OUP user menu

Is Youth Violence Temporally Related to Alcohol? A Time-Series Analysis of Binge Drinking, Youth Violence and Total Alcohol Consumption in Sweden

Johan Svensson, Jonas Landberg
DOI: http://dx.doi.org/10.1093/alcalc/agt035 598-604 First published online: 25 April 2013

Abstract

Aims: The purpose of this study is to investigate the temporal association between violence and binge drinking among Swedish youth. Two time periods are analysed, the first one representing the full observation period 1971–2009 and the second one representing a confined period 1971–2000. Furthermore, the association between population drinking and binge drinking among youths is also investigated with regards to the two time periods. Methods: ARIMA modelling was applied in order to estimate these associations. Results: Based on the confined time period (1971–2000), analyses revealed that four out of six estimates (two where borderline) of the association between binge drinking and violence (self-reported as well as convictions for assaults) were positive and statistically significant. However, most estimates became non-significant when the full study period (1971–2009) was analysed. The analyses of the relationship between total consumption and youth binge drinking revealed that binge drinking among military conscripts and boys was affected by changes in total consumption during the shorter confined study period. These associations became non-significant when the full study period was included in the models. Conclusions: (a) there was a positive relationship between violence and binge drinking among Swedish youth at the aggregated level, but mainly with regards to the shorter study period, (b) changes in per capita alcohol consumption were associated with binge drinking among young men and this was more evident for the shorter study period and (c) there was empirical evidence for the idea that these associations became weaker or non-existent after the year 2000.

INTRODUCTION

Young people are at more risk than the general population of becoming victims of violence. In Sweden, 9% of men aged 16–24 years reported having been victims of assaults during the previous year, whereas the corresponding estimate in the age group 25–44 years is 3% (Estrada 2008). A similar age pattern has been found among women (6% in the younger age group and 2% in the age group above). These self-reported estimates are also confirmed in hospitalizations due to violence, which are the highest among men in their twenties (Danielsson et al., 2009).

The evidence that alcohol consumption is associated with violent behaviour is pervasive (Norstrom, 1998; Room and Rossow, 2001; Haggård-Grann et al., 2006). Individual-level studies have shown that the volume of consumption and in particular drinking to intoxication (henceforth referred to as binge drinking), is associated with an increased risk of alcohol-related aggression among adolescents (Bye and Rossow, 2010). Furthermore, numerous population-level studies using time-series of aggregated data have shown that changes in population drinking are related to changes in rates of various forms of violence, like for instance, homicide, suicide and assaults (Rossow, 1996; Norström and Ramstedt, 2005).

However, most population-level studies have used per capita consumption as an input, and have therefore not been able to directly investigate the temporal association between alcohol and violence in different population subgroups or to test the significance of changes in the prevalence of intoxication. In this study we aimed to investigate the population-level association between drinking and violence among young people in Sweden. More specifically, we have conducted a time-series analysis of data for the time period 1971–2009 to examine the extent to which changes in the prevalence of binge drinking among Swedish youth are related to changes in self-reported victimization of violence and assault convictions.

Although previous research suggests an association between alcohol and violence, the underlying causal mechanisms are complex. The fact that only a small number of all drinking occasions lead to violence suggests that the association is conditioned by several other factors. Individual characteristics such as having an impulsive personality or being inclined to suppress anger have been suggested as significant conditional factors in this context (White et al., 1993). However, drinking to intoxication seems to be one plausible mechanism underlying the association of alcohol and violence. Norström (2011) suggests that there are three elements that are crucial for understanding the process. First, drinking is predominantly a social activity, and a person who has been drinking to intoxication is likely to be accompanied by one (or several) other person(s) who has done this as well. Second, the state of intoxication tends to make individuals more ‘short sighted’, meaning that they postpone future costs and act more upon more immediate cues. And third, during the drinking occasion, it is likely that one person will provoke another simply because intoxicated people more often engage in acts that are hostile and offensive. Norström argues that these mechanisms are consistent with the fact that a large proportion of both perpetrators and victims have been intoxicated prior to the event, and that the perpetrator and the victim often are acquainted and have been drinking together.

