key national alcohol-related surveys. These vary widely with respect to sample size and content, and each makes a unique contribution. The larger ones provide the most precise prevalence estimates for alcohol use disorders. Large samples are also essential for estimating age-, sex-, and ethnicity-specific rates of alcohol abuse and dependence because these disorders are relatively rare in some subgroups. Smaller surveys provide more detailed information by oversampling groups of interest, such as minorities or women, or by expanding their focus to include a broad range of factors thought to influence alcohol-related behavior, such as family history of drinking, social pressure to reduce drinking, and interpersonal relationships.

Alcohol Consumption

Alcohol use is measured at the population level in terms of per capita consumption (Williams et al. 1991). The calculation is based on the total amount of alcohol consumed in the United States, which is estimated on the basis of alcohol sales in each State as determined from tax receipts, sales in State-controlled stores, and reports from beverage industry sources. These overall statistics do not include estimates of home production, illegal production, breakage, or untaxed alcohol brought into the country by tourists; nor do they account for alcohol that is bought and stored rather than consumed. Per capita consumption statistics, then, represent estimates of consumption rather than actual consumption.

"Apparent per capita consumption" is determined by dividing total quantity of alcohol, derived from sales, by the total population aged 14 years or older (Williams et al. 1991). Thus, these estimates attribute average consumption to all persons in this population, regardless of their actual consumption. Per capita consumption is expressed in gallons of pure alcohol, calculated by multiplying total gallons of each beverage type by a conversion factor (0.045 for beer, 0.129 for wine, and 0.414 for spirits) that represents the average alcohol content of each beverage, then summing over all three beverages (Williams et al. 1991).

Although apparent per capita consumption provides an overall estimate of alcohol consumed in the United States, survey data are needed to link consumption data with sociodemographic descriptors of drinking at the individual level. Although other methods have

been proposed to obtain self-reports of alcohol intake (Room 1990, 1991), most U.S. surveys have included quantity-frequency questions asking respondents how often they consume alcohol and how many drinks they usually have. Data are also obtained concerning occasions of heavy drinking, for example, by asking how often individuals have 5 or more drinks at a time, or how often they have 5 to 8 drinks, 9 to 11, or 12 or more at a time.

Although apparent per capita consumption provides an overall estimate of alcohol consumed in the United States, survey data are needed to link consumption data with sociodemographic descriptors of drinking at the individual level.

A limitation of survey data is that people may not report their alcohol intake accurately. Evidence that respondents underestimate their consumption has been provided by studies comparing the amount of alcohol sold in a given region with the amount estimated to have been consumed; self-reported alcohol consumption accounted for only 40 to 60 percent of the alcohol sold (for reviews, see Midanik 1982, 1988). If all drinkers underestimated their intake to the same extent, correlational studies of drinking behavior would still be valid because drinkers would be accurately ranked according to their consumption. However, investigation of this point has yielded inconclusive results. Some researchers have reported fairly good agreement between self-reports and independent assessments of alcohol intake, whereas others found that heavier drinkers and drinkers with problems underestimated their intakes more than light drinkers (Midanik 1988). Despite lingering concerns, alcohol survey researchers have continued to rely on self-report data. While the lack of practical alternatives forces continued reliance on such data, researchers have also been encouraged by studies demonstrating that the validity of self-report data can be maximized by employing good assessment techniques (Babor et al. 1987; Room 1990, 1991; Skinner 1984).

In 1989, apparent per capita consumption of alcohol in the United States was 2.43 gallons of pure alcohol, the lowest level since 1967 (Williams et al. 1991). Translated to more

immediately meaningful data, 2.43 gallons of pure alcohol represents approximately 576 12ounce cans of beer. As indicated in figure 1A, the current per capita level of consumption continues a downward trend from a high of 2.76 gallons in 1980-81 (Williams et al. 1991) (see also figure 2). An examination of trends in per capita consumption for 33 countries indicated comparable declines occurring after large postwar increases (Smart 1989). Although the current downward trend in alcohol consumption counters the modest increase that took place during the 1970s, current consumption is still considerably above the relatively stable per capita intake of around 2 gallons that prevailed during the 1950s, and it still reflects the relatively rapid increase in intake that occurred during the 1960s (Smart 1989).

Figure 1B illustrates that beer makes the largest contribution to per capita alcohol consumption in the United States, wine makes the smallest, and the contribution of spirits is intermediate. Trends in beverage-specific alcohol consumption between 1977 and 1989 are seen more clearly in figure 2, where they are expressed in terms of percentage of change since 1977. There has been a steady decline in the consumption of spirits to 74 percent of their total for 1977; wine consumption peaked in 1986 and has decreased since, but it is still above 1977 levels; and beer intake has been fairly steady, increasing somewhat, then returning to just over 1977 levels (Williams et al. 1991).

