Methods of Measurement

The most common form of alcohol statistic used is unquestionably the annual per-capita consumption attributed to the drinking age population (typically interpreted as older than 14) of a nation or state. Derivation of the annual per-capita figure (technically a “period ratio”) requires enumerations or estimates of two quantities. First, the numerator: the total quantity of pure or “absolute” ethyl alcohol contained in beverages drunk by the population during the year; Second, the denominator: the average number of individuals meeting the specified age criteria living in the area during the year.

Estimates of the denominator are usually derived from census enumerations and extrapolations. These data generally exclude institutionalized populations and are known to disproportionately underestimate various segments of the population by unknown amounts. Inner-city residents, migrant laborers, and illegal aliens are most often cited as undercounted groups in the United States.

The numerator is usually compiled from figures supplied by producers or wholesalers of alcoholic beverages; specifically, the “tax paid withdrawal” of beverages from wholesale stocks available at the beginning or produced during the course of the year, along with their reported alcohol content. In the United States and most industrialized countries, the regulation of manufacturers by government agencies ensures a fairly uniform content of alcohol in respective classes of such beverages; but lapses have occurred, for example, in notorious French and Italian wine frauds. In the liberal and fairly inexpensive mass markets of recent years, it seems safe to assume that the quantity of consumed alcohol that has not been produced and counted under taxation and licensing control is probably small (Mäkelä 1978, Gavin-Jobsen Associates 1978). This includes illicit beverage production for sale or home consumption, laboratory and commercial ethanol diverted to beverage use, trace alcohol in soft drinks and other consumables, and drink produced legally but diverted from taxed inventories. The comparability of U.S. data in this regard with data from other countries is probably satisfactory for recent years in industrialized states.

Taxed withdrawal from local wholesalers' stocks, however, is not a direct measure of consumption by local residents. Consider the rank-ordering of states by per-capita sales in Table 1. The three locations with the greatest apparent consumption—Nevada, New Hampshire, and the District of Columbia—all have small local populations relative to large tourist trades. While all three would be very likely to rank in the upper end of the list without the tourism factor, it is probable that about half of their apparent per-capita consumption can be attributed to visitors not counted in the census of local residents. (This may also account to some degree for the high rate in Vermont due to its ski resorts.) In addition, alcoholic beverages are cheaper in New Hampshire and the District of Columbia than in surrounding areas and the border-crossing “liquor run” is a well-known local custom (Rooney and Butt 1978).

TABLE 1

Apparent Consumption of Absolute Alcohol, Population Aged 14 Years and Older.

As these considerations suggest, gross estimated per-capita consumption figures are subject to error, the more so as migration, transiency, or census errors affect the accuracy of the denominators (people count) and as untaxed production/distribution affects the numerators (gallon count). As a result, both random errors and systematic biases may enter the calculations. As a rule of thumb, in comparisons of a single jurisdiction across time where the systematic bias is not likely to change rapidly, year-to-year estimates using invariant procedures are likely to have confidence margins of 1–5 percent. In comparisons of different areas, error margins on the order of 5–50 percent may be assumed, depending on knowledge of such factors as recordkeeping efficiency, extent of migration and tourism, tax evasion, beverage quality control, and the like.

Uses of Per-Capita Consumption

At this point we an put aside the matter of accuracy of estimates and inquire what it is that per-capita period consumption averages should tell us about alcohol use in populations. First, we should observe that per-capita consumption is an arithmetic mean. Usually a mean is computed from data directly about individuals. For example, we measure the height of everyone in a group, sum the measurements, then divide by the number of individuals. Mean alcohol consumption using tax-paid withdrawals is arrived at quite differently. Instead of observing and measuring the consumption of individuals, we measure produced alcohol at one step in its distribution for sale, assume that all alcohol destined for drinking passes through this step, assume that little is thrown away, assume that retail and private stocks beyond this step do not increase or decrease much relative to the total production and flow of alcohol in a year, and, finally, assume that we have aimed our population count at all and only the people who might possibly drink the alcohol being measured. While none of these assumptions seems to unduly threaten the validity of derived means, except as noted above, all serve to indicate that per-capita consumption is an index that has been constructed to represent an implied set of behaviors.

What, then, do we know about the relation between this index and the set of consumption behaviors? Most of our knowledge about this derives from two sources: surveys of individual drinking practices and comparison of per-capita consumption with public health statistics about diseases strongly related to very heavy drinking. Before discussing this, however, a brief statistical note is in order.

We are accustomed to thinking about the mean, and assigning statistical meaning to it, in terms of the familiar bell-shaped normal distribution. If we think consumption is normally distributed, then the mean lies at the peak of the symmetrical bell-shaped curve that graphs the frequency distribution of individuals across alcohol consumption rates. In a normal distribution, more people consume the mean amount than any other amount (the mean is therefore also the mode); half drink more than the mean and half less (it is therefore equal to the median); and, depending on the degree of dispersion (indexed by the variance or standard deviation) of the curve, the majority of people consume an amount that is fairly close to the mean figure, with fewer people drinking at a given rate the farther it is from the mean.

The mathematics that underlie the bell-shaped curve lead us to expect that an approximate normal distribution will generally occur when there is a large number of independent causes, each of which contributes a small fraction of the total variance in a dependent variable. Thus, if individual rates of annual alcohol use were normally distributed, one would reason that the sufficient causes for single drinking events (person-drink occasions) were many in number and that variations in the occurrence of these causes were largely independent or uncorrelated with each other.

