Journal of Substance Use 6:218-228, 2001




What do we need to know and what research is needed?


Robin Room and

Ingeborg Rossow, PhD





The relation of alcohol to violence has long been a subject of controversy. There is no argument that a large proportion of the perpetrators of crimes had been drinking at the time of the crime was committed. For example about 40 percent of violent offenders in state and local jails in the United States had been drinking at the time of the offence for which they were jailed. The causal role of alcohol in the commission of violent crimes, however, has been a contentious issue.

One source of controversy about the causal nature of alcohol in violent crime has been the definition of “cause”, with some researchers requiring that in order to be considered causal, the presence of alcohol must be necessary and sufficient in order for a violent crime to occur. Rather than adopting such a stringent position, it seems much more reasonable to consider the epidemiology of violence in the same way one would consider the epidemiology of disease. For example, while it is clear that smoking is an important, if not the most important, causal factor in the occurrence of lung cancer, not all smokers get cancer and not all cancer victims are smokers. In the same way, not all violent crimes involve alcohol and not all people under the influence of alcohol commit violent crimes. However, ample evidence exists that more drinking tends to result in more violence and less drinking in less violence.


Evidence of the link between alcohol and violence

The role of alcohol in violence is complex and relates not only to the offender’s drinking but also to that of the victim as well as others in the social context or the gathering of people in drinking places. When assessing the share of violence that can be attributed to alcohol, this complexity is probably better reflected when the relationship between alcohol and violence is estimated at the aggregate level, preferably time series analyses. In the end, the policy question behind the attributable fraction concept is: if a certain fraction of consumption, or of high-risk drinking occasions, could be taken away, how much would the rates of the adverse events fall? from this perspective, the most direct measurement of attributable fraction is from studies at the population level, rather than studies of individual events. over the past decade several studies on changes at the population level have demonstrated that an increase in the population’s alcohol consumption is followed by an increase in rates of violence in that population.

Estimates of the degree to which alcohol contributes to violent crime vary from study to study and country to country. For example, studies of assaults in the United States estimates that from 20 to 46 percent are attributable to alcohol, depending on the assumptions made nad the methods used. While the estimates may vary, there is no doubt that alcohol consumption substantially impacts rates of violence and that violence can be prevented by significant reductions in alcohol use and/or heavy drinking episodes. Several studies support this assertion. Studies in Sweden and Russia, for example, showed dramatic decreases in the occurrence of violence following decreased alcohol consumption.


A cross-cultural perspective

Until recently, the literature on the relationship between alcohol use and violence has assumed a single relative risk for a given volume of drinking. This model is based on the assumption that the relationship between alcohol and violence is similar to the relationship between alcohol and physical disease, such as cirrhosis of the liver. Recently, however, it has become apparent that social problems from drinking are more closely related to pattern of drinking than to the overall volume. That is, more violence is likely to occur when an individual drinks a large quantity of alcohol at one time.

The patterns of drinking associated with different cultures are likely to affect the relationship between overall drinking levels and violence. It is also clear from the research that people in different cultures behave very differently when under the influence of an equal amount of alcohol. Thus the relative risk of violence for a given pattern of drinking can vary from culture to culture.

Studies of different countries confirm these different patterns in the influence of drinking on violence. For example, one study found that the impact of alcohol on homicide was highest in Finland, Norway and Sweden, and lowest in the Mediterranean countries (France, Italy, Portual and Spain). it appears that the stronger impact of alcohol on violent crimes is found in drinking cultures where people are more likely to drink to the point of acute intoxication than where people may drink more often but less at the time.


Challenges for the future

Further study of the potential impact of reductions in alcohol use is needed, yet difficulties in studying this issue exist. Current methods of estimating the share of violence attributable to alcohol use based on individual level data have two flaws. First, while it is possible to determine the presence of alcohol in a violent event and to determine whether participants believe that alcohol played a role, it is not yet possible to determine objectively whether the event would have occurred in the absence of alcohol. A second flaw is that we have not yet found means to measure all the ways in which alcohol may play a causal role. For example, alcohol consumption by the victim as well as by the perpetrator has been found to be important. But in addition, alcohol use by bystanders has been found to play a role. Alcohol may affect a range of interactional and situational aspects of the violent event, but currently we cannot measure this role or establish its causal role. Therefore, studies of changes in alcohol consumption and the relationship with violence can be complicated. For example, a particular policy change may alter not only the quantity of alcohol consumed but also the drinking contexts and the cultural expectancies about the effects of drinking.

It may be useful in examining the role of alcohol in violence to consider three different classes of violent situations:

·        domestic violence involving sexual partners or other family and domestic relationships in which drinking is often a significant feature of the relationship

  • tavern and street violence, primarily between men and often taken for granted culturally as a part of tavern drinking
  • collective violence (for example, wartime atrocities)


Each of these types of violence may yield to different approaches to prevention.

More aggregate level studies are needed to track the impact of policy changes. In addition, individual and event-level studies are needed, including


  • qualitative studies of drinking and violence in particular subcultures
  • studies of expectancies about and experiences with alcohol and violence cross-culturally
  • studies of the use of alcohol as an excuse for violence or as an aggravating factor in legal proceedings
  • observational studies of violent incidents in public drinking places
  • studies of drinking in violent events based on large general population samples
  • case studies of alcohol-involved violent events


Policy implications

Given that a significant proportion of violent incidents can be attributed to alcohol consumption, the question remains as to the best policy options for reducing alcohol consumption and by what mechanism will these policy changes achieve a reduction in violence. For example, while it may seem logical that policies aimed at curbing drinking by heavy drinkers would have the greatest impact on violence, this may not, in fact, be the case. Since moderate drinkers greatly out number heavy drinkers, even changes in drinking among people who are not heavy consumers may result in a significant reduction in violence. there is support for this application of the “prevention paradox”, but there is also reason to believe that alcohol-related violence is most often associated with intoxication and therefore that reducing incidents of intoxication would be an effective strategy for preventing violence.



While more study may help further define the proportion of violent incidents in which alcohol plays a role, too much attention to exact estimates of the relationship may only divert attention from the more important issues of how to prevent alcohol-related violence. A detailed understanding of the various causal pathways and of potential points of intervention that can be used in preventing violence is needed to make a difference in the rate of alcohol-related violence.





