ALCOHOL ADVERTISING IN MAGAZINES AND READERSHIP DEMOGRAPHICS: DO ADOLESCENTS MATTER FOR REGULATION? by Jon P. Nelson Department of Economics 603 Kern Building Pennsylvania State University University Park, PA 16802
[email protected]
Abstract. This paper examines the empirical relationship between alcohol advertising placements in magazines and adolescent readership. Poisson and negative binomial regressions are applied to count data on cumulative placements by beverage for 35 major magazines during the years 1997-2001. The results indicate that demographics other than youth readership determine the placement of alcohol advertisements in magazines. Beer and distilled spirits advertisers tend to favor magazines read by young adults, blacks, and males, but not adolescents. Wine advertisers tend to favor magazines read by older adults and higher-income adults, but not adolescents. Comparing alternative specifications and econometric models, the results fail to support direct or indirect targeting of adolescent readers. The empirical results are applied to regulatory issues as guided by Supreme Court decisions on commercial speech and by the Federal Trade Commission. Keywords: Magazines Advertising; Alcohol Beverages; Adolescents; Regulation JEL classifications: L82, M3, I18, K20
Forthcoming, Regulation: The Cato Review of Business & Government
November 2004 (revised)
INTRODUCTION The regulation of alcohol advertising is major part of the debate over the role of beverage alcohol in society. Deceptive and misleading alcohol advertising is regulated by the U.S. Federal Trade Commission (FTC) and the Alcohol and Tobacco Tax and Trade Bureau (TTB) in the U.S. Department of Treasury. Self-regulation also occurs in the form of industry advertising codes, such as the Beer Institute’s Advertising and Marketing Code. In 2001, the U.S. alcohol industry spent more than $1.5 billion on advertising using conventional broadcast and print media, including expenditures of $397 million on magazine advertisements.1 The magnitude of expenditures and the possibility that some ads might “target” youth has generated concern about the adverse effects of alcohol advertising. A recent report by the National Research Council (NRC 2004, p. 137) includes a recommendation, among others, that alcohol industry trade associations should strengthen their advertising codes in order to preclude placement of commercial messages in media and other venues where a significant proportion of the expected audience is underage. However, demonstrating that advertising has an effect on alcohol consumption, rather than brand equity, has proven to be elusive (Calfee 1997; Nelson 2001, 2005). A report to Congress by the National Institute on Alcohol Abuse and Alcoholism (NIAAA 2000, p. 422) concluded that “the results of research on the effects of alcohol advertising are mixed and not conclusive.” The report of the NRC (2004, p. 134; Grube 2004), which was conducted at the behest of Congress, reached much the same conclusion. Most previous econometric research is based on analysis of alcohol consumption and either advertising expenditures or advertising bans (Nelson 2001, 2004; Saffer 1993). These studies have several problems that might limit the application of empirical results to policy issues. First, expenditure data are largely aggregate and cover the entire industry or major product categories, such as beer, wine, and distilled spirits. In short time-series these data contain relatively little variation, but the larger problem is likely to be that consumption levels also do not vary much in the aggregate. Quarterly data contain more variation and have been used in a few studies. Second, cross-country studies of broadcast advertising bans also are carried out at aggregate levels and require comparisons of countries with different drinking cultures and traditions (Nelson and Young 2001; Saffer 1991). Several different approaches are used to control for cross-country drinking cultures, with conflicting results and outcomes. Third, state-level studies of advertising bans, such as Nelson (2003), use panel data and are based on lower level of aggregation, but the media banned are more limited (e.g., billboards, store displays). 1
The division of magazine advertising expenditures by beverage in 2001 was beer, 9.6%; spirits, 79.4%; and wine, 9.6%. Total advertising expenditures in the U.S. economy in 2001 were about $231billion for all products and services (Bureau of the Census 2003). Hence, alcohol advertising accounts for about 0.6% of all advertising.
2 While none of the studies of bans are directly concerned with youth drinking behaviors, the presumed importance of advertising should be detectable if the bans are long-standing. Examination of adverse behaviors (e.g., traffic fatalities) adds to the policy importance of these studies. However, it desirable to consider an alternative approach to the problem of advertising and youth drinking that avoids some of the difficulties associated with aggregate data.2 Rather than attempting to measure consumption-advertising relationships, an alternative methodology is to examine the relationship between frequency of alcohol advertising in various media and the potential audiences reached by such placements, including adolescents. This approach has the advantage that it can empirically separate advertising placement by demographic group, and it avoids problems associated with measuring expenditures or the difficulty of making subjective judgements about advertising content. The purpose of this paper is to measure the empirical relationship between cumulative alcohol advertising placements in major magazines and youth readership. Using count data on total advertising placements over the five-year period 1997-2001, Poisson regression models are estimated that control for the distribution of readers by age, gender, race, and income. In order to reduce collinearity in the readership numbers, several additional explanatory variables are considered, including the annual number of issues, median age of adult readers, and median reader income. I also demonstrate the effects of using different econometric techniques to account for the skewed distribution of ad counts. For example, data on beer advertising in 35 major magazines indicates there were eight magazines containing no beer advertising whatsoever during the years 1997-2001; ten magazines averaging one or fewer ads per year; and only seven magazines averaging more than ten ads per year or about one beer advertisement per issue for a monthly magazine. As demonstrated below, there are only a few major magazines that contain a large number of alcohol advertisements and these magazines tend to have large readerships among youth, young adults, and older adults. Thus, it is necessary to consider econometric methods that account for zero counts and small positive values of advertising placements. The paper considers several alternative methods for handling these data problems. The remainder of the paper is divided into four sections. Section 1 comments on the policy debate by discussing First Amendment rights of commercial advertisers and provides a brief review of previous research on alcohol advertisements in magazines. Section 2 describes the data and econometric models used in the present study, including statistical distributions for advertisements by beverage.