Investigation of the actual distribution of consumption rates has occurred largely in Europe and Canada, spurred by the work of Ledermann (1956). Ledermann wanted to develop a new basis for estimating the prevalence of harmful effects of alcohol use. He developed a rather bold, theoretically based hypothesis: that frequency distributions of consumption in a population are always logarithmic normal distributions (the normal frequency function, not of the consumption rate, but of its logarithm). The resulting curves are skewed to the right so that the median and mode lie below the mean (see Figure 1). Ledermann fit these data to various small samples of drinkers. Subsequent studies (deLint and Schmidt 1968, Mäkelä 1971, Skog 1971) have confirmed that lognormal curves may be roughly fit to a variety of survey data on consumption involving other, larger populations.

Figure 1

Exemplary logarithmic normal curves.

Bruun et al. (1975) went a step further, arguing that “differences as to dispersion between populations with similar levels of consumption are quite small” (p. 32, authors' emphasis). These authors go on to treat the dispersion across levels as practically invariant.

There are two important aspects to this argument. First, the lognormal function involves only two variables,² the mean and the standard deviation. If dispersion, and thus standard deviation, can be treated as invariant, then we have a “one-parameter” distribution: knowledge of the mean figure alone enables us to know the exact proportions below, within, or above any given consumption level or range. In order to calculate the rate of alcoholism, then, given the mean consumption, one need only define a consumption rate beyond which a diagnosis of alcoholism has a known likelihood: a “hazardous level of consumption” (Schmidt and Popham 1975–1976; Popham and Schmidt 1978). The term “hazardous consumption” can then stand as a quantitative proxy for the prevalence of clinical “alcoholism.”

Under the lognormal/one-parameter (also called the single-distribution) model, the relationship between mean consumption and the prevalence of hazardous consumption is well defined, and the mean, appropriately transformed, becomes an index for the rate of alcoholism. Since the transformation is exponential (parabolic), an increase (or decline) in mean consumption translates into a more pronounced increase (or decline) in alcoholism. Hence, changes in mean consumption may provide quite dramatic evidence in building a public health perspective.

I have stated that there are two important aspects to the argument. The second aspect has to do with interpretation of the mean not only as an index useful for estimating the size of a high-risk subgroup, but as an indicator of the alcohol-using behavior of the population as a whole. If dispersion is truly invariant, then any change in the value of the mean implies change in the number at all consumption levels—not only at the highest ones. The impression is given that a whole population changes its alcohol use together. This impression is reinforced by one of the mathematical assumptions that underlie the lognormal function; namely, that the many causal impulses involved in generating empirical lognormal distributions are not additive but multiplicative. In addition to these considerations is the latent habit of thinking that the mean, as central tendency, directly describes the approximate behavior of a large proportion of the population.

Actually this modal property accrues only if the mean value is close to the mode and if the dispersion of the curve is relatively small. In virtually every empirical consumption survey analyzed in the available literative (Skog 1971, 1979; Guttorp and Song 1977, 1979), one or the other of these conditions does not appear to be met. But this habit is strong, especially in the presence of limited data. Thus, Bruun et al. (1975) are led to say:

The reasons for this invariance are unknown; little research has been done on this subject [The apparent stability in dispersion'] and given our present state of knowledge no plausible explanation can yet be offered, other than that the level of consumption of each individual may be presumed to reflect his social milieu. One might well say, to paraphrase Euripides, “Tell me the company you keep, and I'll tell you how much you drink” (p. 34).

There are three difficulties with this way of thinking about consumption data. First, the invariance of the lognormal curve is not so apparent. Skog (1979), on whose analyses Bruun et al. relied for this finding (Skog being one of the alii), has subsequently argued that the gamma distribution is a better fit to most available survey data than the lognormal and that “there seems to exist an inverse relationship between per capita consumption and the dispersion of the distribution” (p. 3; but see Guttorp and Song 1977, 1979).

Second, adding to the uncertainty regarding the universal adequacy of the single-distribution model, if the consumption surveys that underlie the one-parameter model are used to estimate the mean statistics, their results diverge sharply from the figures derived by the production or sales/census method. Survey-based estimates of mean consumption have been found to be only one-third to one-half as large as those computed for the same populations by the production method (Houthakker and Taylor 1970, Mäkelä 1971, National Institute on Alcohol Abuse and Alcoholism [NIAAA] 1978).

Room (1971) reported that “our best survey questions uncover about two-thirds of the total expected consumption.” By way of comparison, Warner (1978) estimated that in tobacco-use surveys, people report smoking about 75 percent of the cigarettes that manufacturers report selling—a figure that was 90 percent a decade ago.

Third, the theoretical interpretation is not compelling. Let us suppose that the statistical evidence favoring the single-distribution model was strong and exact, rather than tentative and approximate. Still, this evidence is all correlational. Rather than arguing that the overall company of users spreads its influence widely among members, pushing them into (or pulling them back across the edge of) hazardous consumption, one could just as plausibly argue that the small fraction of relatively dedicated drinkers “sets the pace” for the rest. It does this not only by boisterous example, but also by supplying the bulk of the economic ballast for the alcoholic beverage industry. This price-conscious high consumption base provides a foundation for economies of scale in manufacture, for aggressive marketing, and for antitax lobbying—all of which keep prices down for high-end and low-end consumers alike. Thus, it is just as plausible to think that more heavy drinking leads to overall consumption increases, as the other way around. Without further investigation, which has not yet occurred, the mean index, even under strong statistical interpretation, does not yield definite causal accounts of the sort that have been attributed.