A number of studies have demonstrated significant and positive associations between alcohol consumption and rates of criminal violence (Skog & Bjørk, 1988; Lenke, 1990; Parker  & Rebhun, 1995; Norström, 1998; Parker & Cartmill, 1998). A large proportion of violent offenders had been drinking before the violence occurred. About 40 percent of such offenders in state and local jails in the United States had been drinking at the time of the offense for which they were jailed.  In surveys of victims of violent crimes in the United States, about 35% of those who respond to the question perceive the perpetrator to have been drinking (Greenfeld, 1998).  From an epidemiological point of view, stronger evidence is derived from a number of population studies of alcohol consumption and violent behavior, which have also demonstrated a significant association between the two (e.g. Fillmore, 1985; Choquet et al, 1991; Valois et al, 1995; Orpinas et al, 1995; Room et al, 1995; Rossow, 1996a; Borowsky et al, 1997; Grunbaum et al, 1998; oriel & Fleming, 1998; Valois & McKewon, 1998; Rossow et al, 1999; Scott et al, 1999).

The causal significance of these findings, however, is still a matter of dispute. The 19th-century temperance literature, for instance, states that three-quarters of all crime was due to drinking (Levine, 1983:136-138). On the other hand, perhaps partly in reaction against these views, many criminological discussions deny that alcohol has been shown to cause violence.  Thus the Canadian Supreme Court could cite a number of sources to support its conclusion that “studies demonstrate that the consumption of alcohol is not the cause of the crime” (Room, 1996).

The argument is primarily about the meaning of “causation”.  One view, which rules out alcohol as a causal agent, is that the term “cause” should only be used with respect to human agency.  Related to this has been the concern, in the context of discussions of male violence against women, that a formulation like “alcohol causes violence” is among “many distracting theories of violence which act as obstructions to social change” by diverting responsibility from the perpetrator (McGregor, 1990).

Another view which rules out alcohol as a cause of violence is a requirement that the relation be necessary and sufficient before using the term “cause”.  The criminology literature often tends towards this restrictive view of causal connections.  In this vein, Lipsey et al. (1997:277-278) review a variety of publications, each of which, they acknowledge, provide “evidence of an alcohol-violence association which is consistent with a causal interpretation”, yet shy away from attributing causation: “the causal issue is still cloudy and uncertain”.  From their discussion, it seems clear that they hanker for a “broad, reliable, ‘main effect’ of alcohol on violence”, and are reluctant to use causal language in its absence.

At the other end of the spectrum is public opinion about alcohol and violence, which is nearly unanimous, at least in North America.  Fully 86 percent of Ontario adults agree that “if someone is very drunk” it is likely to “make them aggressive and possibly violent”;  87 percent agree that “a lot of violence in society could be eliminated if people didn’t get drunk” (Paglia and Room, 1998).

Somewhere between these polar positions is the stance of the epidemiological literature. Epidemiology hews to guidelines for attributing cause that require neither a necessary nor a sufficient relation. This reflects the fact that it is quite typical for many factors to play a role in an illness or death, and for a particular kind of illness or death to occur also in the absence of a putatively causal factor.  Thus not all cigarette smokers get lung cancer, and not all patients with lung cancer have been cigarette smokers.  In an epidemiological framework, then, causes are often multiple and indeed conditional, and at a population level the causal connection is probabilistic rather than all-or-nothing.  Sets of guidelines to determine causal attribution are commonly explicitly stated in epidemiology; one of the most used was formulated by Bradford Hill (Hill, 1965)[1].

Using an epidemiological frame, we contend that ample evidence now exists in the empirical literature to regard alcohol as a cause of violence.  This does not mean that, at the level of particular events, the relationship is either necessary or sufficient.  Usually, the relationship is conditional: it is drinking in combination with other factors that is implicated.  Nor does the relationship necessarily operate through a psychopharmacological action in a perpetrators brain.  For one thing, drinking by the victim is often a crucial factor, and this factor is often involved in the causal chain. More generally, the relationship may operate through drinking as a cultural and circumstantial practice, that is, through the understandings and behaviors that are part of the drinking occasion.  At a population level, however -- the level that matters for epidemiology and public health -- we can say with some confidence that more drinking tends to result in more violence, and less drinking in less violence (Room, 1983).  Though the strength of this relationship varies in different cultural contexts, new evidence at the aggregate level indicates a relationship across a broad range of cultural contexts (Rossow, 2001).

In the epidemiological literature, the potential impact of a risk factor for some disease (or other outcome measure) is defined as the “proportion of all new cases in a given period that are attributable to the risk factor of interest” (Kleinbaum et al., 1982). This is often termed the “attributable fraction” (Ouellet et al, 1979)[2].  “Attributable”, in this formulation, is “the difference between all new cases in a period and the number of new cases that would have occurred in the absence of the exposure” (Kleinbaum et al., 1982).  This is a meaningful concept from a policy point of view, because it implies the potential to prevent the occurrence of new cases by eliminating the particular risk factor. This is also in line with Pernanen's (1995) suggestion that the causal role of alcohol in violence is more fruitfully approached when applying a more inclusive term that “coincides with the idea of 'had it not been for alcohol'”, rather than a “single-minded focus on hard-science direct causal connections.”


Some methodological considerations in estimating attributable fractions


The attributable fraction (AF) can be calculated on the basis of two estimates derived from analyses of empirical data:  an estimate of the association between the exposure (alcohol) and the outcome (violence), and  an estimate of the magnitude of the exposure (alcohol). This calculation can be done either on individual level data[3] or on aggregate level data[4]. The primary problem in applying these formulae lies in the question of how to measure the association only where the drinking is causally significant.  Arriving at “sound” estimates of the association between alcohol consumption and violent behavior is hampered by various kinds of methodological problems.

Studies based on individual-level data have generally suggested that a substantial proportion of violence is somehow related to alcohol consumption (e.g. Roizen, 1981). But although many studies have found that a high proportion of violent offenders were under the influence of alcohol at the time of the violent crime, it may well be argued that many of these violent crimes might have taken place even if alcohol were not a part of the event. To the extent this is true, the proportion of violent incidents where alcohol was somehow involved is larger than the proportion of incidents that could be attributed to alcohol use. This has been demonstrated, for instance, by Norström (1998), who found that some 75-80 percent of those committing criminal assaults in Sweden were intoxicated at the time of the crime, whereas his estimate of the proportion of assaults attributable to alcohol was far less (47 percent).

