REVIEW OF STUDIES ON LIQUOR CONTROL AND CONSUMPTION Antony Davies, PhD, Duquesne University Policy Variables and Policy Targets Studies on the efficacy of alcohol controls have focused on the three broad categories of market intervention:  Demand mitigation. Laws aimed at mitigating demand target by whom and the manner in which alcoholic beverages can be consumed. Among others, these laws include age restrictions, public consumption, and DUI/DWI laws. What the laws have in common is that they put a restrictive burden on the alcohol buyer.  Supply mitigation. Laws aimed at mitigating supply target by whom and the manner in which alcoholic beverages can be sold. These laws include government ownership of retail or wholesale outlets, restrictions on outlet density, and restrictions on hours of operation. These laws and others like them place restrictive burdens on the alcohol seller.  Transaction mitigation. Laws aimed at mitigating transactions target the act of buying and selling. These laws principally take the form of taxes and, in more rare cases, price controls. While some laws may fall into more than one category (for example, depending on how they are enforced, keg registration laws could be considered demand mitigating or supply mitigating or both), the categories broadly reflect differences in policy efficacy noted in the literature. For example, in their meta studies, Carpenter and Dobkin (2010), Campbell et al. (2002), Grube and Nygaard (2001), and Her et al. (1999) find that studies report markedly different degrees of policy success depending on whether the policies are targeting alcohol demand, alcohol supply, or alcohol transactions. In addition to various policy variables, studies have looked at various policy targets including:  Sales/consumption mitigation. Studies that seek to focus on consumption typically take per-capita sales as a proxy. A reasonable counterargument to using this target is that alcohol consumption is not universally bad nor is a decline in alcohol consumption universally good.  Underage drinking mitigation. Underage drinking data typically comes from surveys and so the data are less reliable. A major source of data in the United States is the National Survey on Drug Use and Health which asks children ages 12 and over to self-report their alcohol use.  Underage binge drinking mitigation. As with underage drinking data, underage binge drinking data is also subject to self-report bias. Given the nature of the question (binge drinking), subjects’ responses are likely to be even less reliable than responses to underage drinking. Particularly where younger teenagers are concerned, there may be incentives to lie in either direction (for detection avoidance or self-aggrandizement). If underage drinking and underage binge drinking data are subject to bias, so long as the bias is consistent, studies that look at changes in these measures are likely to obtain results that are at least directionally correct.  Alcohol-related traffic accident mitigation. In the United States, alcohol-related traffic accidents are accidents in which at least one person involved in the accident, and who is not a vehicle passenger, has a blood alcohol content (BAC) above the statutory maximum. It is not necessary for an automobile driver to be the one with the high BAC. For example, if a car strikes a pedestrian and the pedestrian’s BAC is above the minimum, the accident will be classified as alcohol-related.  Alcohol-related traffic fatality mitigation. An alcohol-related traffic fatality is an alcohol-related traffic accident in which at least one person died.  Alcohol-involved traffic accident/fatality mitigation. An alcohol-involved traffic accident is an accident in which at least one automobile driver has a BAC above the statutory maximum.  Crime. Carpenter and Dobkin (2010) offer a review of studies examining the causal effects of alcohol consumption on crime. Studies that examine crime focus on violent and property crimes as opposed to DUI and public intoxication. The purpose of this paper is to summarize research on the effects of alcohol policy variables on alcohol policy targets, with a primary focus on privatizations. The goal is to determine whether there are trends in the research findings that can inform future policy makers. Outlet Density Controlling alcohol markets by controlling outlet density (the number of retail establishments per square mile) is closely related to controlling markets by monopolization. States that monopolize retail alcohol sales control outlet density by deciding where to place the state’s retail outlets. States with privatized (or partially privatized) retail markets employ density controls to restrict the density of private establishments. With respect to controlling retail prices, the types of beverages sold, hours of operation, and other non-location circumstances, monopolization provides the state with power that outlet density rules do not. However, if a state’s goal is merely to control outlet density, monopolization is not necessary – state licensing within a privatized retail market serves the same purpose. McCarthy (2003) performs a panel analysis of 111 non-metropolitan California cities over the period January 1981 through December 1989. He looks at changes in the densities (establishments per square-mile) of general off-site alcohol licenses (licenses to sell all types of alcohol for consumption elsewhere), general on-site alcohol licenses (licenses to sell all types of alcohol for consumption on the retailer’s premises), and off-site and on-site beer/wine licenses (licenses to sell only beer and wine), and compares these to changes in fatal, non-fatal, and total alcohol-related traffic accidents. McCarthy finds that an increase in the density of general off-site licenses is associated with decreases in fatal, non-fatal, and total alcohol-related traffic accidents, and that an increase in general on-site licenses is associated with increases in non-fatal, but not fatal, accidents. He also finds that, among beer/wine licenses, an increase in off-site license density is associated with a decline in total