Fortunately, the mean statistic, even as derived from production/census reports, need not be supported by invariant distributional properties or strong theoretical interpretations in order to be of value. Even if the observed distributions can be fitted with a wide range of curves, the evidence is persuasive that these curves are all similar in shape: unimodal and skewed strongly to the right. If we were to reduce the overall U.S. consumption curve to a representative sample of 10 drinking-age adults, their annual consumption of absolute ethanol would not be very different from the following rough approximation: 3 nondrinkers, 3 drinking a gallon among them, and the others drinking 1.5, 3, 6, and 15 gallons, respectively (extrapolated from Cahalan et al. 1969). The total is 26.5 gallons or 2.65 gallons per capita, roughly the national figure for the past decade. The point to be noticed is that one drinker (10 percent of the population) consumes 57 percent of the total; two drinkers (20 percent) consume 78 percent of the total. If we examine changes in the mean statistic across time, we will largely be tracking the drinking behavior of this fraction of the population, the 20 percent, 10 percent, or less of heaviest drinkers—whatever changes may be occurring (or not occurring) among the remainder (see also Room 1978). As a statistical observation, we can simply note that given a skewed distribution, the mean is fairly responsive to the behavior of the long tail of the curve.³ This was the original point of Ledermann's (1956) investigations, and without further regard to specific aspects of the distribution function, it justifies the conclusion of Bruun et al. (1975, p. 45): “A substantial increase in mean consumption is very likely to be accompanied by an increased prevalence of heavy users.”

On the other hand, this by no means justifies such unqualified attributions as “a fall in the average level of consumption … will lead to a fall in the number of heavy drinkers” (Hetzel 1978, p. 84). It is at least as plausible to say that the reverse is true, as is implied by Skog's theoretical demonstration of the long-term effect of a large shift in consumption preference by a small fraction of the members of a hypothetical social drinking network (1980).

To sum up the methodological implications thus far: the mean alcohol consumption statistic is a reasonable gross index of the amount of potentially hazardous consumption in a society. It is most valuable for this purpose when used as a time series or longitudinal index. To see this in operation, see Figure 2, which gives a 150-year record of this index in the United States.

Figure 2

Time trend of U.S. consumption of absolute alcohol, per-capita drinking age population, 1830–1977. Note: Data points are less dense and less reliable prior to 1900. All estimates adjusted to reflect population 15 years and older.

The long trend of per-capita consumption in the United States can be divided into four distinct periods: pre-1850, 1850–1914, 1915–1945, and 1946 to the present. Prior to 1850 the trend was downward from the high level (6–7 gallons annually per adult) that characterized colonial and revolutionary America. From 1850 to 1914, the level hovered in the area of 2 gallons, but began to rise appreciably at the turn of the century, reaching its highest level just prior to U.S. entry into World War I. During the third period (the time of the world wars), state and federal prohibition laws reduced per-capita consumption to about half or slightly more of the pre-World War I level, from which it recovered to about two-thirds that level by the end of World War II. After 1945 the rate hovered about the 2-gallon level for some 15 years, then rose during the 1960s back to the pre-World War I level of roughly 2.7 gallons per capita.

For comparative purposes, Table 2 supplies a two-point time series for the postwar era for a number of industrial (mostly European) countries. As indicated above, caution is necessary regarding the accuracy of comparison of these consumption statistics between different countries, but there are two indisputable general observations to be made. First, the United States is unremarkable in its total consumption during this period relative to other similar national societies. It is neither especially dry nor especially wet. Moreover, the postwar increase in U.S. consumption is also unremarkable. With little exception, increases on this order or greater have been common throughout the industrialized world and, for that matter, virtually everywhere else for which statistics have been compiled (Sulkunen 1976, Moser 1979). In both the static and comparative perspectives, the United States is in the middle of the pack formed by the array of mean consumption figures.

TABLE 2

Changes in Apparent Consumption of Absolute Alcohol in 20 Countries in Gallons Per Capita of the Population Aged 15 Years and Over.

Disaggregated Consumption Statistics

I have devoted considerable attention to the average absolute alcohol use measure, first because it has proven to be of strategic importance for one major set of alcohol-related problems (liver disease), and second because its use in Alcohol Control Policies in Public Health Perspective (Brunn et al. 1975) and related literature has generated so much interest and controversy, both scientific and political.

There are, however, compelling arguments for dividing total consumption statistics into finer grain. Currently, four such disaggregating strategies have been employed, singly and/or in combination, in the literature. The most easily accessible of these is the division of aggregate statistics by type of beverage. Pekka Sulkunen (1976) has forcefully argued for the consideration of the “use values” involved in alcohol consumption. At the most basic level, he identifies three sets of values: intoxication, nutrition, and conviviality. Very roughly, these tend to be differentiated in parallel with the three main types of Western alcoholic beverages: distilled spirits are used as an intoxicating drug, wine as the liquid part of a meal, and beer as an accompaniment to sociable relaxation. These usages are only rough guides, of course, and are historically associated with particular ethnic-national practices. Sulkunen points out that in the postwar period, there has been a mild trend toward “homogenization” of drinking cultures worldwide, with one region's drinking preferences and practices being imported and superimposed on the practices native to other regions. However, traditional wine drinkers, for example, in adopting the use of distilled spirits, tend to use them in ways more consonant with their traditional use of wine; e.g., to use distilled spirits in the form of liquors and brandies not far removed from mealtimes.