2
By several measures, the magnitude of youth drinking has declined substantially in the last twenty years, despite high levels of alcohol advertising. According to the Monitoring the Future Survey, drinking in the last 30 days by high school seniors was 29% lower in 2001 compared to 1982; http://www.monitoringthefuture.org. Alcohol-involved auto crashes by drivers aged 16 to 20 fell by 19% between 1982 and 2001 (NHTSA 2003, p. 30).
3 Section 3 presents the empirical results and demonstrates the relationship (or lack thereof) between cumulative advertising placements and readership demographics. The primary relationship of interest is the effect of increased numbers of adolescent readers as an explanatory variable and the resulting proportional change in the cumulative number of alcohol ads. A definitive relationship between the two variables is evidence of targeting of youth (directly or indirectly), and would therefore provide support for the policy recommendation of the National Research Council. The lack of a relationship suggests that alcohol producers exercise appropriate restraint through their self-regulatory codes. Section 4 contains the conclusions from the study.
1. FIRST AMENDMENT AND ALCOHOL ADVERTISING STUDIES A long series of Supreme Court decisions, culminating in Central Hudson (447 U.S. 557) in 1980 and 44 Liquormart (517 U.S. 484) in 1996, demonstrate that commercial speech enjoys attenuated First Amendment rights, and these rights are not trumped by the restrictions on alcohol distribution contained in the Twenty-First Amendment. In Central Hudson, the Court laid out a four-prong approach to regulation of commercial speech. First, the message content cannot be misleading and must be concerned with a lawful activity or product. Second, the government’s interest in regulating the speech in question must be substantial. Third, the regulation must directly and materially advance that interest. Fourth, the regulation must be no more extensive than necessary to achieve its goal; that is, there must be a “reasonable fit” between means and ends, with the means narrowly tailored to achieve the objective. Applying the third- and fourth-prongs, the Court in 44 Liquormart struck down a Rhode Island law that banned retail price advertising of beverage alcohol. In doing so, the Court held that a state’s power to ban alcohol entirely under the Twenty-First Amendment did not include the “lesser power” to restrict commercial speech in violation of the First Amendment (517 U.S. 484, at 510). The common understanding prior to the 1980 decision in Midcal Aluminum (445 U.S. 97) was that the Twenty-First Amendment gave states the absolute power to regulate beverage alcohol, despite interference with other federal laws. Decisions since Midcal have further weakened states’ ability to use their Twenty-First Amendment regulatory power in an anti-competitive manner (Denning 2002; Douglas 2000; Foust 2000). In this context, important alcohol advertising cases include Capital Cities Cable (467 U.S. 691), where Oklahoma’s ban of cable TV advertising was pre-empted by federal law; Coors (514 U.S. 476) in which a federal law banning labels displaying alcohol content of beer were held to violate the First Amendment; Bad Frog Brewery (134 F.3d 87) in which New York’s ban of certain beer labels was not sustained under Central Hudson; and Utah Licensed Beverage Association (2001 U.S. App. Lexis 16624), where Utah’s restrictions on alcohol advertising were not sustained under Central Hudson. In Pitt News (2004 U.S.
4 App. Lexis 15615), the appeals court for the third circuit struck down Act 199, a Pennsylvania law passed in 1996 that banned advertisers from paying for the dissemination of alcohol advertising by any communications media affiliated with a university, college, or other educational institution (i.e., student newspapers). The state argued that newspapers could run free advertisements, but failed to provide evidence that the law would be effective in limiting underage drinking. Applying the third- and fourthprongs of the Central Hudson test, the court ruled that Act 199 was an impermissible restriction of commercial speech. In summary, Central Hudson requires a “balancing-of-interests” test to examine regulation of commercial speech, including alcohol advertisements. The test weighs the government’s obligations toward freedom of expression with its interest in limiting the content of some advertisements. In statistical language, the test reduces the risk of a Type I error (false positive) and the resulting restriction of commercial speech. Reasonable constraints on time, place, and manner are tolerated, and false or deceptive advertising remains illegal. However, in order for a regulation of alcohol advertising to be sustained, it must be demonstrated that the advertising deliberately or unnecessarily targets youth or the advertisements result in alcohol abuse by youth. Thus, Central Hudson delineates the empirical studies and results that are likely to be relevant to policymakers and regulators. Alcohol Advertising in Magazines. Early research on alcohol advertising in magazines was based on the method of content analysis, which attempts to demonstrate targeting of youth by examining the content dimensions of alcohol ads. Alcohol content analysis studies were conducted by Atkin and Block (1981), Breed and DeFoe (1979), and Strictland, Finn, and Lambert (1982). According to Strictland (1984) and the FTC (1985), content-analysis studies tend to employ overly broad categories to codify various “lifestyle themes” in advertising and impute motivations to advertisers based on largely subjective judgements about the themes used in the ads. The evidence provided by content analyses tends to be informal and anecdotal in nature, and it stops short of the evidence that would likely satisfy the Central Hudson test. Placement Studies. Given the shortcomings of content analyses, more recent studies attempt to measure exposure to alcohol advertisements by adolescents and youth, including two studies published by the Journal of the American Medical Association (JAMA). Sanchez et al. (2000) examined advertising placements in a sample of 15 magazines for the period July 1997 to July 1998. However, the authors used the holdings of a local high school library and did not attempt to obtain missing issues for some of the magazines. Instead, they simply estimated the total number of ads based on the issues available at the high school or expanded the sample to include 1998-1999 ads for a few popular magazines. The possible effects of this inconsistent sampling process were not discussed. Adolescent readership was measured by