An alternative to individual-level studies as a strategy for assessing attributable fractions is to apply time series analyses of aggregate-level data. A number of such studies of alcohol and violence have been carried out over the past decade or so (e.g. Skog & Bjørk, 1988; Lenke, 1990; Norström, 1993; Parker and Rebhun, 1995; Norström, 1998; Parker and Cartmill 1998; Rossow, 2001). These studies have generally demonstrated that an increase in the populations drinking level (or an indicator of this) is followed by an increase in rates of violence in that population. The findings are interpreted as a reflection of two underlying mechanisms:  an increase in total consumption implies an increase in the proportion of people who are drinking heavily and in that state are more prone to behave aggressively or violently; and an increase in total consumption also implies an increase in the number of drinking occasions that lead to acute intoxication, which may trigger violent behavior among otherwise moderate consumers. Particularly when the analyses are based on differenced time series, the effect estimate reflects the impact of a change in alcohol consumption on the change in rates of criminal violence.  It is assumed that a spurious correlation due to various individual or social environment characteristics would not be not reflected in annual changes at the aggregate level (Lenke, 1990; Norström, 1998).  Thus, the estimated effect or association based on such analyses can be taken to indicate the causal impact of alcohol consumption on violence. Causality in this respect would also include what Pernanen (1995) refers to as second order causality, for instance the impact of alcohol on violence that is generated through alcohol's role as a 'social attractor' (the gathering of people in specific places) or as a 'courage enhancer'. We should, however, also keep in mind that there is often a problem of Type II error (missing a relationship that is really there) and of  huge confidence intervals around the point estimates from time series analyses on differenced data, due to measurement problems or insufficient length of the data series.  Moreover, the estimate of an attributable fraction is also based on extrapolation from the empirical data. Thus, the estimates of association, as well as the attributable fractions, from these kinds of analyses should be interpreted with some caution.

Spatial correlations or cross-sectional aggregate level analyses have also been applied in a number of studies on alcohol and violence (e.g. Lester, 1995; Parker & Rebhun, 1995; Stevenson et al, 1999). Such analyses generally test for the finding that the higher the alcohol consumption, the higher the violence rate across geographical areas. But, as rates of violence most probably result from a complex of factors, control of all relevant confounders is difficult, and hence the results offer much weaker evidence concerning the existence and strength of a causal relationship.

With many multifactorial phenomena it can be argued that the impact of a specific risk factor depends on other (risk) factors.  As we have indicated, this is also the case with alcohol and violence (Skog & Bjørk, 1988; Lenke, 1990). Both experimental studies and observational studies indicate that the impact of alcohol in aggressive or violent behavior depends on the presence of other factors which may trigger or facilitate violence (see Lenke, 1990; Graham et al, 1996 for reviews). This implies that cases of violence may not be attributed to one single risk factor, but often to the co-occurrence of several risk factors. In turn, this means that summing up the attributable fractions for various risk factors would yield more than 100 percent (Rothman, 1986, p.14). However, this may not gainsay that alcohol makes a substantial contribution, and one that is potentially preventable. The issue around multiple causation, attributable fractions,  and the implications for potential to prevent new cases has been discussed by O'Carroll (1993) in relation to suicide. He argues that even though a very large proportion of cases can be attributed to one or two single factors, this does not mean that little potential exists to prevent new cases by strategies directed towards other risk factors, as these other factors may also contribute significantly and thus be subject to preventive strategies.  His arguments are just as relevant with respect to violence.


Estimated attributable fractions: a review of empirical findings


Individual level data:

In a few review studies over the past decade, some estimates have been given on the proportion of violent assaults that are attributable to alcohol consumption. In their review of U.S. data, Shultz et al. (1991) summarized the fraction of assaults attributable to alcohol as 46 percent, very similar to an estimate of 47 percent in a meta-analytic review oriented to the Australian situation (English et al.,1995). These estimates can be criticized justly; and studies of economic costs of alcohol use in the cost-of-illness tradition have generally used a lower attributable fraction, on the assumption that much drinking prior to violent acts is not causal.  Thus Single et al. (1996) and Xie et al. (1996) used 27 percent as the attributable fraction for assaults in their estimates of the alcohol fraction of assaults in Canada.  Similarly, in their cost-of-illness estimates for the United States, the Lewin Group (1998) used 22.5 percent for sexual assault, and 30 percent for homicide and other assaults (Table 6.8).  These figures were based on taking half of the proportion of inmates in prisons and jails who reported having been drinking before committing a violent crime. The authors acknowledge the rationale for this procedure as “one of default” (Lewin Group, 1998, section, as it is based on a study 20 years ago of attributions by inmates. 

Concluding a literature search for material relevant to estimating attributable fractions for alcohol and other drugs in relation to crimes in Canada, Pernanen and Brochu (1997) conclude that a sizeable presence of alcohol is found in almost all studies on assaults and homicides.  Canadian studies generally find lower rates than U.S. studies, with a modal value perhaps around 40 to 45 percent of perpetrators drinking, while the most representative U.S. studies report proportions of about 50 to 60 percent. The highest values are found in Scandinavian countries with shares of drinking offenders typically in the 70 to 80 percent range.  These figures do not refer to attributable fractions, but rather the proportion of perpetrators who had been drinking; however, the variation in these figures may be indicative of a variation in attributable fractions as well. While the variation in these figures may reflect differences in how carefully the presence of alcohol is reported, probably a core of factual differences also exists in drinking patterns and situations in different cultures. “Factual differences in drinking patterns” (Pernanen & Brochu, 1997, p.8) as well as in norms and expectancies about behavior while drinking, are likely to play a role in the variation.


Aggregate level data:

Based on time series analyses of aggregate level data, Lenke estimated that a little more than half of the assaults in Sweden from 1920 to 1984 were alcohol-related,[5] with even higher fractions in Norway and Finland. His estimated model on French data resulted in a lower fraction -- about 33 percent of the assaults in France from 1919 to 1958 could be attributed to alcohol (our calculations based on Lenke's figures). Skog and Bjørk's  (1988) estimates from similar analyses, but on a shorter times series from Norway (1958-1977), imply that around half of the convicted assaults could be attributed to alcohol. Norström (1998) estimated the fraction of homicides in Sweden (1956-1994) attributable to alcohol to be 69 percent, and the attributable fraction of assaults to be 47 percent. Note that these attributable fractions include the second order causality, in Pernanen’s phrase, from drinking’s concomitants.

