stockholm, Statistics Sweden).
K UÈ HLHO RN , E., A NDERSS O N , J., G U STAVSS ON , J.,
K NUTSSO N ., O LSSON ., R. & R EINIU SSON , M . (1984)
Den svenska vaÊ ldsbrottligheten, Report no. 1984:1
[Criminal Violence in Sweden] (Stockholm ,
National Council of Crim e Prevention).
L ENKE , L. (1990) Alcohol and Criminal ViolenceÐ time
series analyses in a comparative perspectiv e (Stockholm ,
Alm qvist & W iksell).
L IEBERSON , S. (1985) Making It Count. The improvement of social research and theory (C alifornia, University of California Press).
L ILIENF ELD , A. M . & L ILIENFELD , D. E. (1981 ) Foundations of Epidemiology (O xford, Oxford University
Press).
M AC A NDREW , C. & E DG ERTON , R. B. (1969 ) Drunken
Comportment (C hicago, Aldine).
M AYFIELD , D. (1976 ) Alcoholism , alcohol, intoxication
and assaulative behavior, Diseases of the Nervous
System , 37, 288± 291 .
N ORSTROÈ M , T. (1988a ) Deriving relative risks from
aggregate data. 1. Theory, Journal of Epidemiology
and Community Health, 42, 333± 335 .
Crim inal violence and different beverage types
N ORSTROÈ M , T . (1988b ) Real w ages, alcohol consum ption and m ortality in Sweden, 1861± 1913, European
Journal of Population, 4, 183± 196.
N ORSTROÈ M , T . (1989 ) T he use of aggregate data in
alcohol epidem iology, British Journal of Addiction,
84, 969 ± 977.
R OSLU ND , B. & L ARSON , C. A. (1979 ) Crim es of violence and alcohol abuse in Sweden, International
Journal of the Addictions, 14, 1103 ± 115.
S ELZER , M. L., V INOKU R , A. & W ILSO N , T. D. (1977 ) A
psychosocial com parison of drunken drivers and alcoholics, Journal of Studies on Alcohol, 38, 1294±
1312.
S KOG , O. - J. (1986a ) An analysis of divergent trends in
alcohol consuption and econom ic developm ent,
Journal of Studies on A lcohol, 47, 19± 25.
S KOG , O. - J. (1986b ) Trends in alcohol consumption
and violent deaths, British Journal of Addiction, 81,
365± 379.
S KOG , O. - J. (1988 ) Testing causal hypotheses about
699
correlated trends: pitfalls and remedies, Contem porary Drug Problems, 15, 565± 607 .
S KO G , O. - J. & B Jù RK , E. (1988 ) Alkohol og voldskriminalitet: en analyse av utviklingen i Norge 1931±
1982 [Alcohol and crimes of violence: an analysis of
the developm ent in Norway 1931 ± 1982], Nordisk
Tidskrift for Kriminalvidenskab, 75, 1± 23.
S M ART , R. G . (1996 ) Behavioral and social consequences related to the consum ption of different beverage types, Journal of Studies on Alcohol, 57, 77± 84.
W HITE , H. R. B RIC K , J. & H ANSELL , S. (1993 ) A longitudinal investigation of alcohol-use and aggression in
adolescence, Journal of Studies in A lcohol, suppl.,
Report no. 15, Dissertation 11, 62± 77.
W IKSTROÈ M , P. O. H. (1985 ) Everyday Violence in Contemporary Sweden, Report no. 15, Dissertation
(Stockholm, National Council of Crim e Prevention).
W IKSTROÈ M , P. O. H . (1992 ) Context-speci® c trends in
homicide in Stockholm, Studies on Crime & Prevention, 1, 88± 105.