5 the absolute number of readers in the age group 12-17 years, which ranged from 0.7 million for Harper’s Bazaar to 5.2 million for Sports Illustrated. In their reported results, the authors simply provide a tabular ranking of the magazines according to the estimated yearly number of alcohol and tobacco ads, which fails to account for other differences in readership demographics, such as the number of adult readers. Using these counts, the top-ranked magazines were Sports Illustrated and Rolling Stone with 5.2 and 1.9 million youth readers, respectively. Based on this limited and incomplete evidence, the authors concluded that “alcohol advertisers may target youths” (Sanchez et al. 2000, p. 2107). A second JAMA study of magazine placements and youth readership (ages 12-19) was conducted by Garfield, Chung, and Rathouz (2003). This study examined the frequency of alcohol advertising in 35 major magazines during the five-year period 1997-2001, conditional on three readership cohorts. Garfield et al. (2003, p. 2428) report that annual alcohol advertising placements increased proportionately more than adolescent readership, which suggests targeting of youth. Their Poisson regression model controlled for the number of readers in two other age groups (young adults, older adults) and three additional demographic variables (number of readers by gender, race, and income level). There are several econometric problems with this study. First, none of the explanatory variables vary over time, whereas the three dependent variables – annual number of alcohol advertisements by beverage – do vary with time. Consequently, variables for readership cohorts and demographics can only explain cross-sectional variation in these data, and a set of year dummies explains all of the temporal variation. Second, the study suffers from collinerarity problems due to use of number of readers for several cohorts, which simply reflect the total readership of a magazine. Major magazines, such as Sports Illustrated, have large numbers of youth, young adult, and adult readers. Third, the authors report that they used the Poisson “regression model for the log advertisement rate” (Garfield et al. 2003, p. 2426). It is unclear if they are referring to the Poisson model as a nonlinear regression or they have ignored the discrete nature of the dependent variables and zero counts. It is also unclear if they have handled correctly the duration dependence in the count data by simple inclusion of time dummies. Fourth, using simple graphs, Garfield et al. report that the residuals indicate overdispersion in the Poisson model. This result may mean that there is unobserved heterogeneity of magazines and readers that is not accounted for by the authors’ covariates. In this case, it may be desirable to estimate regression models that relax the equidispersion restriction of the Poisson model, but the authors fail to consider alternative models. The results in Garfield et al. (2003) are tentative at best. Lastly, advertising placement studies were conducted by the advocacy organization, Center on Alcohol Marketing and Youth (CAMY 2002a, 2002b). CAMY’s studies cover magazines and other media, but the studies do not contain any statistical analysis whatsoever. Nevertheless, these studies are
6 important for the role that they played in the NRC report (Jernigan and O’Hara 2004); several class action lawsuits filed against alcohol producers; deceptive advertising petitions to the FTC and TTB; and a recent FTC (2003) report on regulation of alcohol advertising. Using data on advertising exposure for “reach” and “frequency,” CAMY measured gross rating points (GRP) for each of three age groups for 98 magazines in 2001, including 24 magazines that CAMY classified as “youth oriented.” However, GRP is an index of relative exposure of a given group and it ignores the total number of exposures.3 For example, in 2001, Sports Illustrated (SI) had a youth audience (ages 12-20) of about 6.1 million and a total readership of 24.4 million, which means that adult readership was 18.3 million or 75% of the total audience reached by SI. Nevertheless, CAMY classified SI as “youth-oriented” and ignored the larger absolute number of adult readers in the analysis and conclusions; that is, 75% of SI ads reached adults, so these ads cannot be seen by more youth than by adults as was wrongly claimed by CAMY.4 According to the FTC (2003, p. 33), the CAMY data show “the distortions that can arise from considering the GRP data in isolation.” Given the shortcomings of previous placement studies, another examination of alcohol advertising in magazines is in order.5
2. ECONOMETRIC METHODS AND DATA The analysis in this section is based on cumulative placements by beverage in 35 magazines for the five-year period 1997-2001. This approach ignores year-to-year fluctuations in placements, which are not of particular interest, and reduces the number of zeros in the ad counts. More important, it is often argued that advertising has cumulative or lagged effects on adolescent behaviors (Martin 1995),
3
A “point” equals one percent of a potential audience; hence, for a given ad, 50% of a potential audience equals a rating of 50 points. If an ad appeared in more than one issue of a magazine, the points for each issue are added together to yield the audience “reach.” “Frequency” is the number of exposures to a given ad, which varies by media and population group. Gross rating points are the product of reach and frequency. 4
According to CAMY, “youth-oriented” magazines are those where 15.8% or more of the total audience readership (not the paid circulation) is estimated to be between 12-20 years. Hence, widely-popular magazines such as Sports Illustrated, Car and Driver, and Popular Mechanics are classified as “youth-oriented,” even if the subscription is paid for by a parent or the vast number of readers are adults. The category cutoff of 15.8% is based on the 12-20 age cohort as a fraction of the total population (12 years and older), and has no other rational basis for determining the nature of a magazine, its intended readership, or other readership demographics (e.g., gender, race, income). This method of categorizing media appears to have originated with an attempt by the Food and Drug Administration to regulate tobacco advertising; see 61 Federal Register 44396, at 44513 (August 11, 1995). 5
Using a sample of 7th and 8th graders, Tomsen and Rekve (2004) examined the effects of exposure to youth-oriented magazines in three categories: music/entertainment, sports, and men’s lifestyle. The outcome variables were beliefs about teenage drinking, drinking expectancies, and drinking frequency in the past 30 days. Youth-oriented magazines are defined using CAMY’s criteria. However, the three categories of magazines were never examined together. The authors’ results show that alcohol expectancies and drinking frequency were unaffected by reading of music or sports magazines, and they note that the results for men’s lifestyle magazines might reflect selectivity bias. The shortcomings of alcohol expectancy studies are discussed in Nelson (2001).