accidents, but that an increase in on-site density is associated with an increase in non-fatal accidents. Table 1. McCarthy (2003) results (density = outlets per square mile). Alcohol-Related Traffic Accidents Non-Fatal Fatal Total Increase in density of off-site general licenses Decrease Decrease Decrease Increase in density of on-site general licenses Increase No change No change Increase in density of off-site beer/wine licenses No change No change Decrease Increase in density of on-site beer/wine licenses Increase No change No change Stockwell et al. (2009) examined the effect of alcohol outlet density and the degree of privatization among retail alcohol stores on alcohol sales in British Columbia. British Columbia provides an interesting case study as the province has permitted a gradual increase in the number of private alcohol stores from 1988 to the present. Unlike McCarthy (2003), who defines density as number of outlets per square mile, Stockwell et al. define density as number of outlets per population. They look at 89 regions within British Columbia over the period April 2003 through March 2008 and claim to find that increased density and increased privatization is associated with increased per-capita alcohol sales. Their results, however, leave unaddressed the question of causality – is increased privatization causing increased sales of alcohol, or is increased demand for alcohol resulting in increased profit opportunities and therefore increased number of private outlets. Finally, the discussion of their statistical results leaves unaddressed potential technical errors that would, if present, render their estimates strongly biased in favor of their reported findings. Stockwell et al. do not discuss whether or not they tested for non-stationarity in their data. It is reasonable to assume that the time series data that they describe using would be non- stationary. Further, the results shown in their Table 4 (p. 1832) imply test statistics that are near- impossibly large for a correctly specified model (maximum = 83.5, average = 22.9). Table 2. Stockwell et al. (2009) results (density = outlets per population 15 and older). Per-Capita Alcohol Consumption Increase in density of beer outlets Increase Increase in density of wine outlets Increase Increase in density of spirits outlets Increase Increase in density of on-site beer/wine licenses Increase Weitzman et al. (2003) examine data on drinking habits among college students at eight public universities and compare the self-reported measures to retail outlet densities. Each of the 3,421 students who participated in the study self-reported to which of the classifications their drinking behaviors belonged: heavy drinking, frequent drinking, drinking-related problems, frequent drunkenness, non-binge drinking, binge drinking, drinks-to-get-drunk, and abstention. Weitzman et al. find positive correlations between retail outlet density and heavy drinking, frequent drinking, and drinking-related problems. As with related studies, the authors stress that their results are correlational and that neither causality nor the direction of causality is implied. Table 3. Weitzman et al. (2003) results (density = outlets per square mile). Correlation with Self-Reported Behaviors Heavy Drinking Frequent Drinking Drinking-Related Problems Outlet density Positive Positive Positive These studies as well as others (Presley et al., 2002; Douglas et al., 1997; Gruenewald et al., 1996) point to a possible relationship between retail outlet density and alcohol consumption. As the studies are not experimental, they leave two important issues unaddressed: (1) Is the positive correlation between outlet density and alcohol consumption causal? While repeated correlational studies might suggest causality, there remains the possibility that there is no causality present. For example, it is possible that college students tend to drink more as a consequence of age, new-found freedom, and propensity to take risks, and it is possible that the density of alcohol retail outlets is higher near universities simply because the density of people is higher near universities. (2) Assuming causality is present, what is the direction of the causality? While there is a natural tendency to blame markets for people’s behaviors, in fact, markets are merely the aggregation of people’s behaviors. In other words, markets do not cause behavior; behavior causes markets. If the correlation between outlet density and alcohol consumption were shown to be causal, it would be tempting to blame increased consumption on the increased availability of alcohol. However, an equally compelling (some may argue, more compelling) argument is that the density of retail outlets is caused by the propensity of the nearby populace to consume alcohol. Finally, if the relationship between density and consumption is causal, it is possible that the causality is bi-directional. It may be the case that both increased density causes increased consumption and that increased consumption contributes to increased density. From a policy perspective, the unanswered causality question is paramount. If the relationship between retail outlet density and alcohol consumption is not causal, or if it is causal but the causality runs from consumption to density or is bi-directional, restrictions on outlet density will have no effect on alcohol consumption. Worse, as is the case with all social policies, implementing the policy may lead people to falsely believe that the government is judiciously spending its resources in pursuit of a valuable social goal, and to erroneously equate spending and regulation directed toward the goal with the achievement of the goal. Privatization In an early review of literature on state monopolization of alcohol markets as a policy tool for reducing alcohol consumption, Holder (1993) looked at the use of state monopolization of alcohol markets as a means of combating alcohol consumption and, by extension, alcohol- related problems. Implicit in Holder’s review, and in many subsequent studies, is the assumption that alcohol consumption causes, rather than is caused by (or