Given that intoxication occasions are associated with an increased risk of interpersonal violence, we would expect that an increase in the prevalence of binge drinking among young people would be associated with an increase in the number of drinking occasions entailing an elevated risk for violence and, other things being equal, that there is a positive relationship between the prevalence of intoxication and rates of violence among youth.

In a broader context, it also seems plausible that the level of drinking in the general population may influence drinking behaviour among young people and, subsequently, the level of alcohol-related violence among youth. This line of reasoning stems from Skog's (1985) theory of ‘the collectivity of drinking cultures’. According to Skog, there is a strong collective component in human drinking behaviour, which implies that changes in the level of consumption in a society tend to reflect a collective shift in the overall population. The hypothesis that there is an association between per capita consumption and youth drinking has recently been supported in a cross-country comparison by Fuhr and Gmel (2011), who found that countries that rank high on per capita consumption also are likely to score high on prevalence of adolescent drinking. Based on this reasoning, a second aim of the present study was to examine whether there is a relationship between population drinking and the prevalence of intoxication drinking among youth.

However, since 2000, there are indications that the Swedish development in youth drinking has deviated from what would be expected based on Skog's collectivity theory. The two main indications of this are that (a) alcohol consumption among Swedish youths (15-year-olds) and binge drinking among military conscripts declined between 2000 and 2005, despite sharply rising population drinking (Leifman and Gustafsson, 2003; Guttormsson, 2007) and (b) the decreasing trend in youth drinking has been paralleled by an increase in alcohol-related hospitalizations among youth (Socialstyrelsen, 2010; Hallgren et al., 2012). In view of these recent Swedish experiences, a third aim of our study was to assess whether the strength of the studied relationships has become weaker over time. This will be tested by estimating the relationships for a shorter period confined to 1971–2000 and a longer period including all study years 1971–2009.

To conclude, on the basis of Swedish data for 1971–2009, the following three research questions will be addressed in the prevailing study:

  1. Is there a temporal relationship between the prevalence of youth binge drinking and youth violence?

  2. Is there a temporal relationship between population drinking and youth binge drinking?

  3. Have the strength of these relationships become weaker since 2000?

DATA AND METHODS

Estimates of the total consumption were based on sales statistics and are expressed in litres of 100% alcohol per year and inhabitants 15 years and older (CAN, 2012). Estimates of binge drinking came from two different sources; the annual school surveys commissioned by the Swedish Council for Information on Alcohol and Other Drugs (CAN), for pupils age 15, and a survey conducted among conscripts into compulsory military training in Sweden. The school survey has been on-going since 1971 and this is also the major reason for the observation period to start at this year. Furthermore, the survey has been completed during March and April every year. The number of respondents has varied; however, for most years the number of respondents has been around 5000. Furthermore, the response rate has been fairly consistently around 85% throughout the period 1971–2009 (Henriksson and Leifman, 2011). The survey to the military conscripts was voluntary and performed during the theoretical testing of the conscripts. The conscripts were men, mainly at the age of 18 years. It can be noted that Sweden had a mandatory military system for all male Swedish citizens at the age of 18, until 2010. Changes in the routines of conscriptions at the end of 2006 led to abolishment of the alcohol and drug survey in this group. The number of respondents has varied from the lowest, about 28 000 (in 1992), to the highest 59 000 in (1972/73) (Guttormsson, 2007).

The binge measurements differed between the conscripts and the pupils. Among the pupils, the question asked concerned a specific volume of alcohol (equivalent to a bottle of wine), whereas the question directed to the conscripts referred to the subjective feeling of being intoxicated. However, both binge questions were aggregated into yearly estimates of the prevalence proportion of binge drinking (at least once a month) among the studied youth groups.

