Evaluating the Impact of Alcohol Warning Labels

Robert N. Mayer, University of Utah
Ken R. Smith, University of Utah
Debra L. Scammon, University of Utah
ABSTRACT - This paper describes the rationale, methodology, data collection, and preliminary results of a study to evaluate the impact of recently mandated warning labels on alcohol beverage containers. Data collection began in April 1989 and has proceeded at three-month intervals through July 1990. The results reported here involve three data points in each of the pre- and post-warning periods. The study also takes advantage of a non-exposed, non-equivalent comparison group to identify the independent effect, if any, of the labels. The results suggest that the warnings have achieved a considerable level of public awareness, with the admonition against drinking during pregnancy being the most memorable portion of the warning. But there is no evidence yet that the labels have affected knowledge of the health and safety risks associated with alcohol or self-reported alcohol consumption.
[ to cite ]:
Robert N. Mayer, Ken R. Smith, and Debra L. Scammon (1991) ,"Evaluating the Impact of Alcohol Warning Labels", in NA - Advances in Consumer Research Volume 18, eds. Rebecca H. Holman and Michael R. Solomon, Provo, UT : Association for Consumer Research, Pages: 706-714.

Advances in Consumer Research Volume 18, 1991      Pages 706-714

EVALUATING THE IMPACT OF ALCOHOL WARNING LABELS

Robert N. Mayer, University of Utah

Ken R. Smith, University of Utah

Debra L. Scammon, University of Utah

[Support for data collection for this project was provided by the College of Social and Behavioral Science, the University of Utah Research Committee, and the Division of Substance Abuse of the Utah State Department of Social Services.]

ABSTRACT -

This paper describes the rationale, methodology, data collection, and preliminary results of a study to evaluate the impact of recently mandated warning labels on alcohol beverage containers. Data collection began in April 1989 and has proceeded at three-month intervals through July 1990. The results reported here involve three data points in each of the pre- and post-warning periods. The study also takes advantage of a non-exposed, non-equivalent comparison group to identify the independent effect, if any, of the labels. The results suggest that the warnings have achieved a considerable level of public awareness, with the admonition against drinking during pregnancy being the most memorable portion of the warning. But there is no evidence yet that the labels have affected knowledge of the health and safety risks associated with alcohol or self-reported alcohol consumption.

INTRODUCTION

Warnings are one of the most common and politically palatable forms of information disclosure to consumers. Federally- or state-mandated warnings exist for cigarettes and smokeless tobacco, saccharin, tampons, aspirin used by children, an anti-acne drug, power lawn mowers, all-terrain vehicles, pressurized cans, and, as of November 1989, alcoholic beverages.

In sharp contrast to the ubiquity of product warnings is the small number of rigorous evaluation studies of their effects, especially studies carried out under real world conditions. And even when resources have been devoted to evaluation research, as in the case of cigarette warnings, it has proven difficult to single out the independent effects of a warning from other contemporaneous events and interventions.

This paper reports the preliminary results of a study designed to evaluate the impact of a newly required warning label for alcoholic beverage containers. All such containers shipped after November 19, 1989 are required to bear the following message:

Government Warning: (1) According to the Surgeon General, women should not drink alcoholic beverages during pregnancy because of the risks of birth defects. (2) The consumption of alcoholic beverages impairs your ability to drive a car or operate machinery, and may cause health problems.

The study reported here is based on data collected over a 15 month period. Three surveys were conducted before implementation of the warning, and three were conducted after; each involved a statewide random sample of adults. The primary outcome measures of interest are awareness of the new labels and their content, knowledge of the risks associated with alcohol consumption, and self-reported frequency and location of alcohol consumption. The study also takes advantage of the characteristics of a single state, Utah, to create a comparison group of devout Mormons who are virtually unexposed to the new warnings. The responses of the devout Mormons are intended to detect the impact of nonwarning factors on knowledge of the health and safety risks of alcohol consumption.

