Oddly, maps of “excessive drinking” in the United States do not correspond particularly well with maps of per capita alcohol consumption. New Hampshire, Delaware, and Nevada, for example, are typically depicted as having the country’s highest levels of per capita alcohol consumption but only moderate levels of excessive drinking. In contrast, Iowa is usually depicted as having one of the highest levels of excessive drinking but only a moderate level of total alcohol consumption. (Unfortunately, maps of per capita alcohol consumption by county are not available.)
Such a seeming discrepancy can easily be explained by differences in drinking behavior. It is possible, for example, that on average Nevadans drink more than Iowans but generally do so in a more responsible manner, engaging less in “binge drinking.” Possible yes, but probable? I doubt it. I think it is more likely that the data used to make these maps are flawed. In the remainer of this post, I will critically examine the sources used for maps of “excessive drinking.” Admittedly, much of the evidence that I present was derived from “conversations” with AI; as such, my arguments should also be regarded with some skepticism.
The foundation of excessive-drinking maps is the data tabulated by the CDC’s Behavioral Risk Factor Surveillance System (BRFSS), derived from state-based telephone surveys. Sample sizes and exact questions, however, evidently vary by state and year, complicating comparisons over time and space. More significant, honest answers about such problematic behavior cannot always be expected. Supposedly, surveys routinely recover only a fraction of the amount of alcohol that sales data imply, often on the order of 30–60%. Such underreporting supposedly stems from forgotten drinks, underestimated pour sizes, and nonresponses by heavier drinkers. Many respondents, deemed “high impression managers,” want to appear respectable by reporting that they drink less than they actually do. According to one academic study, “high impression managers report 20 to 33% less consumption and are about 50% less likely to report risky drinking.” Such considerations worrisomely suggest that actual rates of excessive drinking are considerably higher than they appear to be on these maps.
But it is also known that the bias can run in the opposite direction, particularly among certain groups of college students for whom heavy drinking can be seen as a badge of belonging. Is it therefore coincidental that counties that are often mapped as having the highest rates of excessive drinking also have large college-student populations, such as Missoula and Gallatin in Montana and Dane in Wisconsin? More to the point, more under-reporting is to be expected in areas in which alcohol is deeply stigmatized (Utah and eastern Idaho; interior regions of the U.S. South) than in areas with alcohol-tolerant cultures (the upper Midwest and much of the interior West).
The definition of “excessive drinking” used in these surveys are also problematic. The CDC definition seems straightforward: either “binge drinking” (four or more drinks on one occasion for women, and five or more for men), or “heavy drinking” (eight or more drinks per week for women, and fifteen or more for men). Although the first threshold seems reasonable, the second is questionable. I doubt that most people would regard a woman who drinks a glass of wine with dinner every day, and two on Saturdays, as an “excessive drinker.” By the same token, is a man who drinks two beers a day, and three on Fridays, really a “heavy drinker?” Perceptions obviously differ widely from person to person. More important, actual alcohol consumption depends on what kind of beer the drinker in question consumes. A “light” beer, for example, typically has less than half the alcohol content of imperial IPA of the same size.
A similar problem is encountered with the concept of “binge drinking.” “Five or more drinks on one occasion” might seem like a high figure, but it depends on how long an “occasion” lasts. The standard figure for research purposes is two hours. This duration is used because a typical man who consumes five standard drinks (or 0.6 fluid ounces of ethanol) in two hours will reach a blood-alcohol concentration over 0.08%, which counts as legally “drunk” in most U.S. states. But the surveys on which excessive-drinking data are based do not seem to be rigorous in this regard. According to AI, “respondents are not walked through a two-hour clock. So five beers over a six-hour dinner can be coded the same as five shots in an hour. That is a real measurement gap, not a quibble.”
For both “heavy drinking” and “binge drinking,” individual weight and muscle-mass should also be considered. All other things being equal, a heavier person must drink more alcohol than a lighter person to reach the same blood-alcohol concentration, which is the real measurement of intoxication. A muscular person, moreover, will have a lower blood-alcohol concentration than a less-muscular person of the same weight who drinks the same amount of alcohol over the same period of time.
Ideally, alcohol-consumption surveys would account for such factors, but that is not necessarily the case. Considerations of weight and muscle mass are ignored, and a twelve-ounce beer is usually treated as standard unit regardless of alcohol content. After much back-and-forth, my AI “informant” concluded by telling me that the definition of excessive drinking employed by the CDC’s Behavioral Risk Factor Surveillance System (BRFSS) “is operational and crude on purpose. It is useful for comparing places and tracking a risk category. It is a poor description of how drunk someone got on a given night or how much alcohol they actually swallowed.”
Finally, alcohol-consumption surveys face the same problems of sample representativeness that bedevil all pollsters. Many if not most persons refuse to respond to such surveys, and it cannot simply be assumed that non-responders have the same experiences as responders. Researchers are well aware of this problem and try to compensate, but it is not clear whether they do so adequately when it comes to “excessive drinking.” According to AI:
BRFSS tries to account for refusal and uneven participation—but only in a limited way. It does not fully correct for the possibility that drinkers and non-drinkers refuse at different rates. Most people who are called never complete an interview. Recent median state response rates are around 44% (roughly 31–65% by state). That is normal for large telephone surveys now, not a secret. It assumes that, within those groups, people who answer the phone drink like people who hang up. Such an assumption is questionable.
All of this is not to argue that maps of excessive drinking in the United States are worthless. I find them interesting and regard them as partially valid. But they are more suggestive than definitive. I would also prefer that they be give more modest titles, such as “Rates of Excessive Drinking as Indicated by Public-Health Surveys.” A brief note on how “excessive drinking” is defined would ideally be included on all such maps.