Large confidence intervals surround these estimates; and hence apparent similarities as well as differences should not be given too much weight until they are further validated. However, there is no doubt that the impact of alcohol consumption on rates of violence is substantial, and that incidents of violence can be prevented by significant reduction in the overall intake of alcohol and/or heavy drinking episodes. Corroborating evidence comes from data from sudden and large changes in alcohol consumption (“natural experiments”).  Von Hofer and Tham (1989) observed that when Swedish alcohol consumption fell dramatically from about 6 to 1 liter per capita during World War I (due to rationing), assault rates fell from 70 to 40 per 100 000 inhabitants, and they found no other explanation than the huge decrease in alcohol sales (cited in Lenke, 1990).  Such evidence is much in line with Lenke’s estimate that around half of assaults in Sweden are attributable to alcohol consumption.  Likewise, Shkolnikov and Nemtsov (1997) found that the age-standardized death rate for Russian males from homicide fell dramatically during the Gorbachev-era anti-alcohol campaign, which they estimate reduced the real alcohol consumption (including illicit supplies) from 14.2 liters of ethanol per capita in 1984 to 10.7 in 1987. For the same years, the males deaths from homicide fell from 19.3 per 100 000  to 11.5, suggesting that in Russia at this time considerably more than half of the homicides were attributable to drinking.  In the 1990s, the alcohol consumption increased significantly again in several of the former Soviet republics, and this was followed by a significant increase in homicide rates, for instance in Estonia (Kaasik et al, 1998) and in Russia (e.g. Shkolnikov et al, 1995.

Despite all these studies, a more systematic description of associations between alcohol consumption and violence is still needed from both individual level and aggregate level data and from various countries and drinking cultures. Parker and Rebhun’s analysis (1995), using a combination of time-series with cross-sectional data, and Parker and Cartmill’s (1998) study of beverage specific effects on homicide rates are among the very few studies using time-series methods on population-level data in North America.


The cross-cultural perspective


Until now, the literature on attributable fraction for alcohol has usually implicitly assumed that there will be, in principle, a single relative risk applicable everywhere for a given volume of drinking.  The attributable fraction would then vary in different populations only according to the proportion of the population at each level of volume of drinking (as in the approach taken by English et al., 1995).  This algorithm has been modeled on the assumptions about the relation between volume of drinking and such chronic diseases as cirrhosis of the liver.  Recently, however, it has become apparent that casualties and social problems from drinking are often more closely related to the pattern of drinking -- particularly whether large amounts are drunk on an occasion -- than to the volume of drinking (e.g., Room et al., 1995; Midanik et al, 1996).  Alcohol epidemiologists have thus called for substantial attention to patterns of drinking, as well as the overall volume, in future epidemiological studies (Rehm, 1998).  The influence of pattern of drinking on the link with many social and health problems means that the attributable risk may vary between two societies or groups with the same level of volume of drinking, if their patterns of drinking vary.

Beyond this, with respect to violence in particular, it is clear from the ethnographic record (MacAndrew and Edgerton, 1969) that people in different cultures behave very differently when under the influence of an equal amount of alcohol.  What MacAndrew and Edgerton call “changes-for-the-worse” in behavior after drinking, including violent behavior, is thus influenced not only by individual and situational factors, but also by specific cultural expectations and patterns.  This implies that the relative risk of violence for a given pattern of drinking (and thus also the attributable fraction) may vary from one culture to another.

While the literature offers some typologies of the cultural position of drinking and some hypotheses about the relation of rates of violence to them (Room and Mäkelä, 2000), empirical work on testing and refining such hypotheses is still at its beginning.

Lenke's comparative analyses of alcohol sales and assaults in four Nordic countries and in France (1990) suggested a stronger association between alcohol and violence in Finland, Sweden and Norway, where the drinking pattern is considered more explosive, as compared to Denmark and France. Rossow (2001) has pooled estimates from time series analyses of alcohol sales and homicide rates in 14 European countries (1950-1995) across three regions and found that the impact of alcohol on homicide was highest in the three Northern countries (Finland, Norway and Sweden), and lowest in the Mediterranean countries (France, Italy, Portugal and Spain), with the intermediate western European countries in between.  The differences in associations are statistically significant (see table 1).

A variety of studies in the United States have suggested that there is stronger relationship between alcohol and violence in the South than elsewhere in the country (Levinson, 1983).  However, Parker and Rebhun (1995) found mixed results in their pooled cross-sectional time series analysis concerning a stronger relationship in the South.  Levinson’s conclusion from almost two decades ago remains current: that there is a pressing need for more research on alcohol-related aggression in American subcultures (1983, p. 317)


From the European analyses, it seems that a stronger impact of alcohol on violent crimes is found in drinking cultures where acute intoxication to a larger extent characterizes the drinking pattern. It is worth noting that parallel cross-cultural variations can be found with respect to the alcohol and suicide association. Again based on time series analyses, a generally stronger association between alcohol consumption and suicide rates has been found in the more explosive drinking cultures in Finland, Sweden and Norway than in, for instance, France, Denmark and Portugal (see Norström, 1995a, Rossow, 1996b for reviews). Thus, there appears to be some consistency in the relative strength of association between alcohol consumption and the level of aggressive behavior in a culture, whether the aggression is directed towards self or others. The implications of such findings for multi-cultural societies like the United States remain to be determined. While much diversity exists in drinking patterns between ethnoreligious groups in the United States, the patterns of descendants of immigrants are not the same as either of those of the ancestral culture or those of others in a multicultural society (Room, 1985; Kitano et al, 1988; Caetano & Medina Mora, 1988). Besides potential differences in patterns of drinking, there are also potential differences in the cultural meaning of intoxication and expectancies about violence. In one study, Afro-American informants perceived white Americans as acting “crazier” than Afro-Americans after drinking (Herd, 1983), and some expectancy studies have found white students endorsing more effect expectancies from drinking than Afro-American students (Bell, 1986; Friend, 1993; Miller, 1997). But a review of the literature found the evidence mixed on whetehr a stronger relation of drinking to violence existed among white Americans than among Afro-Americans (Roizen, 1981).



Challenges for the future


Estimating society-specific attributable risks

Attributable fraction is in the end a population-level concept.  It asks what proportion of cause of death or other adverse condition is attributable to a particular risk factor, regardless of whether or not the factor can be causally implicated in particular deaths or adverse events.  In the end, the policy question behind the attributable fraction concept is: if a certain fraction of consumption, or of high-risk drinking occasions, could be taken away, how much would the rates of the adverse events fall?  From this perspective, the most direct measurement of attributable fraction is from studies at the population level, rather than studies of individual events.