7 and examination of total placements for a given time period captures this broader concern. The number of ads per magazine (placements) by beverage is from reports by Competitive Media Reporting (e.g., CMR 2000). Data sources and a list of the magazines are reported in the data appendix. Econometric Methods. The econometric procedures are as follows: first, a Poisson regression model is estimated for the number of alcohol advertising placements per magazine (ad counts). This regression method accounts for the skewed distribution of placements. Both zero counts and positive counts are included in these regressions. Statistical tests indicate overdispersion in some Poisson model specifications (i.e., conditional variance greater than the conditional mean). This means that the same data-generating process may not apply to the decision to advertise in a given magazine and the secondstage choice of the number of ads or, alternatively, there is unobserved heterogeneity across magazines that is not accounted for by the explanatory variables. In light of the policy interest in zero values, econometric results are reported for the negative binomial model and the zero-inflated Poisson model, which allow for heterogeneity and overdispersion (Greene 2000, pp. 886-91; Winkelmann 2003). Explanatory Variables. The key explanatory variables in the study are magazine readerships for three age groups, which are obtained from Mediamark Research Inc. (MRI 2003a, 2003b). For adults, the MRI readership data are based on face-to-face interviews and self-reported readership by 13,000 persons. For youth, a mail survey is used (MRI 2003a). MRI takes the readership estimates for each magazine and extrapolates them to the nation. Readers are divided into three age groups: adolescents (12 to 19 years); young adults (20 to 24 years); and older adults (25 years or older). These data represent the audience “reach” of an alcohol advertisement, which is defined as the number of people in a given population group that has the potential to see an advertisement. In addition, demographic data by magazine were obtained from MRI (2003b) for the number of male readers (men); number of black readers (blacks); and number of lower-income readers (no. low income). Additional explanatory variables in the present study are the number of issues per year (no. of issues), median reader age for adults 20 years and older (median age), and median reader income (median income). The results reported below suggest that multicollinearity is reduced by using the median age and income variables. In lieu of information on the page size of ads, the number of issues is included as a proxy variable. The cost of an advertisement per exposure should be roughly constant in a cross-section of magazines, and Garfield et al. (2003, p. 2427) report that their cost measure was not significant. The null hypothesis is that increased youth readership results in an increase in the number of alcohol placements. Expected signs and magnitudes for the other explanatory variables are largely uncertain. In principle, the objective of the study is to “unbundle” the advertising decisions of alcohol producers by examining those influences that systematically increase (or decrease) the frequency of placements.
8 Placement Distributions. Figure 1 shows the skewed distributions of the magazine placement data by beverage for the advertisement counts. Variances are greater than the unconditional means by a ratios of 59 to 435, which suggests overdispersion. The empirical question is whether this overdispersion disappears on inclusion of the regressors. Most magazines contained only a few alcohol ads during the five-year period, but a few magazines had a large number of placements (Sports Illustrated for beer; Sports Illustrated, Rolling Stone, and Entertainment Weekly for spirits; and In-Style for wine).
3. EMPIRICAL RESULTS Multicollinearity in Readership Numbers. As discussed above, the focus of the empirical investigation is the number of adolescent readers. However, because major magazines have a large number of readers in several age groups, this variable is correlated with several other explanatory variables that are based on readership numbers. When collinearity occurs, it is difficult for a multivariate regression to estimate the independent contribution of a regressor to the observed variation in the dependent variable. The consequence is that standard errors are inflated or the empirical results are sensitive to particular data points in the sample or the model specification (Kennedy 2003). I first demonstrate the degree of correlation among the readership variables and then introduce two alternative variables that are less correlated with the number of adolescent readers. Zero-order correlations (rho) among the explanatory variables are shown in Table 1 for each beverage. Examining each beverage is necessary due to the possible restrictions placed on the estimation by the zero placements as well as the controversies that surround the advertising practices of different producers. Table 1 shows several high correlations (rho > 0.50) among the number of readers in each age group, especially between youth and young adults (rho > 0.75). This is not surprising because a popular magazine with a larger readership tends to have a large number of readers in all age groups. The simple correlation between youth and young adults is particularly high, and may illustrate a spillover dilemma faced by regulators; that is, what is read and enjoyed by young adults is also likely to read by youth (see the data appendix). Youth readership also tends to be correlated with readerships for men and lowerincome readers. However, the last two columns show lower correlations when two alternative readership variables are introduced; these are the median age of adult readers and median income of adult readers. None of the correlations in the last two columns are statistically significant. In the regression analysis, I show the effect of replacing the variables for young adult and older adult readers with the median age of adult readers (ages 20+). Because it is important to the policy results, the variable for number of youth is always included in the regressions. I also show the effects of replacing the number of lower-income readers with the variable for the median income of readers.
9 Poisson Regression Results. In Table 2, the dependent variables are total numbers of beer, spirits, or wine magazine advertisements for the period 1997-2001, including zero counts. The estimated coefficients indicate the proportional increase in the number of placements for a one unit increase in an explanatory variable (Cameron and Trivedi 1998, p. 81). More generally, positive coefficients indicate that advertisers tend to favor or “target” magazines with the reader characteristic in question, while negative coefficient indicate the opposite effect. Further, in the Poisson model, the explanatory variables interact in a multiplicative manner. Robust standard errors are used to determine statistical significance at the 95% confidence level, two-tailed test. The results in Table 2 fail to demonstrate that targeting of youth is occurring. For beer and wine, the youth coefficient is never statistically significant, while the significantly positive coefficient for spirits in regression (4) is not robust. Four of nine estimates for youth are negative. Coefficients for young adults are significantly positive for spirits and wine, while the coefficient for older adults is significantly negative for beer. The alternative demographic for adult readers – median age for adult readers – is significantly negative in five of six regressions, but the youth coefficients in these regressions are insignificant.6 Median age and median reader income are significantly positive for wine. Overall, the results indicate that beer and spirits advertisers favor magazines with more readers among younger adults, men, and blacks, but not adolescents. Wine advertisers favor magazines with more adults and higher-income readers, but not adolescents. With regard to collinearity, median age and income of readers provide a better fit for beer as indicated by the log-likelihood values. All of the regressions are statistically significant based on a Wald test, except regression (8) for wine. Tests for Overdispersion in Table 2. A well-know problem with the Poisson model is the presumed equality of the conditional mean and variance functions (equidispersion). This restriction may not hold due to unobserved heterogeneity or because the zero outcomes of the data-generating process are quantitatively different from the positive outcomes. The proportion of zeros in the placement data are 22% for beer; 11% for spirits; and 31% for wine. It also is conceivable that different models apply to different magazines and beverages, and this is reflected in different proportions of zeros in the sample. For example, it is not clear if a zero placement arises because advertisers did not happen to use a particular magazine during the study period or because that magazine would never be chosen for a particular beverage (e.g., a zero count for wine ads in Popular Mechanics). Additional tests may be required using alternative econometric models.