RELEVANT RESEARCH

Warnings are an appealing policy tool because of their relatively low cost (both in administration and compliance) and their consistency with individual freedom of choice. Despite the advantages of warnings and their frequent use by policy makers, there is only limited evidence that warnings are effective in altering individual behavior. Warnings may fail for any number of reasons. An individual warning may be hard to read, difficult to comprehend, or stated too generally. Even a well crafted warning may go unnoticed if it exists in an environment full of other warnings. Referring to this possible saturation effect, Paul Schlemm, chairman of Vintners International Inc., predicted that the new alcohol warning labels would be ineffective because "the public has become completely immune to warning labels" (Freedman 1989).

The warning policy that has been most extensively researched (and which is most akin to the case of alcohol) concerns cigarette warnings. Starting in 1965, the federal government has imposed ever stronger and more conspicuous warnings in an effort to reduce cigarette consumption, and indeed cigarette consumption has declined in the United States over this period. But there is no strong evidence that the warnings reduced consumption (FTC 1969; McAuliffe 1988; Murphy 1980; US DHHS, 1987a). Besides concerns that the cigarette warnings are unreadable or go unread (Davis and Kendrick 1989; Fischer et al. 1989; Myers et al. 1981), the fundamental problem is separating the possible effects of the warning from contemporaneous events (e.g., the release of reports by the Surgeon General or the death of a celebrity from lung cancer) or general trends (e.g., increased health awareness) that might also discourage cigarette consumption. A 1989 report by the Surgeon General (US DHHS 1989) found that "there are no controlled studies to permit a definitive assessment of the independent impact of cigarette warnings on knowledge, beliefs, attitudes, or smoking behavior" (pp.4 78-9).

There have been several laboratory studies of the effectiveness of alternative label formats (Karnes and Leonard 1986; Morris 1980; Viscusi and Magat 1987), but a recent review of research on the impact of health warnings concluded that no field study to date has used a pre- and post-test design with matched treatment and controlled groups (US DHHS 1987a). In the absence of matched treatment and control groups, evaluation research on the actual impact of product warnings has relied on statistical control of other potential influences. In an interrupted times series design, researchers compare trends before and after policy implementation, controlling for other factors that might affect the criterion variable under investigation. For example, in examining the impact of the saccharin warning label on the consumption of diet drinks, Schucker et al. (1983) controlled for advertising expenditures and prices for diet and non-diet drinks as well as for media coverage of the risks associated with saccharin. The value of interrupted time series designs is directly dependent on the researcher's ability to identify and accurately measure all the non-warning factors that might serve as competing explanations for any shifts in policy-relevant variables.

Sales or consumption data are usually the outcome measure used in interrupted time series designs. The main advantage of this type of data is that it is relatively free from measurement error, but it has several limitations as well. For one, there is usually a serious delay between policy implementation and accumulation of sufficient sales and consumption data to evaluate a policy; as a result, no short-run policy adjustments can be based on these evaluations. More important, using sales or consumption data to evaluate the impact of a warning ignores the possibility that a warning can be effective without changing aggregate data (e.g., by reinforcing "safe" behavior or shifting behavior from a less safe to a more safe setting).

Given the limits of interrupted times series studies, some researchers have used cognitive outcome measures and/or alternative methods of isolating the independent impact of a warning label. To evaluate the effectiveness of cigarette package warnings, for example, the Federal Trade Commission conducted a survey of consumer awareness and knowledge of the health hazards of smoking cigarettes (Myers et al. 1981). There are obvious problems in trying to draw conclusions from a cross-sectional study conducted after a policy has been in place; baseline data are needed.

In a 1978 study by the Swedish National Smoking and Health Association, nationally representative samples of adults were interviewed both before and after the 1977 implementation of rotating cigarette warnings. In addition to measuring knowledge of the health hazards covered by the warnings, respondents were queried about health hazards not covered by the rotating labels. The purpose of the latter questions was to measure background factors that might account for any effect of the new cigarette labels. Then, any effect attributed to the labels would need to be discounted by any increases in awareness of the non-covered health hazards. While this approach is ingenious, the possibility also exists that increased awareness of risks not covered in the labels might be an indirect or "spillover" effect of the labels.