The measurements of violence are two different indicators and came from two different sources, criminal statistics from the Swedish National Council for Crime Prevention (BRÅ) and self-reported experience of violence from the Living Conditions Surveys (ULF) collected by Statistics Sweden (SCB). The ULF survey is an annual survey conducted since 1975. The sample size has been between ∼5000 and 14 000, with a response rate varying from 75 to 86%. To measure experiences of violence, we used a question that asked whether the respondent had been exposed to violence leading to visible marks on the body or bodily harm during the past year. This question was aggregated into yearly estimates of the proportion of respondents experiencing violence among the age group 16–20 years. The other indicator of violence is rates of (per 100 000) convictions for assaults in the age group 15–20 years.

The series on consumption—total alcohol consumption and binge drinking among boys and girls aged 15 years and among conscripts—starts in 1971 (1972 among the conscripts) and ends in 2009 for the pupils, and in 2006 for the conscripts. The series on self-reported violence starts in 1980 and continues for the rest of the period, and finally, the series on convictions for assaults stretches from 1971 to 2009.

This study was approved by the regional ethics committee in the Stockholm region (Protocol: 2012/5:3).

The analyses begin with a descriptive analysis of the data and its trends over time. Next, the data were analysed by autoregressive integrated moving average (ARIMA) time-series models (Box et al., 1976). A first step in this approach is to make the series stationary by differencing. This means that we were analysing the relationship between the yearly changes in the data, rather than analysing the relationship between the raw time-series. This process strongly reduces the risk of obtaining a spurious relationship, as an omitted variable is more likely to be correlated to common trends than due to synchronization in the yearly changes. Furthermore, the noise term, consisting of measurement errors and causal factors not included in the model, is incorporated into the model. The noise term is estimated in terms of autoregressive or moving average parameters, which are of two kinds, regular AR (n) and MA (n), respectively, where n denotes the order of the parameter (for further methodological assumptions regarding this type of analysis, see Norström and Skog (2001)).

A log-log model of the following specification was used:Embedded Image

The operator ∇ denotes that the series are differenced. D is the dependent variable, in this case either binge drinking or violence (self-reported or convictions for assaults), and IV is the independent variable, total alcohol consumption or binge drinking. b is the effect parameter that is to be estimated. N (noise) is the noise term that includes other factors. An important model fit criterion is that the residuals are white noise, which means no temporal structure in the residuals. The models here have met this criterion according to the Box–Ljung test.

RESULTS

Descriptive analysis

Table 1 presents descriptive statistics (period averages) of the prevalence of binge drinking, self-reported violence and rates of convictions for assaults. The average prevalence of binge drinking was higher among conscripts compared with pupils aged 15 years. Almost half of the conscripts had been binge drinking during the last month compared with about one-fourth of the boys and roughly one-fifth of the girls. Furthermore, boys were more often victims of violence compared with girls, according to the self-reported estimates.

View this table:
Table 1.

Descriptive presentation of variables included in the modelling

VariableObservation periodAverage (over observation period)
Binge %Violence
Conscripts1972–200644.6
Pupils aged 15 years
 Girls1971–200922.1
 Boys1971–200927.5
Self-reported violence (%)1980–2009
 Girls11.2
 Boys15.4
Convicted for assaults (per 100 000)1971–2009297.4
  • Observation periods, approximate mean binge, convictions for assaults and self-reported violence.

The trends in convictions for assaults and youth binge drinking are presented in Fig. 1. In general terms it appears as if they follow the same pattern up until the turn of the millennium. However, thereafter binge drinking among conscripts and boys decreased, whereas binge drinking among girls was rather stable. The rates of convictions for assaults decreased during the first part of the 2000s, and then increased during the latter part of the 2000s. Taken together, it seems as if the trend in convictions for assaults roughly followed the trends in binge drinking for all three youth groups during the first part of the study period, that is, up to around the first years of the 2000s, after which the trends in conviction and binge drinking among youths diverged.