If we follow Sulkunen's arguments and, as in Table 3, identify countries by their predominant (according to absolute alcohol volume) beverage, we see that the highest-ranked countries (by mean per-capita total alcohol) are wine-drinking, the lowest-ranked consume mainly spirits, and the heavy beer-drinking countries are mainly found in between. There is also, among the European countries, a regional concentration of wine countries in the southwest, where the grape vine grows plentifully. The dispersion and range down the table are greatest for wine: a number of countries are separated by geometric ratios in excess of 25 and as high as 50, much more than the differences for beer (10x) and spirits (3x). We do know that ratios of consumption between beer and spirits have been known to shift abruptly in European countries, such as in Denmark, when in 1917 the beverage taxation rates were quite drastically modified (Wilkinson 1970).

TABLE 3

Apparent Consumption of Absolute Alcohol in Gallons Per Capita of the Population Aged 15 Years and Older, in 25 Countries, by Beverage Type.

Regional distributions are also evident in the United States (Table 4), although they are much milder. Wine varies by about 5x, spirits by 2x, and beer by no more than 1.5x. Although the limited comparability of regional indices is, as indicated above, an inherent caution against generalization, it seems very safe to say that the states are, as a whole, quite uniform in their prevalence of heavy beer drinking and less so in regard to spirits. (Note the high rate in New England versus the low rate in South-Central.) The greatest regional difference lies between the wine preferences of the Pacific states in contrast to the Midwest—although these wine-growing coastal states unquestionably lie well below the prevalence of heavy consumption found in the European wine countries.

TABLE 4

Apparent Consumption of Absolute Alcohol, in U.S. Gallons Per Capita of the Population Aged 14 and Older and the Percentage Contribution of Each Class of Beverage to the Total, in U.S. Geographical Regions.

We may also examine changes in the consumption of different beverages across time. A 150-year trend for the three main classes of beverage types in the United States is displayed in Figure 3. The overall configurations are quite different. (Note that these graphs are scaled in gallons of beverage rather than gallons of absolute alcohol.)

Figure 3

Time trend of U.S. consumption of types of alcoholic beverages, 1830–1977. Note: The solid lines for the years 1920–1935 are consistent with Warburton's (1932) estimates, while the dotted lines represent Jellinek's (1947–1948) (more...)

Wine remained at fairly low stable levels throughout the 19th century and into the 20th century. The World War I mobilization, along with state (then national) prohibitions, succeeded in suppressing the apparent mini-boom about 1910, but home winemaking became popular during the 1920s. Heavy wine drinking recovered rapidly from the Great Depression, and average wine consumption was the most vigorous beverage index during the consumption boom of the late 1960s.

Beer, by contrast, only began to be drunk in the United States about 1850, but its heavy use grew remarkably (the index increasing by 10-fold) in the next 65 years. The drastic effects of Prohibition were not shaken off for a full generation, and the mean consumption of beer today is less than it was in 1915. Most notably, the mode of drinking beer in the prewar period—in draught from tavern barrels—has been very largely replaced by canned and bottled beer (Rooney and Butt 1978).

Finally, consumption of heavy spirits dropped steadily during the 19th century, finally leveling off about 1880. The wartime state prohibitions and the Volstead Act depressed this index for a few years, but it rebounded during the mid-1920s to about 30 percent higher than its Victorian level. The Great Depression appears to have strongly affected the index, although the switch in data base from indirect measures of illicit production (1930) to tax records of legal distillation (1934) exaggerate this. The index was somewhat below the 1880–1915 level after World War II, then rose by 1970 to its recent plateau—still below the levels of heavy spirits drinking estimated during Prohibition.

In brief we find that consumption trends for the three alcoholic beverage types diverged widely during the 19th century and through about 1935. Wine use (nutritional) was stable (until a growing popularity after 1900), intoxication (spirits) fell to 25 percent or less of Jacksonian levels, and heavy beer drinking in the convivial atmosphere of the tavern grew steadily until Prohibition effectively ended it. Since 1935, which is to say since the reinstatement of federal control over production and state control over distribution, the consumption trends of all three types have been quite similar. The largest difference is in the more rapid growth of wine drinking, which now contributes about 13 percent of all absolute alcohol consumed in the United States.

Drinking Patterns in Survey Data

The total consumption approach draws on sales data for its mean statistic and on sample surveys of drinking for its single-distribution argument. A rather different use of sample survey data is made by those who are interested in the frequency and distribution of intoxication, what has come to be called “drinking practices” or “patterns.”

Although the study of drinking patterns has been carried on largely by questionnaire survey research, a few smaller studies have used “daily drinking diaries,” which have recently been compared with survey methodology and have demonstrated relatively minor deviations in summary results (Gerstel et al. 1975, Harford 1979). Since large-scale survey research is basically a post-World War II phenomenon, detailed quantitative data about drinking patterns are virtually confined to this period.