6
As an experiment, I replaced the number of youth readers with the percent of total readers who are youth. This change did not alter the youth results in the median age regressions for either beer or wine. The youth coefficients for spirits were negative and significant, which does not support targeting of youth.
10 Using the results in Table 2, two tests for overdispersion were conducted. The regression-based test due to Cameron and Trivedi (1990) is based on the actual residuals, while the test suggested by Wooldridge (1996) is based on the standardized residuals. For beer advertising, overdispersion was rejected for all three of the regressions. For spirits, overdispersion was found for all three regressions using Wooldridge’s test, but not using Cameron and Trivedi’s test. The overdispersion results were sensitive to the treatment of the error terms, with robust standard errors tending to lead to a rejection of the null of overdispersion. For wine, overdispersion was found for regressions (7) and (9) using Cameron and Trivedi’s test, but not using Wooldridge’s test. In light of these mixed findings, it is desirable to examine the assumptions of the data-generating process by estimating the negative binomial model and the zero-inflated Poisson model. The negative binomial model relaxes the presumed equality of the mean and variance functions and allows for between-observation heterogeneity. The zero-inflated Poisson model relaxes the equidispersion assumption and allows the zero counts to be generated by a binary selection process (Winkelmann 2003). Alternative Econometric Estimates. Table 3 displays the estimates for negative binomial and zero-inflated Poisson regressions. Two alternative specifications are reported for spirits in light of the positive youth coefficient found in Table 2. None of the youth coefficients in Table 3 are statistically significant, including those for distilled spirits. In other respects, the results are similar to Table 2.7 Median reader age for adults is significantly negative for beer and spirits, while median age is significantly positive for wine in regression (4). Male readership is significantly positive for beer and spirits, while median income is significantly positive for wine. Black readership is significantly positive for all beverages in the negative binomial regressions, but not in the zero-inflated Poisson regressions. The regressions fail to demonstrate targeting of youth by magazine advertisers. The alpha overdispersion parameters in the negative binomial model are given in the last row of Table 3, which suggest overdispersion in the Poisson model (Cameron and Trevidi 1998, p. 79). The undeflated Poison model is nested within the negative binomial model (Winkelmann 2003, p. 104). Using comparable results in Tables 2 and 3, a likelihood ratio (LR) test strongly rejects the Poisson model in favor of the negative binomial model. The LR test statistics are 496.4 (beer), 3800 (spirits), and 631.7 (wine). The critical value of the chi-square distribution with one degree of freedom is only 50.9 at the 99% confidence level. The Poisson model does not nest the zero-inflated Poisson model. Vuong’s (1989) test for nonnested models can be applied, and the test statistic has a standard normal distribution
7
The full complement of specifications in Table2 were estimated for both models, but the results parallel those in Table 2 and are not reported here. The zero-inflated negative binomial model was not concave for wine and these results are omitted for all beverages. Excluding the zero placement counts does not change my results.
11 (Greene 2000, p. 891). Applying this test yields statistics with positive values of 2.03 (beer), 0.46 (spirits), and 2.17 (wine). Values less than 2.0 do not favor one model over another, which is the case for spirits. The values for beer and wine are close to 2.0, suggesting that the preferred model is the negative binomial. In any event, the primary finding is that the null hypothesis of non-targeting of youth cannot be rejected in these data.
4. CONCLUSIONS Based on the results in this study, beer and spirits producers tend to advertise in magazines with more young adult readers (ages 20-24), men, and blacks. Wine advertisers tend to target older readers (age 25+) and higher-income readers. Adolescent readers are not important statistically for any beverage. This result holds for all beverages, including beer, which is generally regarded as the alcohol beverage that is preferred by youth. The study by Garfield, et al. (2003) wrongly concluded that beer and spirits producers were targeting adolescent readers through magazine placements. This conclusion does not withstand analysis that accounts for the collinearity that exists among the explanatory variables for numbers of readers, the lack of time variation in the covariates, and overdisperision of the residuals. All three econometric models failed to provide evidence that supports the criteria of the Central Hudson test. The results in the present study do suggest a dilemma for policymakers: major magazines have large numbers of readers in several demographic groups, including youth.8 However, the relationship between the number of readers and the percent of readers in each age group will vary by magazine. At the request of the U.S. Congress, two recent reports by the Federal Trade Commission (FTC 1999, 2003) examined the ability of the alcohol industry to self-regulate its advertising practices. Through 2002, most of the industry followed a standard that at least 50% of an advertisement’s audience should consist of adults aged 21 and over. This limit should be reflected in the empirical results above. However, as noted by the FTC (2003), this standard still allows alcohol ads to reach substantial numbers of youths. The main outcome of the FTC investigation was a suggestion that the industry should limit its ads to media with at least 70% adults in the audience, which would prohibit alcohol ads in many major magazines with large numbers of adult readers. Because the relationship between alcohol consumption and advertising is non-existent or weak at best, the dilemma posed by this partial ban is that costs of regulation are incurred without any obvious benefits in terms of reductions in underage alcohol consumption. 8
In Bolger (463 U.S. 60), the First Amendment issue was a Postal Service regulation that prohibited unsolicited mailings of advertisements for contraceptives. Speaking for the Court, Justice Marshall reaffirmed that “the government may not reduce the adult population . . . to reading only what is fit for children” (463 U.S. 60, at 73). He went on to discuss the marginal degree of protection provided by the advertising ban and the benefits denied to adults and adolescents alike from greater amounts of information.