Fig. 1.

Prevalence (%) of binge drinking among boys and girls (aged 15), conscripts (aged 18) and rates of convictions for assaults, in the age group 15–20 years, during the period 1971–2009.

With regards to the full observation period there is no evident trend in self-reported violence, among either boys or girls (Fig. 2). Beginning in the 1980s, when data for self-reported violence were first available, the trends were decreasing until the mid-to-late 1980s. During this period the trend for binge drinking among girls also decreased. This period was followed by an increase in self-reported violence as well as binge drinking among all youth groups up until the mid-1990s. From the mid-1990s to the turn of the millennium self-reported violence decreased, while binge drinking was fairly stable. During the last years of the study, binge drinking among conscripts and boys declined, whereas the trends of self-reported violence were fairly stable.

Fig. 2.

Prevalence (%) of binge drinking among boys and girls (aged 15), conscripts (aged 18) and violence (self-reported victims of violence (1980 onward), during the period 1971–2009.

Figure 3 displays the trends in total consumption and prevalence of binge drinking among the three youth groups. After a peak in the mid-to-late 1970s the trends in both total consumption and binge drinking among all youth groups decreased until the early-to-mid-1980s. This was followed by an increase of all trends for binge drinking up till the late 1990s, during the same period the total consumption decreased. After the turn of the millennium the trend in total consumption increased sharply, whereas binge drinking among all three youth groups in general decreased during the remainder of the study period.

Fig. 3.

Total alcohol consumption in litres (100%) per inhabitant aged 15+, prevalence (%) of binge drinking among conscripts, girls and boys (aged 15), during the period 1971–2009.

Based on visual inspections of raw data, it is anticipated that there is an association between the prevalence of binge drinking among youths and both convictions for assaults and self-reported violence, at least for the shorter study period. Likewise, visual inspections reveal that trends in total alcohol consumption and binge drinking among youth were similar during the shorter study period but that the trends started to diverge from around the year 2000. To fully examine these associations and to reduce the risk of obtaining spurious relationships due to common trends, a more thorough analysis using ARIMA modelling is required.

Estimation of time-series models

The results from the ARIMA modelling on the relationship between prevalence of binge drinking among young people and violence, for both the short and the full study periods are presented in Table 2. In model A the dependent variable is convictions for assaults (per 100.000), and in model B the dependent variable is self-reported victims of assaults (%).

View this table:
Table 2.

Estimated effects (log-log models) of prevalence of monthly binge drinking among youth (conscripts and pupils aged 15 years) on A, convictions for assaults and B, self-reported assaults

InputModel A convictions for assaultsModel B self-reported violence
EstimateSEModelaQbPcEstimateSEModelaQbPc
Period 1972–2009Period 1980–2009
Conscripts1.496*0.6800.1.02.5860.7631.039(*)0.5750.1.16.2660.281
Pupils aged 15 years
 Girls0.400ns0.2310.1.21.3090.93400.400*0.0181.1.03.8430.572
 Boys0.592(*)0.3110.1.21.8310.872−0.099ns0.7540.1.02.6820.749
Period 1972–2000Period 1980–2000
Conscripts1.488*0.4010.1.22.8440.7241.768*0.7170.1.13.9270.560
Pupils aged 15 years
 Girls0.424(*)0.2340.1.12.4520.7840.962*0.2430.1.12.0800.838
 Boys0.63*0.190.1.23.1660.6741.245(*)0.6210.1.14.1730.525
  • Estimated on data for the periods 1972–2000 and 1972–2009.

  • aARIMA models are indicated by, in turns of order, order of autoregressive parameters (AR), order of differentiating and order of moving average parameters (MA).

  • bBox–Ljung test for residual autocorrelation at lag 5.

  • cP-values for the Box–Ljung test.

  • (*)P < 0.1; *P < 0.05; **P < 0.01.