 In estimating attributable fraction, studies at the level of individual events have two major flaws. 

  • There is no established and valid way to estimate the specific role of alcohol in a violent event.  The case-control procedures that are commonly used to establish attributable fractions for accidental injuries or deaths depend on an assumption that is not applicable for violent behavior: that no human intention to injure or be injured is involved.  Once intention is admitted to the model, as it must be with respect to violence, a case-control model built around time of day and week and physical circumstances, as in traffic casualty studies,  becomes problematic, since the intention to commit the violence may dictate the whereabouts in space and time, and the intention may in turn be affected by intoxication.  Whether drinking or intoxication is involved in individual violent situations and events can be measured.  Whether the participants believe that drinking was causal in the event can be asked.  But so far, there is no way of establishing an objective attributable fraction that reflects whether or not the event would have occurred without the drinking.  This is what has led the economic-costs studies to use default rules-of-thumb, e.g.  attributing half the reported proportion of had-been-drinking perpetrators.  Using all of the reported proportion would clearly be indefensible; but at this point there is defensible rationale either for using any particular fraction of the proportion.

  • Most studies at the level of individual events miss measuring some of the potential ways in which drinking may be causally implicated.  To measure only the intoxication of the person defined as a perpetrator misses the potential role of intoxication by the person who ends up as victim.  To measure only the intoxication of the victim, as in coroners studies of homicide victims, misses the potential role of the perpetrator.  Both kinds of study miss the potential role of the intoxication of bystanders or participants in the event, a role that can be quite crucial (Wells and Graham, 1999).  Studies in the tradition of Wolfgang’s classic work (1958) on alcohol in homicide do manage to catch some of alcohol’s role in the interactional and situational aspects of the violent event, but must essentially rely on police and investigative attributions of the alcohol relationship.  Again, these are not strong evidence for arriving at a causally oriented attributable fraction.


Studies of change over time in a given polity have the potential to overcome these flaws of individual-level studies.  The causal attribution is based on a direct measurement of how much the dependent variable changes when the alcohol variable changes.   Obviously, interpreting this covariation as causal depends on all else being equal.  The different aggregate-level studies have different ways of controlling for other factors.  One major control, used by the strongest studies, is to look at variation over time in a single polity. This automatically controls many of the particular cultural and socioeconomic features which might otherwise enter into the explanation. ARIMA-type analyses typically use differenced data to control out long-term trends.  Studies of particular events in the alcohol market -- for instance, liquor store strike studies -- are often able to use a control period in the same population in another year.

Nevertheless, estimates from aggregate-level studies have their own drawbacks.  Cross-sectional aggregate analyses are particularly vulnerable to influence by unmeasured variables, and in our view are of little value for estimating attributable fractions.   But the studies of change over time have problems, too, even if they are less serious. As we have already mentioned, the small numbers of data points in many ARIMA analyses make the confidence intervals large, and the conservative procedures used also raise the possibility of important relationships remaining statistically insignificant. It should also be kept in mind that the cause of  the change that is registered in the studies may be a broad penumbra of effects related to whichever particular alcohol variable is measured (often, the per-capita consumption).  The attributable fraction measured by these aggregate studies is not the fraction attributable to ethanol as a psychoactive substance, but rather to amounts and practices and the social interactions that are attached to drinking.  The attributable fraction thus tells us something about the size of the problem, but does not dictate the policies which might counteract it.  Besides changing the overall amount of drinking, policy options may include, for instance, changes in drinking contexts, or in cultural expectancies about behavior while drinking. 

As we have mentioned, some North American studies show at the aggregate level the effects of changes in drinking, but this tradition of studies has been much more strongly developed in the Nordic countries.  While the Nordic countries have particularly rich data for such analyses, much can be done with the data available in any developed society, as the current European Comparative Alcohol Study is showing (Rossow, 2001).  Future applications in North America of attributable fractions for violence should be able to depend on studies of this sort done on North American populations.

The patterns of change in the aggregate-level analyses can also contribute hints about the social processes involved, and thus about potential targets and entry-points for efforts at social change.  The dramatic fall in homicides in the Gorbachev era in Russia, for instance, suggests that homicide-prone drinkers must have been affected more than other drinkers by the changes in the market.  Other data are beginning to suggest that problem-prone drinkers may be particularly affected by some alcohol control strategies.  But, in general, aggregate-level studies are not an effective way to study and tease apart the mechanisms involved in the relations between drinking and violence. 



The policy implications


Given that a significant proportion of violent incidents seem to be attributable to alcohol consumption, a question for policy is whether preventive strategies should be aimed at particular  high-risk individuals or the whole population -- i.e., should the focus be on what are generally termed high-risk strategies or on population strategies?  This will basically depend on what the risk curve between alcohol exposure and violent behavior looks like, but given a (more or less) straight linear risk function, it can be assumed that those with a moderate volume of consumption are responsible for the majority of the violent incidents that are due to alcohol (see Skog, 1999 for details). This is often referred to as the prevention paradox (Rose, 1981; Kreitman, 1986; Skog, 1999). Doubts have been raised with regard to whether the prevention paradox is valid for various acute alcohol problems (Stockwell et al, 1996). However, several individual level studies have indicated that there seems to be a (more or less) linear relationship between alcohol consumption and violent behavior, implying that the prevention paradox may be valid for violent behavior (e.g. Room et al., 1995; Rossow, 1996a; Rossow et al., 1999). One might therefore argue that a prevention strategy aimed at all consumers (population strategy) would yield a larger reduction in violent incidents than a prevention strategy aimed only at high-risk individuals (high-risk strategy). Norström (1995) argued that the advantage of a population strategy is most marked in connection with accidents and suicide (which probably have a risk function more like that for violence than like that for cirrhosis).  But even a high-risk strategy, he felt, would yield a significant impact on accidents and suicide as well. Since alcohol-related violence is associated particularly with intoxication, preventive strategies might be aimed not only at reducing overall consumption but also at preventing drinking events becoming occasions of intoxication (Grover, 1999).  However, a need still exists for more studies evaluating the outcome of such strategies.