One approach that has been widely assimilated is to divide drinkers into categories such as abstainer, infrequent, light, moderate, and heavy drinkers, taking into account the frequency of drinking occasions, the usual quantity consumed, and the frequency and quantity consumed during maximum drinking episodes. This is the quantity-frequency-variability or Q-F-V approach (Mulford 1964, Knupfer 1966, Cahalan et al. 1969, Cahalan and Room 1974). Mapped against the skewed distribution of annual per-capita alcohol consumption, each of these discrete subtypes of drinker would generate its own characteristic curve, “underneath” the aggregate total consumption curve, which the summation of the five would yield. Many of the surveyed individuals whose drinking pattern might qualify them as light or moderate drinkers (small amounts every day; seldom noticeably intoxicated) have a total annual consumption greater than individuals whose pattern classifies them as heavy drinkers (twice a month “blind drunk”). The typological approach identified here is especially sensitive to such occasions of extreme drunkenness and thus to “binge drinking,” a type that has haunted the alcohol literature, both empirically and conceptually, since Jellinek (1960) originally formulated this as a distinct type. The typological method clearly assumes that particular consumption characteristics, beyond period consumption per se, dramatically escalate any risks associated with alcohol.

According to successive surveys, the proportion of reported abstainers (no alcohol drunk in the past year) among U.S. adults was about 45 percent in the middle and late 1950s, but declined to about one-third by 1970; the figure has been unchanged since that time. However, there is considerable difference between adult men and women. About 27 percent of men, but 42 percent of women, are now abstainers.

It is conventional to distinguish heavy from moderate drinkers. The criteria for drawing this line vary from study to study. In the major national survey carried out in the 1960s, the designation “heavy” was used for a conglomerate of patterns; it included those who had five to six drinks twice a month, a drink or two nearly every day, five to six drinks “once in a while,” and a drink three times a day; and it encompassed patterns combining or exceeding these frequencies. By these lights, 1 in 5 men, and 1 in 10 women were heavy drinkers in 1965. Annual surveys initiated in the 1970s (see Johnson et al. 1977) using a simpler definition of “heavier drinking” (an average of 2 or more drinks per day) yielded virtually identical proportions. These figures too have shown no significant change since 1970.

In a comparative perspective, the United States is remarkable among Western industrial nations for its proportion of adult abstainers, which is higher in the United States than in Canada or in any nation in western Europe (these countries range from 3 percent to 30 percent abstainers; [NIAAA 1978]). There is considerable regional variation in the United States, however. The proportion is lowest in the Mid-Atlantic states (17 percent) and highest in the Bible Belt states of the South (65 percent). It is not possible to make broad international comparisons for proportions of “heavy drinkers” using such survey data due to incompatible or nonexistent data for the different countries.

It is important to note that even where drinking patterns for a population appear to be stable across a number of years, there is continual flux in the drinking patterns of many individuals within the population.

At the extreme, many clinically diagnosed alcoholics alternate between periods of heavy consumption and periods of abstention. There is also a tendency for drinking to decline markedly beyond the age of 50, and there are often significant differences in drinking rates between age and sex cohorts as well as within the same cohort at different points in time. Skog (1979) has argued that these different drinking subpopulations are the primary explanation for the departure of consumption curves from Ledermann's predictions.

Use and Consequences: The Problem of Attribution

It is one thing to know how much or how often people drink; it is another to pinpoint the effects of that drinking. For virtually every possible consequence of alcohol use that we may be interested in, alcohol is neither a necessary nor a sufficient cause, but rather one in a series of factors that may combine in various ways to yield effects. The presence of specifiable other factors, some permanent but others subject to manipulation, is just as important as drinking per se to the production of alcohol-related effects.

For example, sustained heavy alcohol use is demonstrably associated with the pathological and potentially fatal liver condition called cirrhosis. The data on this connection are as strong as those linking heavy smoking with lung cancer. But the mechanism by which alcohol use causes this gradual accumulation of scar tissue in the liver is still speculative. It is known that cirrhosis can occur in people who have never drunk alcohol. It is known that various nutritional deficiencies or imbalances can not only cause cirrhosis directly, but also can markedly change the vulnerability of liver tissue to alcohol-related cirrhosis. Good nutrition probably cannot prevent alcoholic cirrhosis; but bad nutrition can certainly hasten it along.

Another example is automobile accidents. Drunken driving is notoriously a precursor of traffic fatalities. But at least half of all traffic fatalities in the United States evidence no alcohol “involvement” at all. Perhaps 1 in 500 to 2,000 of legally drunk-driving episodes result in arrest—and there are far more arrests than fatalities. Moreover, an individual can hardly cause a traffic fatality, no matter how drunk, in the absence of a lethal vehicle. This may seem trivial in a nation of 100 million automobiles, but it is not trivial if changes in gasoline costs drastically depress people's use of cars or change the way autos are built and driven. In short, certain of the socially most important consequences of alcohol use are rare, may occur with no alcohol present, and/or require the presence of additional factors in order to happen at all.

These two examples involved the negative effects of alcohol on physical well-being. In looking at psychological and social well-being, the difficulties do not decrease. A number of studies have found some common properties among the personalities of diagnosed alcoholics: depression, anxiety, low frustration tolerance, feelings of powerlessness, etc. But virtually the same properties have been deemed to cause alcoholism. If depression is both a cause and an effect of heavy drinking, how can one decide what part of the alcoholic's depression, or of suicides that follow from depressive episodes, is due to drinking?

Similarly, attempting to pinpoint the exact part played by drinking in problematic social behavior is exceedingly difficult, compounded as it is by interaction with others during numerous situations over time. Statistical methods to separate alcohol consumption from other causes do offer some help in disentangling multiple causes, but since the problems are generally long-term ones, in which both the drinker and “significant others” build up a history of cumulative perceptions and judgments, such attempts to isolate and assign weight to one factor may have little relationship to reality.