12 References Atkin, C. and M. Block (1981), Content and Effects of Alcohol Advertising, Report of Michigan State University Study, NTIS No. PB82-123142 (Washington, DC: Bureau of Alcohol, Tobacco, and Firearms). Breed, W. and J.R. DeFoe (1979), “Themes in magazine alcohol advertisements: A critique,” Journal of Drug Issues, 9 (4), 511-22. Calfee, J.E. (1997), Fear of Persuasion: A New Perspective on Advertising and Regulation (Monnaz, CH: Agora and AEI Press). Cameron, A.C. and P.K. Trivedi (1990), “Regression-based tests for overdispersion in the Poisson model,” Journal of Econometrics, 46 (December), 347-64. Cameron, A.C. and P.K. Trivedi (1998), Regression Analysis of Count Data (Cambridge: Cambridge University Press). Center on Alcohol Marketing and Youth (2002a), Overexposed: Youth a Target of Alcohol Advertising in Magazines (Washington, DC: CAMY). URL http://camy.org/research/mag0902. Center on Alcohol Marketing and Youth (2002b), Television: Alcohol’s Vast Adland (Washington, DC: CAMY). URL http://camy.org/research/tv1202. Competitive Media Reporting (2000), Beverage Advertising by Medium, 1992-1999. URL http://www.tnsmi-cmr.com/products/index.html Denning, B.P. (2002), “Smokey and the Bandit in cyberspace: The dormant Commerce Clause, the Twenty-First Amendment, and state regulation of internet alcohol sales,” Constitutional Commentary, 19 (2), 297-343. Douglass, D.B. (2000), “Constitutional crossroads: Reconciling the Twenty-First Amendment and the Commerce Clause to evaluate state regulation of interstate commerce in alcoholic beverages,” Duke Law Journal, 49 (6), 1619-62. Federal Trade Commission (1985), Omnibus Petition for Regulation of Unfair and Deceptive Alcoholic Beverage Advertising and Marketing Practices (Docket No. 209-46) – Appendix A: Alcohol Advertising, Consumption, and Abuse (Washington, DC: FTC). Federal Trade Commission (1999), Self-Regulation in the Alcohol Industry: A Review of Industry Efforts to Avoid Promoting Alcohol to Underage Consumers – A Report to Congress (Washington, DC: FTC). URL http://www.ftc.gov/reports/alcohol/alcoholreport.htm. Federal Trade Commission (2003), Alcohol Marketing and Advertising – A Report to Congress (Washington, DC: FTC). URL http://www.ftc.gov/os/2003/09/alcohol08report.pdf. Foust, J. (2000), “State power to regulate alcohol under the Twenty-First Amendment: The constitutional implications of the Enforcement Act,” Boston College Law Review, 41 (3), 659-97.
13 Garfield, C.F., Chung, P.J., P.J. Rathouz (2003), “Alcohol advertising in magazines and adolescent readership,” Journal of the American Medical Association, 289 (May 14), 2424-29. Greene, W.H. (2000), Econometric Analysis, 4th edn. (Upper Saddle River, NJ: Prentice-Hall). Grube, J.W. (2004), “Alcohol in the media: Drinking portrayals, alcohol advertising, and alcohol consumption among youth,” in National Research Council, Reducing Underage Drinking: A Collective Responsibility (Washington, DC: National Academies Press), pp. 597-624. Jernigan, D. and J. O’Hara (2004), “Alcohol advertising and promotion,” in National Research Council, Reducing Underage Drinking: A Collective Responsibility (Washington, DC: National Academies Press), pp. 625-53. Kennedy, P.E. (2003), A Guide to Econometrics, 5th edn. (Cambridge, MA: MIT Press). Martin, S.E. (1995), The Effects of the Mass Media on the Use and Abuse of Alcohol (Washington, DC: NIAAA). Mediamark Research Inc. (2003a), The MRI TEEN Study, Teenmark 2000 Technical Guide. URL http://www.mediamark.com/mri/techguide/tg_tm00.htm. Mediamark Research Inc. (2003b), Pocketpiece Top-Line Reports. URL http://www.mriplus.com. National Highway Traffic Safety Administration (2003), Alcohol Involvement in Fatal Crashes, 2001, DOT-HS-809-579 (Washington, DC: NHTSA). National Institute on Alcohol Abuse and Alcoholism (2000), “Alcohol advertising: What are the effects?” in 10th Special Report to the U.S. Congress on Alcohol and Health (Washington, DC: NIAAA), pp. 412-26. \National Research Council, Institute of Medicine (2004), Reducing Underage Drinking: A Collective Responsibility (Washington, DC: National Academies Press). Nelson, J.P. (2001), “Alcohol advertising and advertising bans: A survey of research methods, results, and policy implications,” in M.R. Baye and J.P. Nelson (eds.), Advances in Applied MicroEconomics: Advertising and Differentiated Products (Amsterdam: JAI Press), pp. 239-95. Nelson, J.P. (2003), “Advertising bans, monopoly, and alcohol demand: Testing for substitution effects using state panel data,” Review of Industrial Organization, 22 (1), 1-25. Nelson, J.P. (2004), “Advertising bans, U.S.,” in R. Whaples (ed.), EH.Net Encyclopedia, pp. 1-29. URL http://www.eh.net/encyclopedia/?article=Nelson.AdBans Nelson, J.P. (2005), “Beer advertising and marketing update: Structure, conduct, and social costs,” Review of Industrial Organization (forthcoming). Nelson, J.P. and D.J. Young (2001), “Do advertising bans work? An international comparison,” International Journal of Advertising, 20 (3), 273-96.