The Swedish study attempted to isolate the effects of the cigarette warnings in a second way. The study approximated treatment and control groups by defining smokers as having been exposed to the labels, and nonsmokers as not having been exposed. Besides relying on self-reports of smoking behavior, the main problem with this approach is that it creates experimental groups based on one of the supposed effects of the treatment. If the labels worked as intended and caused people to switch from being smokers to nonsmokers, then the latter group might show larger increases in risk knowledge--the exact opposite of what was predicted. Stated more generally, differential increases in knowledge of health hazards between smokers and nonsmokers could reflect sample selection (i.e., the changing composition of the smoker group) as opposed to the impact of the labels.

In sum, evaluation studies require the external validity that is offered by field settings, but research designs that are feasible in the field pose threats to internal validity. Even when adequate data are collected in the pre-policy period, there is no single way of fully isolating the independent effects of a product warning. The identification of these effects remains a major challenge in conducting real world evaluations of the impact of product warnings.

METHODS

Research Design

In the absence of a control group unexposed to the warning, the best research design for studying the impact of a nationwide product warning is usually an interrupted time series design. As mentioned above, however, a warning must be in place for several years before an interrupted time series analysis can be carried out. Moreover, the sales and consumption data typically used in interrupted time series analyses may not be the most appropriate measure of a warning's impact.

The project described here attempts to blend some of the advantages of an interrupted times series design with the advantages of cognitive outcome measures collected directly from individuals. The study reported here has five key features:

1. data collected from statewide samples at multiple points both before and after the appearance of the alcohol warning labels;

2. multiple, cognitive outcome measures;

3. questions on risks not covered in the warnings;

4. a naturally occurring comparison group of people (devout Mormons) unexposed to the warning; and

5. a setting that minimizes the possibility of new, anti-alcohol interventions during the study period.

TABLE 1

SCHEMATIC OF STUDY DESIGN

Samples and Data Collection

Data collection for this project took place at three month intervals beginning in April 1989. Questions were attached to an ongoing omnibus statewide survey. There were three waves of data collection before implementation of the new alcohol warning labels (April, July, and October of 1989). To date, there have been three waves of data collection after the warnings began to appear in November 1989 (January, April, and July 1990). For all six waves, data were collected from a representative, statewide sample of at least 400 Utah households. Within each household, data were gathered from a randomly selected person aged 18 or older.

Data collection was conducted by interviewers from a University-based survey research center via computer-assisted telephone techniques. The cooperation rate (the number of known eligible respondents completing the survey divided by the number of eligible households reached) across the six surveys was 81%.

Measures

Ultimately, the impact of the alcohol warning labels should be gauged in terms of the number of traffic fatalities and birth defects it prevents. As a proxy, alcohol sales or consumption data might also be used. In the short run, however, cognition-based outcome measures are both more feasible and appropriate. If the labels are going to affect behavior, first they must be noticed and change consumer perceptions of the risks of alcohol consumption. Accordingly, this study uses a combination of awareness, knowledge, and self-reported behavioral items to measure the possible impact of the alcohol warning labels. Table 1 shows the measures used in each of the six surveys.

The awareness measures are straightforward; they ask respondents whether alcohol warnings appear on beverage containers and in establishments that serve alcohol. The inclusion of a measure gauging perception of the risks associated with alcohol consumption is suggested by several theories of behavioral change (e.g., health belief model). Respondents can be asked to rate the degree of danger or harm associated with a particular object or activity. An attributable risk perception measure was chosen for this study because it has an objective referent and can therefore be used as an indicator of knowledge. The attributable risk question posed to respondents took the following form: What would be your best estimate of the percentage of traffic deaths in Utah that are related in some way to the consumption of alcohol? (For the sake of comparison, a relative risk measure was also used, but the results are not reported here.)

It is important to note that the attributable risk measures refer both to risks covered on the labels (drinking when driving or when pregnant) and risks not covered on them (cirrhosis of the liver and leukemia). There is a strong and well-known relationship between alcohol consumption and cirrhosis of the liver. If some non-warning factor or event heightens public perception of alcohol-related risks in general, this should be reflected in the cirrhosis measures in addition to the two risks listed on the warning labels. Perceptions of alcohol's role in causing leukemia (for which there is no scientific evidence) are used to control for the effects of social desirability in social surveys (i.e., the tendency to alter answers in line with what a respondent believes is expected or most acceptable).