All of these problems in attribution apply to the positive effects of alcohol use as well. They are somewhat aggravated by the disinclination of researchers to investigate positive effects, which means that less data are available for analysis. For example, the best-known study of economic effects of alcohol is on the costs of alcohol abuse and alcoholism and does not attempt to estimate effects that may yield economic positives. Mortality studies ask how many deaths alcohol may have been at least partly responsible for and do not concern themselves with the lives alcohol may preserve, even though studies on heart disease have indicated certain advantages possibly associated with moderate drinking.

There is no avoiding the difficulties in attributing the effects of alcohol use; however, it is necessary to have some scheme of accounts for these effects if we are to consider alcohol-related policy. If a central principle of any policy is to preserve and promote good effects while minimizing bad ones, then it is difficult to evaluate any policy without identifying the effects of relevance and observing how they change over time. Since the consequences of alcohol use are manifold, it is important in the first place to identify the ones of greatest social importance, to assess the relative importance of alcohol in generating them as best we can, and then to be systematic and comprehensive in thinking about how these effects can be modified.

A more exact feel for the difficulties of attribution may be achieved by examining recent responsible attempts to estimate the effects of drinking in the United States. Three of these were prepared in the 1970s under the auspices of the National Institute on Alcohol Abuse and Alcoholism (NIAAA) and were prominently featured in the Third Special Report to the U.S. Congress on Alcohol and Health. Each examines a different dimension or measure of effects: mortality (deaths), economic costs (dollars), and psychobehavioral problems (symptoms).

Mortality

Table 5, reproduced in Alcohol and Health 3 (NIAAA 1978), estimates that the number of deaths related to alcohol use in the United States in 1975 was between 61,000 and 95,000. The associated text notes that other mortality studies have respectively estimated 140,000; 185,690; and “as high as 205,000 deaths per year, which was 11 percent of the total 1.9 million deaths in 1975.” The largest figure has received the widest publicity.

TABLE 5

Estimated Deaths Related to Alcohol in the United States, 1975.

A comprehensive study of deaths among clinical alcoholics (Schmidt and Popham 1980, also see Polich et al. 1980) suggests how the totals in Table 5 could have been inflated. For 12,000 former alcoholism patients in Ontario, Canada, who were followed up at an average of 8.5 years, 1,062 excess deaths (compared with age-standardized population norms) were attributed to the causes listed in Table 6.

TABLE 6

Percentage of Excess Deaths in a Clinical Alcoholic Population in Canada.

The authors of this study caution that alcoholics smoke cigarettes at considerably higher rates than the comparison population and, hence, that excess heart disease and cancer deaths are associated to an unclear degree with tobacco rather than alcohol exposure. They also warn that nutritional and other life-style differences intervene in the relation between drinking alcohol and dying.

Still, 40 percent of the deaths in Table 5 were from causes not directly considered in Table 6; hence, an absolute maximum figure in the range of 150,000 for deaths “related to alcohol” might be acceptable, presuming that all untreated alcoholics have the same death risks as clinical patients. But what does this mean? If we wish to determine how alcohol affects the overall death rate in the United States, i.e., to discover its net causal impact, we would have to estimate both deaths resulting from and deaths prevented by alcohol use. (There is some evidence, for example, that moderate drinking is correlated with decreased risk of death, particularly death involving ischemic heart disease.) However, being related to or correlated with does not amount to causal proof. In this sense, the 150,000 figure is simply a pool within which whatever deaths might be caused by alcohol—and thus prevented by minimizing its use or misuse—are to be found. The number will be no more—it will be much less. How much less is not certain, but a figure in the area of 50,000 theoretically preventable deaths seems reasonable. This is not 200,000—but neither is it negligible.

Economic Costs

Table 7 is a tabulation of how much alcohol abuse and alcoholism cost the United States in 1975. This $42.75 billion figure is not a “net cost” estimate; that is, it is not intended to represent the net reduction in gross national product resulting from alcohol use. It is a conglomeration of dollar values: discounted lifetime earnings attributable to about 69,000 people whose deaths in 1975 were estimated to be alcohol-related (25 percent of the $42.75 billion); lessened production as estimated by certain household wage differentials (35 percent); assignment of health care resources to alcohol-related problems (21 percent); and assorted other estimates. This collection of dollar values is held by its authors (Berry et al. 1977, Berry and Boland 1977) to be the “external cost” of alcohol abuse—paid not by the alcohol abuser but by the rest of society.

TABLE 7

Economic Costs of Alcohol Misuse and Alcoholism in the United States, 1975.

It is difficult to assign precise significance to this result. For one thing, a large fraction of the costs tabulated here do appear to be borne by the alcohol abuser: from 25 percent to 50 percent of the total amount cited. There are implicit assumptions about the labor supply in 1975—namely, that it was short—that are not consistent with the then-prevailing economic indicators (unemployment was at 8.5 percent). The transposition of “associated with alcohol” into “due to alcohol abuse,” which is to say transformation of an estimated correlation into a direct cause, occurs in the discussion despite explicit cautions by the authors to the contrary.