Studying the patterns of connection between drinking and violence

In studying patterns of connection between alcohol and violence, an important step is to differentiate between different kinds of violence, since the patterns of linkages between the violence and the drinking may well differ for different types of violence. A rough differentiation which may prove useful is between three classes of violent situations:

  • Domestic violence, involving sexual partners and other kinds of family and domestic relationships, often long-lasting.  The violence in such situations often enforces (or, less often, counters) patterns of intimate domination, and the links with drinking may be ties to the signification and significance of drinking in the relationship (Room, 1980).
  • Tavern and street violence. Disputes and brawls, primarily between men, are a taken-for-granted aspect of tavern drinking in many cultures.  Frequently such disputes also spill onto or start on the street.  Parker and Rebhun (1995) suggest that their changing results for the American South may reflect a shift in the mixture of violence between this and the first type of violence.
  • Collective violence. The role of alcohol in wartime and other collective violence has not been systematically studied.  It is clear that drinking has long had a role in war, as much in celebrations and in expressions of solidarity as in preparing for and enacting violence. A consistent theme in reportage of wartime atrocities is the drunkenness of those committing the atrocities, whether in Nazi-occupied Europe or more recently in “ethnic cleansing” episodes in the Balkans.


We have already marked on the need for further aggregate-level studies, and in particular for the tradition of differenced time-series studies on aggregate data to be used more extensively on North American data -- for instance on U.S. states and Canadian provinces.  Beyond this, a variety of studies at the level of particular events and individuals are needed, to elucidate patterns of relationship.  These include:

  • qualitative studies of drinking and violence in particular subcultures
  • studying expectancies and experiences about alcohol and violence cross-culturally

  • studying the excuse-value (and the opposite, the value for incrimination) of intoxication in legal proceedings, as an indicator of how societies conceptualize and deal with alcohol-violence relationships (such studies should where possible be cross-culturally comparative).
  • quantified observational studies of violent incidents in public drinking places (in bars, among sports spectators, etc.)
  • collecting data about participants drinking in violent events in large general-population samples (such data has the potential to contribute information on attributable fractions from individual-level data).
  • detailed studies of alcohol-involved violent events.  These might be modeled on the multidisciplinary accident investigations (Shinar et al., 1983) which have been used for transportation crashes.



We are thus still a considerable distance from being able to assign relative risks and attributable fractions for alcohol’s role in violence in a given society, on the basis of a well-developed epidemiological literatures for a range of types of societies.   It is worth keeping in mind, however, that the search for defensible attributable risks applicable cross-culturally is not the end of our journey, and too much fetishistic attention to this search may divert us from more important paths.  In the end, the exact size of an attributable risk is primarily of importance to politics rather than to policy or public health programming.  A large attributable risk may serve to draw political attention to a factor otherwise neglected -- what is sometimes called the gee whiz factor.  The ability to compare attributable risks for different factors which are computed according to common principles may have some importance in priority setting in the policy process.  Conceivably, doing this might be used to elicit support for general policy measures – for instance, alcohol rationing or higher alcohol taxes – for which there is evidence of effects on alcohol-related violence, but which are at present politically unpopular or inconceivable. in an era where political realities favor strategies that prevent the violence without substantially interfering with the drinking (Room, 2000) a detailed understanding of the various causal pathways, and of the potential points of intervention in them, is needed to make a difference in the rate of alcohol-related violence.   





Bell, DD. (1986) Alcohol related expectancies and patterns of use in Black and White male college students. Dissertation Abstracts International 46 (9): 3208B


Borowsky IW, Hogan M and Ireland M (1997) Adolecent sexual aggression: Risk and protective factors. Pediatrics 100: E71-E78.


Boyatzis RE (1975) The predisposition toward alcohol-related interpersonal aggression in men. Journal of Studies on Alcohol 36: 1196-1207.


Caetano R and MedinaMora ME (1988) Acculturatio and drinking among people of Mexican descent in Mexico and the United States. Journal of Studies on Alcohol 49: 462-471.


Choquet M, Menke H and Manfredi R. (1991) Internpersonal aggressive behaviour and alcohol cosumption among youth urban adolescents in France. Alcohol and Alcoholism 26: 381-390.


Cole P. and MacMahon B. (1971) Attributable risk percent in case control studies. British Journal of the Society of Preventive Medicine 25: 242-244.


English DR, Holman CDJ, Milne E, Winter MG, Hulse GK, Codde JP, Bower CI, Corti B, de Klerk N. Knuiman MW, Kurinczuk  JJ, Lewin GF and Ryan GA (1995) The quantification of drug caused morbidity and mortality in Australia, 1995 edition. 2 vols. Commonwealth Department of Human Services and Health, Canberra.


Fillmore KM (1985) Social victims of drinking. British Journal of Addiction 80: 307-314.


Friend R. (1993) Drinking practices and expectancies in undergraduate males as a function of ethnicity and fraternity membership. Dissertation Abstracts International. 54 (2): 1095B.


Graham K, Schmidt  G  and Gillis K. (1996) Circumstances when drinking leads to aggression: An overview of research findings. Contemporary Drug Problems 23: 493-558.


Greenfeld  LA. (1998) Alcohol and Crime: An Analysis of National Data on the Prevalence of Alcohol Involvement in Crime. Bureau of Justice Statistics, U.S. Department of Justice,  Washington, DC


Grover PL. (1999) (ed) Preventing problems related to alcohol availability: environmental approaches: reference guide. DHHS Publication No. (SMA)99-3298.  U.S. Center for Substance Abuse Prevention, Rockville, MD. Available at


Grunbaum JA, BasenEngquist K and Pandey D. (1998) Association beteween violent behaviors and substance use among Mexican-American and non-Hispanic white high school students. Journal of Adolescent Health 23: 153-159.


Gustafsson R. (1995) Is it possible to link alcohol intoxication causally to aggression an violence? A summary of the Swedish experimental approach. Studies on Crime & Crime Prevention 4: 22-42.


Herd D (1983) Commentary. In: Room R and Collins G eds. Alcohol and disinhibition: Nature and meaning of the link. NIAAA Research Monograph no. 12. DHHS Publication no. ADM 83-1246. National Institute on Alcohol Abuse and Alcoholism, Rockville, MD.


Hill AB. (1971) Principles of medical statistics. 9th Edition, pp. 309-323. Oxford University Press New York.


Kaasik T, Andersson R and  Horte LG. (1998) The effects of political and economic transitions on health and safety in Estonia: An Estonian-Swedish comparative study.  Social Science & Medicine 47: 1589-1599.