14 Saffer, H. (1991), “Alcohol advertising bans and alcohol abuse: An international perspective,” Journal of Health Economics, 10 (May), 65-79. Saffer, H. (1993), “Advertising under the influence,” in M.E. Hilton (ed.), Economics and the Prevention of Alcohol-Related Problems (Washington, DC: NIAAA), pp. 125-40. Sanchez, L., et al. (2000), “Tobacco and alcohol advertisements in magazines: Are young readers being targeted?” Journal of the American Medical Association, 283 (April 26), 2106-07. Strickland, D.E. (1984), “Content and effects of alcohol advertising: Comment on NTIS Pub. No. PB82123142,” Journal of Studies on Alcohol, 45 (January), 87-100. Strickland, D.E., T.A. Finn, and M.D. Lambert (1982), “A content analysis of beverage alcohol advertising: Magazine advertising,” Journal of Studies on Alcohol, 43 (July), 655-82. Thomsen, S.R. and D. Rekve (2004), “The differential effects of exposure to ‘youth-oriented’ magazines on adolescent alcohol use,” Contemporary Drug Problems, 31 (Spring), 31-58. U.S. Bureau of the Census (2003), Statistical Abstract of the United States, 2003, 123rd edn. (Washington, DC: USGPO). Vuong, Q.H. (1989), “Likelihood ratio tests for model selection and non-nested hypotheses,” Econometrica, 57 (2), 307-33. Winkelmann, R. (2003), Econometric Analysis of Count Data, 4th ed. (Berlin: Springer). Wooldridge, J.M. (1997), “Quasi-likelihood methods for count data,” in M.H. Pesaran and P. Schmidt (eds.), Handbook of Applied Econometrics, vol. 2 (Oxford: Blackwell), pp. 352-406.
15 Data Appendix Variable
Description and Data Source
Advertising placement counts for beer, spirits, and wine and wine coolers
Cumulative number of alcohol advertisements per magazine for the years 1997-200; Beer ads include favored malt beverages and wine ads include wine coolers; Competitive Media Reporting and Garfield et al. (2003)
Readers by age group
Number of readers in millions for the age groups 12-19 years, 20-24 years, and 25 years and more; Mediamark Research Inc. and Garfield et al. (2003)
Median age of adult readers
Median age of adult readers (20+ years) in 1999; Mediamark Research Inc., Pocketpiece Reports (Fall 1999)
Men readers
Number of adult male readers in millions; Mediamark Research Inc. and Garfield et al. (2003)
Black readers
Number of adult black readers in millions; Mediamark Research Inc. and Garfield et al. (2003)
Lower-income readers
Number adult readers with annual household incomes of less than $30,000 in 1999. Mediamark Research Inc. and Garfield et al. (2003).
Median household income of readers
Median household income of readers in 1999; Mediamark Research Inc., Pocketpiece Reports (Fall 1999)
Issues per year
Number of annual issues per magazine; author’s count based on Mediamark Research Inc., Pocketpiece Reports (Fall 1999) List of magazines (Y indicates estimated youth readership (ages 12-19) is 14% or more): Allure (Y), Better Homes and Gardens, Car and Driver (Y), Cosmopolitan (Y), Ebony (Y), Elle (Y), Entertainment Weekly (Y), Essence (Y), Family Circle, Field and Stream, Glamour (Y), Good Housekeeping, In Style (Y), Jet (Y), Life (Y), Maria Claire (Y), Motor Trend (Y), National Geographic (Y), Newsweek (Y), People Weekly, Popular Mechanics (Y), Popular Science (Y), Reader’s Digest, Road and Track (Y), Rolling Stone (Y), Soap Opera Digest (Y), Spin (Y), Sport (Y), Sporting News (Y), Sports Illustrated (Y), Time, TV Guide (Y), Vibe (Y), Vogue (Y), and Women’s Day.
16
Table 1. Zero-order correlations among selected explanatory variables Youth & Young Adults
Youth & Adults
Young Adults & Adults
Youth & Males
Youth & Blacks
Youth & LowIncome
Youth & Median Age
Youth & Median Income
Beer
0.796*
0.584*
0.498
0.768*
0.348
0.643*
0.160
-0.278
Spirits
0.782*
0.714*
0.664*
0.773*
0.326
0.769*
0.298
-0.208
Wine
0.763*
0.459
0.358
0.821*
0.197
0.470
0.089
-0.219
Beverage
All 0.782* 0.604* 0.536* 0.798* 0.350 0.634* 0.210 -0.189 Notes: Asterisks indicate statistically significant correlations at the 99% level; critical values are beer, 0.529; spirits, 0.494; wine, 0.561; and all, 0.465. Different samples by beverage reflect exclusion of zero counts, except in the last row. Youth variable is the estimated number of adolescent readers (ages 12-19) for a magazine; young adults is number of readers ages 20-24 years; adults is number of readers ages 25+ years; males is number of male readers; blacks is number of black readers; low income is number of readers with annual household income of less than $30,000; median age is average age of readers 20 years and older; and median income is average household income of adult readers in 1999 dollars.
17 Table 2. Poisson regression analysis of alcohol advertisements counts (dependent variables: Magazine advertising placements by beverage, zero counts included; z statistics shown below parameter estimates) (1) Beer
(2) Beer
(3) Beer
(4) Spirits
(5) Spirits
(6) Spirits
(7) Wine
(8) Wine
(9) Wine
Constant
1.910 (3.10)*
8.991 (6.22)*
11.75 (6.77)*
3.635 (8.31)*
9.623 (5.44)*
10.74 (6.11)*
1.848 (3.48)*
2.411 (1.11)
-5.852 (2.46)*
Youth (no.)
0.416 (1.36)
0.032 (.100)
-0.120 (.620)
0.397 (2.04)*
0.047 (.200)
-0.166 (.760)
-0.013 (.030)
-0.029 (.070)
0.347 (1.00)
Young adult (no.)
0.227 (1.42)
---
---
0.895 (6.82)*
---
---
1.000 (4.64)*
---
---
Adults (no.)
-0.288 (2.38)*
---
---
0.123 (1.73)
---
---
0.774 (3.37)*
---
---
Median age (yrs)
---
-0.225 (5.11)*
-0.246 (8.19)*
---
-0.164 (3.11)*
-0.213 (5.26)*
---
-0.004 (.050)
0.106 (2.80)*
Men (no.)