For the shorter time period (1972–2000) all estimates of the association between the prevalence of binge drinking and convictions for assaults were positive. Moreover, for conscripts and boys the estimates were statistically significant (P < 0.05), whereas for girls the estimate was only close to significant (P < 0.1). The estimates imply that a 10% increase in the prevalence of binge drinking among conscripts was associated with an ∼15% increase in the convictions for assaults, whereas the effects on convictions associated with a 10% increase in the prevalence of binge drinking among girls and boys were 4 and 6%, respectively. The estimates of the association between binge drinking and being a victim of violence (self-reported) (model B) were all positive and significant for the shorter time period. Here, the estimates imply that a 10% increase in the prevalence of binge drinking among conscripts was associated with an 18% increase in the self-reported violence among young men aged 15–20, the corresponding effects of a 10% change in the prevalence of binge drinking among girls and boys were 10 and 12%, respectively.

When the full study period (1972–2009) was included in the models, the estimates of the relationship between prevalence of binge drinking and convictions for assaults were positive for all studied youth groups and fairly similar to those obtained for the shorter time period. However, the only estimate that was statistically significant at the 5% level was found for conscripts. A similar pattern was obtained in the models for self-reported violence. Here, the estimates for conscripts and boys failed to reach significance, whereas the estimate for girls remained significant and suggests that a 10% increase in the prevalence of binge drinking among girls was associated with a 4% increase in the self-reported violence for the same group.

In Table 3, the association between total alcohol consumption and prevalence of binge drinking for each of the three youth groups is presented. In the models including the short study period, the association between total alcohol consumption and binge drinking was positive and statistically significant for both conscripts and boys, whereas no significant association was found for girls. The estimates in these models were interpreted as above, that is, as indicating that a 10% increase in the total alcohol consumption was associated with a 3 and 2% increase in the prevalence of binge drinking among conscripts and boys, respectively. In the models including the full study period, none of the estimates were significant at the 5% level.

View this table:
Table 3.

Estimated effects (log–log models) of total alcohol consumption on monthly binge drinking among youth (conscripts and pupils aged 15 years)

InputEstimateSEModelaQbPc
Period 1972–2009
Conscripts0.308(*)0.1680.1.01.000.962
Pupils aged 15 years
 Girls0.002ns0.0020.1.01.3930.925
 Boys0.105ns0.07520.1.01.3840.926
Period 1972–2000
Conscripts0.336*0.1690.1.11.4900.914
Pupils aged 15 years
 Girls0.056ns0.5100.1.01.0640.957
 Boys0.182*0.0890.1.04.0360.544
  • Estimated on data for the periods 1972–2009 and 1972–2000.

  • aARIMA models are indicated by, in turns of order, order of autoregressive parameters (AR), order of differentiating and order of moving average parameters (MA).

  • bBox–Ljung test for residual autocorrelation at lag 5.

  • cP-values for the Box–Ljung test.

  • (*)P < 0.1; *P < 0.05; **P < 0.01.

DISCUSSION

Given our initial research questions, the following three conclusions can be made from the analyses: (a) there was a positive relationship between binge drinking and violence among Swedish youth at the aggregated level, but mainly for the earlier part of our study period, (b) changes in per capita alcohol consumption were associated with binge drinking among young people, but this only applied to young men, and more clearly for the earlier study period and (c) there is empirical evidence for the idea that these associations became weaker or non-existent after the year 2000.