Kitano HHL, Chi I, Law CK, Lubben J and Rhee SY. (1988) Alcohol consumption of Japanese in Japan, Hawaii, and California. In: Towle LH and Harford TC eds. Cultural influences and drinking patterns: A focus on Hispanic and Japanese populations. NIAAA Research Monograph no. 19. DHHS Publication no. ADM 88-1563. National Institute on Alcohol Abuse and Alcoholism, Rockville, MD.


Kleinbaum DG, Kupper LL  and Morgenstern H. (1982) Epidemiologic research. Principles and quantitative methods. van Nostrand Reinhold, New York.


Lenke L. (1990) Alcohol and criminal violence - Time series analyses in a comparative perspective. Almquist & Wiksell, Stockholm.


Levin ML. (1953) The occurrence of lung cancer in man. Acta Unio International Contra Cancrum 9: 531-541.


Levine HG. (1983) The good creature of God and the demon rum: colonial American and 19th century ideas about alcohol, crime and accidents, pp. 111-161. In: Room R and Collins G, eds. Alcohol and Disinhibition: Nature and Meaning of the Link, NIAAA Research Monograph No. 12. DHHS Publication No. (ADM) 83-1246. National Institute on Alcohol Abuse and Alcoholism, Rockville, MD.


Levinson D. (1983) Alcohol use and aggression in American subcultures, pp. 306-321. In: Room R and Collins G. eds. Alcohol and disinhibition: nature and meaning of the link. NIAAA Research Monograph No. 12. Washington DC, U.S.G.P.O., DHHS Publication No. (ADM) 83-1246.


Lewin Group, The (1998) The Economic Costs of Alcohol and Drug Abuse in the United States.  Report prepared for the National Institute on Drug Abuse and the National Institute on Alcohol Abuse and Alcoholism, May.  Available at


Lilienfeld AM and Lilienfeld DE (1980) Foundations of epidemiology. 2nd ed. Oxford University Press, New York.


Lipsey MW, Wilson DB, Cohen MA and Derzon JH. (1997) Is there a causal relationship between alcohol use and violence? A synthesis of evidence, pp. 245-282.  In: Galanter M. ed. Recent Developments in Alcoholism: vol. 13, Alcohol and Violence. Plenum,  New York.


MacAndrew C and Edgerton RB. (1969) Drunken Comportment. Aldine, Chicago.


McGregor H. (1990), Domestic violence: alcohol and other distractions - a grassroots perspective, pp. 59-66. In: Venon  J, ed. Alcohol and Crime. Australian Institute of Criminology Canberra.


Midanik LT, Tam TW, Greenfield TK and Caetano R. (1996) Risk functions for alcohol-related problems in a 1988 U.S. national sample. Addiction 91: 1427-1437.


Miettinen OS. (1974) Proportion of disease caused or prevented by a given exposure trait or intervention. American Journal of Epidemiology 99: 325-332.


Miller, TL. (1997) Psychological risk factors for alcohol use and abuse among Black Americans. Dissertation Abstracts International. 58(4): 2131-B.


Norström T. (1989) The use of aggregate data in alcohol epidemiology. British Journal of Addiction 84:  969-977.


Norström T. (1993)  Family violence and total consumption of alcohol. Nordic Studies on Alcohol and Drugs 10:  311-318.


Norström T. (1995)  Prevention strategies and alcohol policy. Addiction 90: 515-524.


Norström T. (1998) Effects on criminal violence of different beverage types and private and public drinking.  Addiction  93: 689-699.


O'Carroll P. (1993)  Suicide causation - pies, paths, and pointless polemics. Suicide & Life-Threatening Behavior 23:27-36.


Oriel KA and Fleming MF. (1998) Screening men for partner violence in a primary care setting: A new strategy for detecting domestic violence. Journal of Family Practice 46: 493-498.


Orpinas PK, BasenEngquist K, Grunbaum JA and Parcel GS. (1995) The co-morbidity of violence-realted behaviors with health-risk behaviors in a population of high-school students. Journal of Adolescent Health 16: 216-225.


Ouellet BL, Romeder JM and Lance JM (1979) Premature mortality attributable to smoking and hazardous drinking in Canada.  American Journal of Epidemiology 109: 451-463.


Paglia A. and Room R. (1998) Alcohol and aggression: general population views about causation and responsibility. Journal of Substance Abuse 10:199-206.


Parker R. N. with Rebhun M A. (1995): Alcohol and Homicide. A deadly combination of two American traditions. State University of New York Pressm,  Albany.


Parker RN and Cartmill RS. (1998) Alcohol and homicide in the United States 1934-1995, or one reason why U.S. rates of violence may be going down. Journal of Criminal Law and Criminology 88: 1369-1398.


Pernanen K. (1991) Alcohol in human violence. Guilford Press, New York.


Pernanen K. (1995).  The social cost of alcohol-related crime: conceptual, theoretical and causal attributions. Document available at:



Pernanen K and Brochu S. (1997) Attributable fractions for alcohol and other drugs in relation to crimes in Canada: Literature search and outlines of data banks. Prepared for the Canadian Centre on Substance Abuse. Document available at and continuation addresses.


Rehm J. (1998) Measuring quantity, frequency, and volume of drinking. Alcoholism: Clinical and Expermintal Research 22 (Suppl. 2), 4S-14S.


Roizen J. (1981) Alcohol and criminal behavior among Blacks: The case for research on special populations. pp. 207-252.  In: Collins JJ ed. Drinking and crime. Guilford, New York.


Roizen  J. (1989) Alcohol and trauma. pp. 21-66 In: Giesbrecht N, Gonzales R. Grant  M, Österberg E, Room R, Rootman I and Towle L.  eds. Drinking and casualties: accidents, poisonings, and violence in an international perspective. Tavistock/Routledge, London.


Room R. (1980) Alcohol as an instrument of intimate domination, presented at the annual meeting of the Drinking and Drugs Division, Society for the Study of Social Problems, New York.


Room R. (1985) Foreword, pp. xi-xvii. In: Bennett LA and Ames GM eds. The American experience with alcohol: Contrasting cultural perspectives. Plenum, New York.


Room  R. (1996). Drinking, violence, gender and causal attribution: a Canadian case study in science, law and policy. Contemporary Drug Problems 23: 649-686.