Utah-Specific Characteristics of the Study

The study attempts to take advantage of the unique characteristics of a Utah sample. Whereas Utah's low level of alcohol consumption makes it an unlikely candidate for an evaluation study of alcohol warning labels, Utah's devout Mormons constitute the closest group possible to a naturally occurring, non-exposed, non-equivalent comparison group (Phillips and Calder 1980).

All states have people who abstain from alcohol, but even such abstainers may come into contact with the new labels in grocery stores or at social gatherings. In contrast, many Utahns are devout Mormons who not only abstain from the consumption of alcoholic beverages but are extremely unlikely to have first-hand contact with the new warning labels. (Alcoholic beverages in Utah, with the exception of reduced-alcohol beer, can only be purchased in state-owned liquor stores.) Thus, like a true control group, devout Mormons in Utah are highly unlikely to be exposed to the experimental stimulus (i.e., the warning labels). Unlike a true control group, however, they may differ from other study participants who may be exposed to the new labels. The inclusion of devout Mormons in the study not only makes the results representative of the entire state but, more importantly, helps identify non-label influences on public awareness and knowledge.

A Utah-based study has one final advantage. Unlike other states in which warning posters for alcohol have recently been required in establishments that sell alcohol (e.g., California), Utah's posting requirement has been in place since 1978. One does not have to worry, therefore, that the effects of the new warning labels might be confounded by the introduction of new warning requirements for establishments that sell alcoholic beverages.

RESULTS

Before examining the data for possible changes between the pre- and post-warning periods, it is necessary to examine changes within these two periods. In addition, the accuracy of the responses can be assessed, and the responses of study subgroups can be compared.

Differences Among Groups in the Pre-Warning Period

In this study, religion (including religiosity) was used to differentiate the sample by degree of potential exposure to the alcohol warning labels. Using religion is preferable to using self-reported alcohol consumption because the latter can be influenced by exposure to the warning labels. That is, if the warning has the desired effect of changing drinking behavior, the composition of the experimental groups will change as well.

Using information on religion and the frequency of church attendance, the respondents can be divided into three groups: devout Mormons (50.7% of all subjects), less-devout Mormons (19.2%), and non-Mormons (30.1%). Devout Mormons attend church at least once a week. Less devout Mormons attend church less often than once a week.

Table 2 shows that the three study groups differ from one another. Not surprisingly, self-reported alcohol consumption varies substantially, with devout Mormons reporting almost total abstention. More interesting is the fact that devout Mormons consistently attribute a greater role to alcohol in causing health and safety problems than do the non-Mormons. The attributions of less devout Mormons generally fall between the other two groups.

Accuracy of Responses

Table 2 also reveals several important inaccuracies in respondent awareness of alcohol warnings and knowledge regarding the health and safety risks of alcohol consumption. A substantial number of people in the pre-warning surveys incorrectly reported that the alcohol warning labels existed when they did not. Post-warning estimates of label awareness need to be discounted by the amount of these "false positives."

Respondents, regardless of religion, tended to overestimate the role of alcohol in causing various health and safety problems, especially in the case of birth defects. Scientific estimates of the role of alcohol in causing traffic fatalities, birth defects, cirrhosis of the liver, and leukemia vary. Using national data, the best. estimates seem to be 40% for traffic fatalities (NHTSA 1988a, 1988b), 5% for birth defects (US DHHS 1987b), at least 50% for cirrhosis of the liver (Alcohol and Health 1987), and 0% for leukemia. Because of Utah's low level of alcohol consumption, these figures would be somewhat lower in Utah.

The self-reported items on alcohol consumption raise a final issue regarding accuracy. Self-report items are always suspect, especially when they involve socially stigmatized behavior. In this study, 73.4% of respondent said that they consumed alcohol less than one time per month or not at all. A 1985 study conducted in Utah reported that 69.3% of adults described themselves as nondrinkers (Smith et al. 1990). In addition to these similar results, a 21-state study found that self-reported alcohol consumption is highly correlated (r = 0.81) with per capita sales of alcoholic beverages.