There are some problems raised by this approach of estimating costs in absolute, autonomous terms. Economic cost estimates are usually related either to a specific program of action or to a specific annual budget. In the first instance, one might estimate the costs and benefits of a specific program to modify alcohol abuse. In the second instance, one would try to estimate the proportion of, for example, the federal budget or aggregate personal income, or capital stock, expended on, diverted by, or destroyed as a result of alcohol-related problems (Walsh 1979, Mäkelä and Österberg 1979). Absent from such contexts, a multibillion dollar cost figure is simply a large price tag on an empty box. It draws attention, but gives little guidance.

Behavior

In Table 8, approximately 1 in 10 drinking American adults (about 10 million people) are classified as “problem drinkers”; an additional 1 in 4 (25 million) as “potential problem drinkers”; the other 65 million or so drinkers reportedly had “no problems.” What precisely do these rather ominous figures mean? In the original table (Johnson et al. 1977), the authors listed these categories as “frequent problem drinking symptoms,” “potential symptoms,” and “no symptoms.” The most common of these symptoms (“taking 2 or 3 drinks at one sitting”—“sometimes but not often” or “frequently”; “going several days or weeks without taking a drink and then having several drinks at one time”—“sometimes but not often” or “frequently”) are indistinguishable from criteria for moderate drinking patterns. The next most common symptoms (“talking a lot about drinking”—“sometimes but not often” or “frequently”; “showing the effects of liquor more quickly than most people”—“sometimes but not often” or “frequently”; “taking a drink to feel better”—“sometimes but not often” or “frequently”) are not, on their face, “problems.” Sixty percent of the reported symptoms fall into one of the categories above. The authors warn that “this problem index is presented for comparison purposes over time and should not be used as an absolute definition of problem drinkers.” As an index it simply demonstrates that there has been relative stability between 1973 and 1975 in reports of these drinking patterns.

TABLE 8

Rates of Problem Drinking Among U.S. Drinkers, by Drinking Population, 1973–1975.

These data really shed little light on the relationship between alcohol use and consequences. A few categories of symptoms (not the most common ones) are markedly more prevalent among heavier drinkers than other drinkers. But this report's conglomeration and renaming of categories obscures more than it enlightens. It is impossible to attach meaning to Table 8 beyond the indication of stability across time.

Socially Important Effects of Alcohol Use

It is clear that there are problems with efforts to assay the effects of alcohol by toting up detached dollar values, “associated” deaths, and “symptoms.” There are a number of alternative ways to present and organize information about these effects. No single way is right for all purposes, but my concern here is to highlight the socially important effects, to examine the degree to which alcohol is a principal cause, to roughly estimate the population affected, and to note what other generative factors besides alcohol use might be important and subject to modification.

Taking this view, I can identify five principal environments that, when combined with alcohol, produce effects. Each is best visualized as a system. First is the internal organic environment of the body, in W. B. Cannon's (1932) sense: its basic, enduring physiological forms and processes. Our principal interests here are the organs most sensitive in the long term to alcohol exposure: the heart and the liver.

The second environment is the personality. It is also internal to the person and one can even speak of it as based in an organ: the brain. But the physiological base does not suffice to complete our thinking about effects on the personality system. We need to use psychological concepts that we would not think of applying to hearts or livers.

The third system is external and the one that we so often think of as “the” environment: i.e., the physical one. It is of great importance that the physical environment we touch most directly is heavily transformed by human activity. The world of moving vehicles, building elevations, and concrete and floating surfaces is no less physical for being fabricated.

The fourth system, external like the last, is the system of intimate social relations, most particularly those we traditionally regard as primary and important to us: kin and coworkers or, to stress the long view, family and occupational careers.

Finally, there is the institutional environment, the relatively impersonal social matrix of secondary relations that we can ordinarily relate to only in abstract or remote terms, although it diffusely and sometimes very directly transfigures our lives. Most particularly, we are interested here in the broad public systems of responsibility for health care, civil safety, and economic well-being.

Physiological Effects

When the possibilities of sustained changes in physiological systems due to alcohol use are examined, very different effects come into view depending on whether the depth of intoxication, drinking patterns, or total consumption is considered. Due to the surprising capacity of the liver to rapidly convert alcohol to usable carbohydrate, the physiological effects of intoxication tend to be short-lived. The principal exception is intoxication that leads quite immediately to death. In examining drinking patterns, on the other hand, the primary effect involves physical dependence on alcohol, which induces vulnerability to the pathological syndrome of withdrawal or abstinence. The medical recognition and management of this syndrome, however, has relegated it to a fairly insignificant role in terms of physiological health—in contrast to the effects (which physiological alcohol addiction may accentuate or support) of drinking patterns on psychological and interpersonal matters. Finally, clear links between drinking patterns and physiological changes have been investigated most thoroughly in relation to total consumption. In this connection, rates of coronary heart disease and cirrhosis of the liver are by far the most prominent physiological aspects known at this time.

Deaths from alcohol overdoses, alone and in combination with other drugs, account for close to 10,000 deaths and perhaps 100,000 episodes of medical intervention each year. The finding of BAC in excess of 0.3 in autopsy toxicological findings is generally considered to strongly suggest the possibility that the death was the consequence of an acute alcohol reaction. Approximately 5,000 such deaths occur each year (Day 1977). In addition, approximately 5,000 fatal overdoses involving alcohol in combination with other drugs occur in the country each year, based on extrapolation from Drug Abuse Warning Network (DAWN) data (IMS America 1976). Since these reports indicate only measurable presence of blood alcohol and not specific BAC and do not speculate about relative mechanisms, it is difficult to know in how many of these deaths alcohol was a sufficient or necessary cause. The incidence of suicide within this group is not known, although in the alcohol-in-combination cases the death certificates estimate suicidal intent in about 40 percent of the cases.