Total per capita consumption in gallons of alcohol for 1989 is presented by State in table 2, together with adjusted per capita consumption, which takes into consideration abstention levels reported for selected States (Williams et al. 1991). This adjustment was made to correct for underestimates of per capita consumption that result because abstainers contribute to the population denominator, even though they do not actually consume any of the ethanol that makes up the numerator. Thus, a State with a high proportion of abstainers may appear to have a low per capita alcohol consumption even though drinkers in that State drink heavily, simply because the per capita calculation assumes that everyone age 14 and older drinks the same amount. Data on abstention by State were obtained from the Behavioral Risk Factors Survey (BRFS), which is conducted in most States in coordination with the Centers for Disease Control. In the BRFS telephone survey, persons aged 18 years and older were asked about their

alcohol consumption in the previous month. It should be noted that individuals classified in the BRFS as "abstainers" may actually be infrequent drinkers who drank during the past year but not during the past month; “abstainers" may also be heavier drinkers who had not consumed alcohol during the month prior to the interview because of illness, alcoholism treatment, or other circumstances (Williams et al. 1991).

Based on BRFS data, Wisconsin had the lowest percentage (30.1) of abstainers, whereas West Virginia had the highest percentage (69.7). For States where such an adjustment was possible, taking abstention rates into account appeared to have a strong effect on the per capita consumption ranking. For example, the high abstention rates reported for several States in the South most likely contribute to the historically low per capita consumption rates in this region (Williams et al. 1991).

Drinking Patterns and
Trends

Data on alcohol consumption patterns in the
United States and trends in these patterns were
reported in the 1983 and 1988 Alcohol Supple-
ments of the National Health Interview Survey
(NHIS) (Williams and DeBakey 1992). For both
surveys, information on consumption of beer,
wine, and spirits was collected for the 2-week
period immediately prior to the interview; if the
respondent had not consumed any alcohol in
those 2 weeks, consumption was recorded for
the 2-week period prior to the respondent's last
drink within the past year. Abstainers fell into
three categories: lifetime abstainers (fewer than
12 drinks in a lifetime); former drinkers (12 or
more drinks in 1 or more years, but no drink in
the past year); and infrequent drinkers (an aver-
age of less than 0.01 ounce of alcohol daily in
the past year). Drinkers were classified on the ba-
sis of average alcohol intake per day: Individuals
consuming 0.01 to 0.21 ounce of alcohol (about
1 to 13 drinks per month) were categorized as
light drinkers; 0.22 to 0.99 ounce of alcohol
(about 4 to 13 drinks per week) as moderate
drinkers; or 1 or more ounces (about 2 or more
drinks per day or 14 or more drinks per week)
as heavier drinkers.

Drinking patterns reported in the 1988 NHIS are summarized for males and females according to selected demographic characteristics in tables 3A and 3B (Williams and DeBakey 1992).

Compared with females, males were less likely to abstain and more likely to drink heavily. In both genders, abstention rates increased with age and generally decreased as family income and education went up. Abstention was also more prevalent among blacks and Hispanics than whites, among those not in the labor force (including many older individuals who have retired), and among those living in the South. The highest prevalence of heavier drinking was observed among unemployed men.

The prevalence of abstinence reported in the 1983 NHIS is compared with that reported in the 1988 NHIS according to selected sociodemographic characteristics in table 4 (Williams and DeBakey 1992). These data indicate widespread increases in abstention in both men and women. In general, abstention increased significantly for individuals (both sexes) at all educational levels and among those with a family income of $10,000 or more; abstention rates also increased among those who were married or divorced/ separated. Increases in abstinence were significant for whites and persons not of Hispanic origin; changes were smaller and failed to reach statistical significance among blacks and Hispanics. With respect to age groups, a larger number of younger (ages 18 through 29) and middleaged (45-64) persons abstained from consuming alcohol in 1988 than in 1983; however, for other age groups, abstinence increased only among males aged 30 through 44 and females aged 65 and over. These trends are consistent with limited trend data available since 1986 from the Behavior Risk Factor Surveillance System (Williams et al. 1991).

Significant increases in abstention between 1983 and 1988 were accompanied by significant decreases in heavier drinking across a wide range of sociodemographic characteristics (Williams and DeBakey 1992); these patterns were observed for both men and women in the following groups: the young (18-29), whites, persons not of Hispanic origin, employed persons, and persons at the highest income level. A concomitant increase in abstention and reduction in heavy drinking was seen in men aged 30 through 64, at all educational levels, and with family incomes above $25,000. With respect to marital status, increases in abstention and decreases in heavy drinking were observed for married males and for divorced/separated persons of both genders. Blacks of both genders and Hispanic males displayed decreases in heavier drinking without accompanying increases in abstention.