Stability of the Data

The stability of the data within the pre- and post-warning periods has both substantive and methodological implications. Substantively, it is important to know whether the outcome variables were changing even before implementation of the warning. If they were, an observed pre-post change might simply be the function of long-term trends rather than the warning label itself. Similarly, the shape of the post-warning curve needs to be known to see whether any effects increase, decrease, or remain constant over time. From a methodological point of view, the stability of the pre- and post-warning data influences the methods of data analysis used. If, for example, the data are stable within both periods, the results of different surveys can be combined and the data analysis can be greatly simplified.

TABLE 2

DIFFERENCES WITHIN AND ACROSS STUDY PERIODS BY RELIGIOUS GROUPS

Table 2 shows that the outcome measures were extremely stable within the pre- and post-warning periods. Within the pre-warning period, there were only two measures whose value changed using a statistical significance level of p < .05. Among both non-Mormons and devout Mormons, the perceived role of alcohol in causing birth defects declined in the final pre-warning survey. While the reason for this decline is not known, one can at least rule out the possibility that some non-warning factor was operating to increase sensitivity to the connection between alcohol and birth defects. For the post-warning period alone, none of the measures changed within any of the three subgroups.

Pre-Post Differences

The stability of the data within the pre- and post-warning periods permits one to collapse within the two periods and compare across them. Awareness of the warning labels increased between the two periods, but surprisingly, it did so in all three groups, including the devout Mormons. This suggests that awareness of the warning, even among potential drinkers, might be partly attributable to factors such as media coverage of the labels. The researchers monitored television and newspaper coverage during November 1989. Interestingly, coverage in the media owned by the Mormon Church (KSL-TV and the Deseret News) was virtually nonexistent. While many devout Mormons are exposed to media other than these two, the increased awareness among devout Mormons could be due to some other unidentified factor.

Given the increased awareness of the new warning labels, what individual characteristics predict the likelihood that a person will report awareness? Table 3 presents the results of a logistical regression based on the three post-warning surveys. The analysis separates people who reported awareness of the warning labels from all other respondents (i.e., people who didn't know whether the labels existed are combined with those who incorrectly said that the labels did not exist).

TABLE 3

LOGIT ANALYSIS OF WARNING LABEL AWARENESS FOR MERGED POST-WARNING SURVEYS

The results indicate that awareness was related to a respondent's age and religious group. Older respondents and devout Mormons were least likely to report awareness of the new labels. Educational attainment, sex, and area of residence were not related to warning label awareness. Additional analyses included a sex by age interaction term and attributable risk responses (as a measure of possible predisposition to the warning label) but yielded the same results. The key finding is that younger people (18-30 years old in this survey), who are at greatest risk of driving under the influence of alcohol or drinking while pregnant, were most aware of the warning label.

Among the attributable risk measures, there were some pre-post differences, most of which are difficult to explain. Counter to what might have been expected as a result of the warning labels, the perceived role of alcohol in causing birth defects declined between the pre- and post-warning periods. Among non-Mormons only, perceptions of the role of alcohol in causing cirrhosis of the liver increased. There were no changes among the three groups in attributions involving traffic fatalities or leukemia.

Among less-devout Mormons, awareness of warnings posted in establishments that sell alcohol increased between the two study periods. If the new labels somehow heightened awareness of already existing posted warnings, why would this effect not be equally pronounced among non-Mormons? A final anomalous finding is that among less-devout Mormons who report drinking, the percentage of those who drink primarily at home decreased between the pre- and post-warning periods. This change cannot be explained in terms of seasonal changes (i.e., the Christmas drinking season).

A final set of findings, not shown in the tables, is potentially quite important. In the three post-warning surveys only, people who answered "yes" to the question of whether warning labels appear on beverage containers were asked follow-up questions. First, they were asked whether they had actually seen the labels. If they had, they were asked in four separate questions whether the label contained information about the role of alcohol in traffic fatalities, birth defects, leukemia, and cirrhosis of the liver. Combining the three post-warning surveys, the most readily recalled part of the warning was the message about birth defects, with 66% of the relevant respondents correctly noting the inclusion of this information in the labels. In comparison, only 54% of the respondents correctly recalled the label's caution against drinking and driving. Small percentages of respondents incorrectly stated that the labels mentioned cirrhosis of the liver (14%) and leukemia (4%).