Our findings are thus not clear-cut, and raise some new questions. The finding that there is a positive relationship between binge drinking and violence for the earlier part of the study period is as expected based on Skog's ‘collectivity of drinking’ theory (Skog, 1985). However, it raises the question as to why the association was not found over the full study period. One hypothesis is that other factors besides alcohol may have affected the association with youth violence on the latter part of the study period. One indication of this is presented in a study by Granath (2011), where all violent crimes with a deadly outcome in Sweden during the periods 1990–1996 and 2002–2008 were studied. Here, Granath found that alcohol involvement in relation to violent crimes had been reduced. Furthermore, it should be noted that the rate of abstention has increased considerably among youths in Sweden since around the year 2000: from 18–19%, among both boys and girls, to 39% among boys and 32% among girls in 2010 (Henriksson and Leifman, 2011). This implies that the prevalence of binge drinking partly has decreased since a lesser number of youths drink, whereas binge drinking among those who still drink has decreased to a lesser amount. The questions concerning how, frequency of binge drinking has evolved, particularly among different consumer groups, and the underlying mechanisms behind the increasing rates of abstainers, are both important for further research in order to more fully understand recent changes in drinking among youths in Sweden. Furthermore, a change in convictions for assaults does not necessarily indicate a change in actual violent events. It may rather reflect changes in policies or an increase in a proneness to report assaults (BRÅ, 2011). Still, self-reported estimates of violence revealed results similar to those of convictions for assaults.

The findings concerning the association between changes in per capita alcohol consumption and binge drinking among young men should also be discussed. Again, the associations are more pronounced for the earlier part of the period than for the whole period. In these analyses we used sales statistics as an indicator of the total alcohol consumption. However, since Sweden joined the EU in 1995, the share of unrecorded alcohol consumption in relation to the total alcohol consumption increased considerably, from about 25% in 1996 to about 38% in 2004 (Ramstedt et al., 2009), and this was mainly due to variations in import by travellers. This may have resulted in poorer validity of the applied alcohol measure, which in turn may have affected the estimates to become weaker when using the latter part of the study period. Furthermore, in 2001 Sweden implemented a national action plan to prevent alcohol-related harm (Prop., 2000/01:20), by counteracting alcohol policy implications of the Swedish entry into the EU. Following this, many local preventive programmes targeting youth were implemented and disseminated, partly funded by the government. Thus, if the action plan had any effect on alcohol consumption and harm, it is likely that it affected drinking among youth more than drinking in the general population.

The main finding from this study must be assessed in the context of its limitations. We have mentioned the weakness of the binge drinking measurement and the possibility that the measurement does not capture changes across those groups who binge drink over time. Furthermore, the binge measurement among the conscripts refers to the subject feeling of being intoxicated, whereas the binge measurement among the pupils refers to a given amount of alcohol. It is worth noting that it has been shown that there are variations over time in the number of drinks required to feel intoxicated (Kerr et al., 2006). As a consequence the amount drank to feel intoxicated may vary from different periods. Hence, it would have been preferable to have used the same binge measurement among all youths. The violence measurements and the use of sales statistics also run the risk of being biased from contextual meanings from different time periods.

Moreover, there is the risk of omitted variable bias. The use of ARIMA modeling entails a low risk of obtaining biased estimates due to omitted variables. In order to have an effect on the models estimates omitted variables must be correlated with yearly changes in the input as well as the output variables, that is, synchronized in time. Although ARIMA entails a low risk of obtaining biased estimates, we should acknowledge that the number of data points is on the verge of fulfilling the criteria needed for ARIMA modelling.

In this study we have analysed data spanning 35 years, and the overall findings are compatible with Skog's theory of collectivity of drinking, since we found an association between the consumption among the general population and binge drinking among youth (conscripts and boys) for the shorter time period. From a policy point of view our findings also lend support to the idea that policy measures that limit the total consumption, such as high taxation and low availability of alcohol, which has been important in Sweden during the study period, has had an effect on violence among youth. However, the finding that a recent trend of a decline in binge drinking among youth has not been accompanied by a similar decline in violence calls for further research on the complex associations between alcohol and violence among youth.

Funding

This research was funded by the Swedish Council for Working Life and Social Research (FAS) grant 2009–1784. Conflict of interest statement. None declared

Acknowledgements

The authors thank Mats Ramstedt for comments on earlier versions of this article.

REFERENCES

View Abstract