Room R. (2000) Preventing alcohol problems: Popular approaches are ineffective, effective approaches are politically impossible. Presented at “Youth and the prevention of tobacco, alcohol and drug misuse: does it matter?”, Oslo, Norway, March 23-24.


Room R., Bondy SJ and Ferris J. (1995) The risk of harm to oneself from drinking, Canada 1989. Addiction 90: 499-513.


Room R and Mäkelä K. (2000).  Typologies of the cultural position of drinking.  Journal of Studies on Alcohol 61(3): 475-483.


Rossow I. (1996a) Alcohol related violence: The impact of drinking patterns and drinking context. Addiction 91: 1641-1651.


Rossow I. (1996b) Alcohol and suicide -- beyond the link at the individual level. Editorial. Addiction 91: 1413-1416.


Rossow I. (2001) Drinking and violence - a cross-cultural comparison of alcohol consumption and homicide rates in 14 European countries. Addiction, 96 (Supplement 1): S77-S92.


Rossow I, Pape H and Wichstrøm L. (1999) Young, wet and wild? Associations between       alcohol intoxication and violent behaviour in adolescence. Addiction 94: 1017-1031.


Rothman KJ. (1975) Alcohol. , pp. 139-150.  In: Fraumeni JF. ed.  Persons at high risk of cancer: an approach to cancer etiology and control. Academic Press,  New York.


Rothman KJ. (1986) Modern epidemiology. Little, Brown & Company, Boston.


Scott KD, Schafer J and Greenfield TK. (1999) Role of alcohol in physical assault perpetration and victimization. Journal of Studies on Alcohol 60: 528-536.


Shinar D, Treat JR and McDonald, ST. (1983) Validity of police reported accident data. Accident Analysis and Prevention 15: 175-191.


Shkolnikov V,  Mesle F  and Vallin J. (1995) Health crisis in Russia. 2. Changes in causes of deaths -- comparison with France and England and Wales from 1970 to1993.  Population 50: 945-982.


Shkolnikov VM and Nemtsov A. (1997) The anti-alcohol campaign and variations in Russian mortality. In: Bobadilla  JL, Costello CA and  Mitchell F. eds. Premature death in the new independent states. National Academy Press, Washington, DC.


Shultz JM, Rice, DP, Parker DL, Goodman RA, Stroh G Jr. and Chalmers N. (1991) Quantifying the disease impact of alcohol with ARDI software,  Public  Health Reports 106: 443-450.


Single E, Robson L, Xie X, and Rehm J. (1996) The Costs of Substance Abuse in Canada. Canadian Centre on Substance Abuse, Ottawa, Canada. Highlights on web at


Skog O-J. (1999) The prevention paradox revisited. Addiction 94: 751-757.


Skog O-J and  Bjørk E. (1988) Alkohol og voldskriminalitet. En analyse av utviklingen i Norge 1931-1982, (Alcohol and violent crimes. An analysis of the 1931-1982 trends in Norway). Nordisk Tidsskrift for Kriminalvidenskab 88: 1-23.


Stevenson RJ, Lind B and Weatherburn, D. (1999) The relationship between alcohol sales and assault in New South Wales, Australia.  Addiction 94: 397-410.


Valois RF, McKeown RE, Garrison CZ and Vincent ML. (1995) Correlates of aggressive and violent behaviors among public high school adolescents. Journal of Adolescent Health 16: 26-34.


Wells S. and Graham K. (1999) The frequency of third party involvement in incidents of barroom aggression. Contemporary Drug Problems 26:457-480.


Wolfgang  ME. (1958). Patterns in criminal homicide. University of Pennsylvania Press, Philadelphia.


Xie X., Rehm J, Single E and Robson L. (1996) The economic costs of alcohol, tobacco and illicit drug abuse in Ontario: 1992. ARF Research Document No. 127. Addiction Research Foundation, Toronto.








Table 1. Parameter estimates of the association between alcohol consumption and indicators of violence from time series analyses (ARIMA) in various countries or areas.




Violence indicator

Parameter estimate

Skog & Bjørk, 1988

Norway 1930-1982

Assault rates



Lenke, 1990

France 1919-1958

Assault rates



Lenke, 1990

Sweden 1950-1980

Assault rates



Lenke, 1990

Sweden 1921-1984

Homicide rates



Norström, 1998

Sweden 1956-1994

Assault rates



Norström, 1998

Sweden 1956-1994

Homicide rates



Rossow, 2001

North Europe 1950-1995

Homicide rates



Rossow, 2001

Central Europe 1950-1995

Homicide rates



Rossow, 2001

South Europe 1950-1995

Homicide rates




     Note: The estimates in the studies by Skog & Bjørk (1988), Lenke (1990) and Rossow (forthcoming) are based on semi-logarithmic models, which means that the parameter estimate indicates a relative (i.e. percentage) change in the dependent variable with a one unit (1 liter pure alcohol per inhabitant > 15 yrs) change in the independent variable, whereas Norströms study is based on linear models, which means that the parameter estimate indicates the absolute change in the dependent variable. Moreover, the estimates in Rossows study are based on pooling of estimates across several  countries (i.e. North Europe comprises Finland, Sweden and Norway; Central Europe comprises Austria, BRD, Denmark, Netherlands, Belgium, UK and Ireland; and South Europe comprises Italy, France, Spain and Portugal).




End notes:

[1]. Hill criteria for causal inference from associations: 1) strength, 2) consistency, 3) specificity,

4) temporality, 5) biologic gradient, 6) plausibility, 7) coherence, 8) experimental evidence, 9) analogy.


[2]. The terms used by various authors have differed.  Other terms used have included: “attributable risk” (Levin, 1953); “population attributable risk percent” (Cole & MacMahon, 1971); “etiologic fraction” (Miettinen, 1974), and “attributable proportion” (Rothman, 1986).


[3].  (1)   AF=    q(RR-1)       

    q (RR-1) + 1

where q is the proportion of the population exposed to the risk factor and RR is the relative risk (Lilienfeld & Lilienfeld, 1980).



[4]. (2) AF=   bX        


where b  is the (unstandardized) regression coefficient, X is the risk factor, and Y is the outcome measure, when a straight linear model is specified, or


(3) AF= 1 B e -bX, when a semi-logarithmic model is specified (Norström, 1989).


[5]. Lenke (1990) applied the term alcohol-related for violent crimes that  “probably not would have been committed without the presence of alcohol” in contrast to the term “alcohol involvement” implying that one or more of the persons involved had been drinking prior to the crime.