In light of the fact that alcohol accounts for a far greater percentage of traffic fatalities than birth defects, it is striking that a larger percentage of respondents remembered that the labels contained information about birth defects than about traffic fatalities. One possible explanation is that the labels explicitly mention birth defects while traffic fatalities are only alluded to with the phrase "consumption of alcoholic beverages impairs your ability to drive a car." A second possibility is that the link between alcohol and traffic fatalities is common knowledge, but the link between alcohol and birth defects has only recently become a matter of scientific consensus. The connection between alcohol and birth defects still has the quality of "news," and seeing it on government mandated labels may, for the moment at least, be capable of catching a person's attention.

DISCUSSION AND CONCLUSIONS

Given the limitations of the present study as well as the fact that additional post-warning data remains to be collected, what can be said about the success or failure of the alcohol warning labels? From the point of view of public health promotion, there are three positive points. First, the new labels have been noticed by all population subgroups. Among non-Mormons, awareness of the new labels already exceeds awareness of the warning posters that have been required in Utah for 12 years! Second, awareness of the labels is greatest among young people, that is, among the group at highest risk of driving under the influence of alcohol (young males) or drinking during pregnancy (young females). Third, public recall of label content is quite accurate; moreover, information about birth defects seems to be particularly memorable, probably due to its quality as new information.

The results of this study are very similar to those reported by Mazis, Morris, and Swasy (1991). Comparing results from two national surveys conducted in May 1989 and May 1990, they found increased label awareness across time, with awareness being greatest among younger respondents. (They also reported substantial numbers of false positives in the pre-warning survey.) The major difference between the findings of Mazis et al. and this study concerns risk perception. While we found no evidence to this point that the labels have affected knowledge and perception of the role of alcohol in causing health and safety problems, Mazis and his colleagues found a small but statistically significant increase in the number of people who described alcohol as "very harmful." (There was no significant change between their two surveys, however, when "very harmful" and "somewhat harmful" responses were combined.)

While the absence of a clear impact on risk perception and knowledge may be disappointing to public health professionals, perhaps this was to be expected. Consider that the labeling requirement has been in effect for less than a year. With the exception of beer, alcoholic beverages continue to be sold in containers that were shipped prior to the time of the warning requirement. In addition, many of the early warning labels suffered from problems of noticeability and legibility, resulting in new requirements, beginning in November 1990. Finally, research on other forms of consumer information disclosure suggests that cognitive changes (awareness and knowledge), precede changes in affect, which, in turn, precede changes in behavior.

Other methodological approaches might also yield more positive assessments of the alcohol warning label's effects. Clearly, data collection needs to continue beyond the first few months of the post-warning period. Eventually, however, one runs into the problem of differentiating the warning from other historical factors. Another step might be to focus exclusively on high risk populations (e.g., young people or Native Americans). A final approach would be to use a panel design rather than a trend design. In a panel design, one can directly observe individual change. In a trend design, one can only measure net change across the population.

Even with other approaches, one might still conclude that the alcohol warning label, even in its revised form, has not been and will not be successful. If so, one should start with the content of the label for an explanation. The label states the obvious with respect to drinking and driving, while remaining vague about the health risks of alcohol consumption (other than during pregnancy). In contrast to the weak content of the warning labels, the current legislative proposal to require rotating warnings in advertising is very hard-hitting. One of the warnings would state that "alcohol is a drug and may be addictive." Another would say that "the consumption of this product, which contains alcohol, can increase the risk of developing hypertension, liver disease, and some cancers." Judging from the results of this study, a warning will be effective to the extent that it contains specific information that is not already widely known. Without such information, one should not be surprised if people ignore labels and rationalize their content.

REFERENCES

Alcohol and Health. Seventh Special Report to the U.S. Congress from the Secretary of Health and Human Services. Rockville, MD: National Institute on Alcohol Abuse and Alcoholism, 1990.

Davis, R.M. and Kendrick, J.S. (1989), "The Surgeon General's Warnings in Outdoor Cigarette Advertising," Journal of the American Medical Association, 261, pp. 90-94.