Cirrhosis of the liver is a disease process known to have multiple causes, that are not all well understood (French 1971). Certain protein deficiencies are invariably followed by appearance of cirrhosis—the dissolution of liver cells and their replacement by scar tissue. Cytological studies on the effects of alcohol use have shown that there is a buildup of fatty yellow liver cells as a consequence of metabolizing alcohol. In laboratory studies of short-term high-level exposure to alcohol, mitochondria (essential organelles within liver cells) disintegrate. Both of these processes may contribute to liver cell death and fibrosis (cirrhosis). The supply of certain amino acids in the diet also appears to have a strong influence on how well liver cells resist deterioration; but this protective effect is by no means absolute (Lieber et al. 1971).

These microbiological findings are strongly supported by macroepidemiological studies (Schmidt 1977), which show that gross rates of cirrhosis deaths track shifting rates of total consumption in a population. Among clinical alcoholic populations in which consumption of 5 ounces of alcohol daily for long stretches of time is an approximate lower limit of alcohol use, a prevalence of cirrhosis damage of 8 percent has been reported, far in excess of the general population; another 25 percent suffer acute liver inflammation, generally regarded as a precursor to cirrhosis. In postmortem studies of individuals identified as alcoholics in New York City (Haberman and Baden 1978), the incidence of moderate-to-severe fatty changes in the liver exceeded the rate of cirrhosis by approximately the same proportions. In general, it is expected that a long period of exposure to high levels of consumption, on the order of 15 years or more, is necessary to bring the disease process to a life-threatening state; however, cessation of alcohol use can freeze the process.

The rate of liver cirrhosis death in the United States at several points, beginning in 1961, is reported in Table 9. It has been estimated that if no alcohol were consumed, the death rate from cirrhosis would approximate 3–4 per 100,000 (see Schmidt 1976, 1977; Skog 1979). The rise from 11.3 to 15.4 between 1961 and 1971 coincides with the 26-percent increase in per-capita U.S. alcohol consumption in this period.

TABLE 9

Statistics Relevant to Alcohol-Related Mortality in the United States, Death Rates per 100,000 by Year and Cause of Death.

Since 1971, however, alcohol use has held steady while the cirrhosis death rate has declined to 13.6. The reason for this decline is simply not known. It is expected that cirrhosis death rate changes lag somewhat behind changes in total consumption, so we would have actually expected the increase to continue somewhat into the 1970s. However, general death rates in the United States have been declining steadily during the 1970s, and the decrease in cirrhosis death is in approximate line with this general decrease.

In the 35–54 age band, cirrhosis accounted for 6.5 percent of all deaths in 1979. Of the 1.93 million deaths in the United States in 1979, 30,000 were due to cirrhosis. At best guess, between 20,000 and 25,000 of these were a primary result of alcohol consumption. Studies in Ontario, Canada, which has a total consumption similar to the United States as a whole, lead to the estimate that roughly one-half of this last figure involved individuals whose total consumption is above the lower limit observed in clinical alcoholic populations (Schmidt 1977). We would therefore estimate that about 10,000–12,000 deaths from cirrhosis occur among clinical alcoholics in the United States, and another 10,000–12,000 among people who would most likely not meet clinical criteria for chronic alcoholism.

Alcohol use has also been investigated in relation to heart disease. The findings here are not clear-cut (see Klatsky et al. 1978, Hennekens et al. 1978). Very high consumption of alcohol has been linked to cardiomyopathy and hypertension; however, moderate consumption has been linked in a number of prospective and case control studies to reductions in ischemic heart disease, the major cause of death in the United States, claiming about 525,000 lives in 1979. Relative to nondrinkers, these studies indicate risk reduction of 20 percent to as much as 70 percent. However, the samples are specialized and the studies are inconsistent in regard to whether or when increases in consumption beyond two drinks per day cease being “protective.”

From 1961 to 1971, while alcohol use on the whole increased 26 percent, the rate of death from ischemic increased 18 percent; but this change was confined to cohorts born before 1910—there was virtually no change in rates of coronary death for age brackets below 65. Since 1971, ischemic heart death has been declining dramatically in all age categories. This could be a lagged effect of earlier consumption changes, a result of improved cardiological care or general health, or the result of other causes entirely.

In summary, the major significant physiological effects of alcohol in the United States today are overdose, liver disease, and heart disease. Alcohol is involved in roughly 10,000 overdose deaths annually, half in combination with other drugs. It is difficult to place any boundaries around the precise population at risk for all of these overdoses, although it has been shown that in the combination deaths the sexes are evenly split and most were people under 30, whereas alcohol overdose deaths follow the general demographic profile of clinical alcoholic populations: mostly 30- to 55-year-old men. Alcohol is also involved in 20,000–25,000 cirrhosis deaths annually, of which about half occur in populations that probably meet appropriate criteria for clinical alcoholism. There is reason to believe that alcohol use may have a significant effect on coronary disease. If alcohol use were responsible for a 5-percent net decrease in coronary mortality, this would be 33,000 lives saved, and concomitant decreases in nonfatal heart attacks and other disabilities. The population of reduced risk in this instance would extend to 55 or 60 percent of the adult population of the United States.