unrelated to), social ills. If, in fact, the causality is reversed or not present, we would expect that reducing alcohol consumption would have no effect on social ills. Holder concludes that research demonstrates that limitations on the availability of alcohol can reduce the consumption of alcohol and that this effect is most pronounced when alternative, unrestricted forms of alcohol do not exist. However, this result seems to be tautological in that it is not possible to consume what does not exist. MacDonald (1986) looked at privatization of wine sales in Idaho and Maine (both in 1971) and found that wine sales increased significantly following privatization, but that beer and spirits sales did not. In 1969, grocery stores in Washington were allowed to sell imported wines. Following this privatization, MacDonald detected an increase in wine sales – despite two mitigating factors: (1) grocery stores already sold domestic wines, and (2) grocery stores charged prices 25% higher than those in state-owned stores. As confirmed by later studies, MacDonald found that the wine privatization had no effect on beer and spirits sales. Virginia’s privatization of fortified wine sales in 1974 was not associated with an increase in wine, beer, or spirits sales. MacDonald suggests that this result may be due to the fact that fortified wine comprised a very small portion of the overall market for wine. Table 4. MacDonald (1986) results. Beer Sales Wine Sales Spirits Sales Privatization of retail wine stores (Idaho, Maine) No change Increase No change Privatization of retail wine stores (Washington) No change Increase No change Privatization of retail fortified wine stores (Virginia) No change No change No change Holder and Wagenaar (1990) found that, following Iowa’s privatization of liquor stores, sales of spirits rose significantly (9.5%), sales of wine fell significantly (13.7%), and sales of beer did not change. Wagenaar and Holder (1995) look at the privatization of wine sales in Alabama (1973 and 1980), Idaho (1971), Maine (1971), Montana (1979), and New Hampshire (1978) over the period 1968 through 1991. Employing a Box-Jenkins modeling technique to measure the relationship between privatization and alcohol sales, they find that each of the states experienced significant increases in wine sales following privatization. However, they found no significant change in beer and spirits sales following privatization. These results are consistent with their 1991 study in which they find similar results for privatization in Iowa (1985) and West Virginia (1981). Table 5. Holder and Wagenaar (1990) results. Beer Sales Wine Sales Spirits Sales Privatization of retail spirits stores (Iowa) No change Decrease Increase Table 6. Wagenaar and Holder (1991) results. Beer Sales Wine Sales Spirits Sales Total Sales Privatization of retail wine stores (Iowa, West No change Increase No change Increase Virginia) Table 7. Wagenaar and Holder (1995) results. Beer Sales Wine Sales Spirits Sales Privatization of retail wine stores (Alabama, Idaho, No change Increase No change Maine, Montana, New Hampshire) Conversely, Mulford, Ledolter, and Fitzgerald (1992), who also examined data before and after Iowa’s privatization, found that the privatization effect on wine sales was temporary (dropping to insignificance by two years after privatization), and found no evidence of an increase in spirits sales following privatization. Mulford et al.’s analysis differs from Wagenaar and Holder’s in several important respects. Mulford et al. have 29 more months of data following privatization. With the additional data, Mulford et al. are more likely than Wagenaar and Holder to detect the temporary nature of the privatization effect, if indeed the effect were temporary. Mulford et al. also express concern that Wagenaar and Holder’s model was misspecified in that Wagenaar and Holder’s model implicitly assumes that any change in baseline alcohol sales following privatization would be permanent. Thus, not only are Wagenaar and Holder’s data less able to detect temporary effects, but their model expressly assumes that privatization effects are permanent. Also, Wagenaar and Holder incorrectly include sales of wine coolers in their data. Because the privatization had no statutory effect on wine cooler distribution, sales of wine coolers should not be included in their measures of wine sales. Not only should wine cooler sales not have been included in the data, but, by coincidence, there was a surge in popularity of wine coolers that occurred around the time of Iowa’s privatization. This coincidence, combined with the erroneous inclusion of wine cooler sales, causes Wagenaar and Holder’s results to be biased toward showing a positive privatization effect. Table 8. Mulford, Ledolter, Fitzgerald (1992) results. Beer Sales Wine Sales Spirits Sales Temporary increase; Privatization of retail wine stores (Iowa) No change No change No change after two years Finally, it is worth noting that Wagenaar and Holder do not mention testing for non- stationarity – an anomaly that frequently plagues time series data. The presence of non- stationarity in a time series data set results in spurious parameter estimates – results that are biased toward significance. In their 1991 paper, Wagenaar and Holder report that they made their data stationary by using first differences, and, in their 1995 paper, do not mention stationarity at all. In 1991, they do not report the results for stationarity tests nor discuss performing post- regression tests for stationarity in the residuals. Were it not for their reported regression results, this would be less of an issue. However, their regression results (both in the 1991 and 1995 papers) exhibit extremely high multiple correlation coefficients and extremely large test statistics – both of which can indicate the presence of unaddressed non-stationarity. As discussed later, unaddressed non-stationarity frequently results in erroneous findings of significant relationships. Rehm and Gmel (2001) raise concerns about this problem, particularly as it relates to published research investigating alcohol use.