Federal Trade Commission (1969). Report to Congress Pursuant to the Federal Cigarette Labeling and Advertising Act. Washington, D.C.: Federal Trade Commission.

Fischer, P.M., Richards, J.W., Jr., Berman, E.J., and Krugman, D.M. (1989), "Recall and Eye Tracking Study of Adolescents Viewing Tobacco Advertisements," Journal of the American Medical Association 261, 1989, pp. 84-89.

Freedman, A.M. (1989), "Rebelling Against Alcohol, Tobacco Ads," Wall Street Journal, November 14, 1989, p.B1.

Karnes, E.W. and Leonard, S.D. (1986), "Consumer Product Warnings: Reception and Understanding of Warning Information by Final Users." In: Karwowski, W., ed., Trends in Ergonomics/Human Factors 111. Holland: Elsevier, pp. 9951003.

Kruger, Elizabeth L. (1988), "Mitigating Alcohol Health Hazards Through Health Warning Labels and Public Education," Washington Law Review 63, pp. 979-996.

Mazis, Michael B., Louis A. Morris, and John L. Swasy (1991), "An Evaluation of the Alcohol Warning Label: InItial Survey Results," Journal of Public Policy and Marketing 10, No.1, in press.

McAuliffe, Robert E. (1988), 'The FTC and the Effectiveness of Cigarette Advertising Regulations," Journal of Public Policy and Marketing 7, pp. 49-64.

Myers, M.L., Iscoe, C., Jennings, C., Lennox, W., Minsty E. and Sacks, A. (1981). Staff Report on the Cigarette Advertising Investigation. Washington, D.C.: Federal Trade Commission.

Morris, L.A. (1980), "Estrogenic Drugs--Patient Package Inserts." In: Morris, L.A., Mazis, M. and Barofsky, I., eds., Product Labeling and Health Risks--Banbury Report 6. Cold Spring Harbor, NY: 1980, pp. 23-36.

Murphy, R.D. (1980), "Consumer Responses to Cigarette Health Warnings." In: Morris, L.A., Mazis, M. and Barofsky, I., eds., Product Labeling and Health Risks--Banbury Report 6. Cold Spring Harbor, NY: 1980, pp. 13-21.

National Highway Traffic Safety Administration (1988a). Alcohol Involvement in Fatal Crashes 1986. Report No. DOT HS 807 268. Washington, DC: NHTSA.

National Center for Statistics and Analysis (1988b). Drunk Driving Facts. Washington, DC: NHTSA.

Phillips, L.W. and Calder, B.J. (1980), "Evaluating Consumer Protection Programs: II. Promising Methods," Journal of Consumer Affairs 14, pp. 9-36.

Schucker, R.E., Stokes, R.C., Steward, M.L., and Henderson, D.P. (1983), "Impact of the Saccharin Warning Label on Sales of Diet Soft Drinks in Supermarkets," Journal of Public Policy and Marketing, Volume 2, 46-56.

Smith, Perry F., Patrick L. Remington, David F. Williamson, and Robert F. Anda (1990), "A Comparison of Alcohol Sales Data with Survey Data on Self Reported Alcohol Use in 21 States," American Journal of Public Health, 80 (March), 309-12.

Swedish National Smoking and Health Association Study of Sweden's rotational label system (1978). Described in U.S. Department of Health and Human Services, 1987a.

U.S. Department of Health and Human Services (1987a). Review of the Research on the Effects of Health Warning Labels: A Report to the United States Congress. U.S. Department of Health and Human Services.

U.S. Department of Health and Human Services (1987b). Program Strategies for Preventing Fetal Alcohol Syndrome and Alcohol-Related Birth Defects. Rockville, MD:: National Institute on Alcohol Abuse and Alcoholism.

U.S. Department of Heath and Human Services (1989). Reducing the Health Consequences of Smoking: 25 Years of Progress. A Report of the Surgeon General. Rockville, MD: U.S. Department of Health and Human Services, Public Health Service, Centers for Disease Control, Center for Chronic Disease Prevention and Health Promotion, Office on Smoking and Health.

Viscusi, W.K. and Magat, W.A. (1987). Learning About Risk. Cambridge, MA: Harvard University Press.

----------------------------------------