0.164 (1.89)
0.218 (2.93)*
0.254 (4.25)*
-0.023 (.420)
0.150 (2.47)*
0.163 (2.46)*
-0.383 (1.81)
-0.028 (.250)
-0.151 (1.32)
Blacks (no.)
0.004 (.040)
0.214 (2.33)*
0.129 (1.06)
0.236 (2.40)*
0.142 (2.23)*
0.134 (1.76)
0.301 (2.06)*
-0.127 (1.18)
0.043 (.320)
No. low income
0.412 (1.17)
-0.063 (.620)
---
-1.048 (3.52)*
-0.173 (1.26)
---
-2.516 (3.28)*
0.094 (.550)
---
Median income
---
---
-0.035 (1.39)
---
---
0.008 (.680)
---
---
0.065 (3.44)*
No. of issues
0.009 (.880)
0.019 (1.92)
0.019 (2.12)*
0.026 (2.67)*
0.026 (3.14)*
0.030 (3.85)*
0.122 (1.21)
0.015 (.900)
0.024 (2.03)*
pseudo-R2
0.684
0.723
0.740
0.717
0.640
0.624
0.546
0.038
0.252
Variable
Log L -458.8 -402.3 -378.1 -1580 -2004 -2099 -251.3 -532.4 -413.9 Notes: Dependent variables are total number of advertisements during the period 1997-2001, with zero observations included. Poisson estimates obtained using Intercooled Stata 8.2. Z-statistics in parentheses are based on robust standard errors; an asterisk indicates a statistically significant coefficient at the 95% level, two-tailed test. All regressions are statistically significant based on a Wald chi-square test, except regression (8). Readership numbers (no.) are expressed in millions of people. Median income is in thousands of 1999 dollars. Log L is the log-likelihood value.
18 Table 3. Count data regression analysis of alcohol advertisements (dependent variables: Magazine advertising placements by beverage, zeros included; z statistics shown below parameter estimates) Model
----------- Negative Binomial -----------
-------- Zero-Inflated Poisson --------
(1) Beer
(2) Spirits
(3) Spirits
(4) Wine
(5) Beer
(6) Spirits
(7) Spirits
(8) Wine
Constant
11.07 (5.81)*
12.65 (5.56)*
1.211 (.650)
-16.85 (4.49)*
11.27 (7.79)*
10.53 (6.01)*
3.150 (3.87)*
-4.101 (1.08)
Youth (no.)
-0.287 (1.24)
-0.554 (1.56)
0.050 (.190)
0.814 (1.91)
-0.096 (.610)
-0.162 (.740)
0.161 (.750)
0.494 (1.18)
Young Adult (no.)
---
---
0.865 (3.00)*
---
---
---
0.639 (3.89)*
---
Adults (no.)
---
---
-0.224 (4.65)*
---
---
---
-0.140 (4.00)*
---
Median age (yrs)
-0.228 (5.40)*
-0.340 (3.83)*
---
0.160 (3.04)*
-0.211 (7.89)*
-0.202 (4.83)*
---
0.096 (1.75)
Men (no.)
0.273 (2.97)*
0.329 (2.25)*
0.169 (1.76)
-0.323 (2.93)*
0.229 (5.00)*
0.163 (2.46)*
0.048 (.910)
-0.124 (.950)
Blacks (no.)
0.192 (2.51)*
0.313 (2.42)*
0.186 (2.12)*
0.420 (2.06)*
0.045 (.380)
0.125 (1.70)
0.031 (.420)
-0.145 (.750)
Median income
-0.030 (1.63)
0.053 (1.17)
0.046 (1.48)
0.200 (4.77)*
-0.039 (1.95)
0.007 (.610)
0.017 (1.45)
0.048 (2.10)*
No. of issues
0.018 (1.10)
0.029 (2.89)*
0.026 (2.60)*
0.027 (2.07)*
0.018 (2.19)*
0.027 (3.31)*
0.024 (2.68)*
0.016 (1.30)
pseudo-R2
0.085
0.065
0.072
0.103
na
na
na
na
Log L
-129.9
-199.1
-197.7
-98.07
-269.9
-2026
-1970
-307.1
Variable
alpha 1.584 1.196 1.102 1.718 --------(s.e.) (.507) (.288) (.244) (.568) Notes: Dependent variables are total number of advertisements during the period 1997-2001, with zero observations included. Poisson estimates obtained using Intercooled Stata 8.2. Z-statistics in parentheses are based on robust standard errors; an asterisk indicates a statistically significant coefficient at the 95% level, two-tailed test. All regressions are statistically significant based on a Wald chi-square test. Readership numbers (no.) are expressed in millions of people. Median income is in thousands of 1999 dollars. Log L is the log-likelihood value and alpha is the overdispersion parameter in the negative binomial model.
19
Figure 1. Distribution of Alcohol Advertisement Counts in Major Magazines, 1997-2001
30 S e rie s : B E E R A D S
C o un ts 25 20 15
Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis
34.34286 5.000000 367.0000 0.000000 70.87103 3.444242 15.54153
Jarque-Bera Probability
298.5807 0.000000
10 5
To ta l A d s p e r M a g a z in e
0 0
50
100
150
200
250
300
350
28 S e r ie s : W IN E A D S
C o un ts 24 20 16 12 8 4
Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis
12.08571 2.000000 108.0000 0.000000 23.63742 2.580516 9.563047
Jarque-Bera Probability
101.6602 0.000000
To t a l A d s p e r M a g a z in e
0 0
20
40
60
80
100
20 C o u nts
S e rie s : S P IR IT S AD S
8
Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis
214.4000 126.0000 1470.000 0.000000 305.4086 2.517676 9.901815
4
Jarque-Bera Probability
106.4435 0.000000
16
12
To ta l A d s p e r M a g a z in e
0 0
200
400
600
800
1000
1200
1400
