Alcohol advertising in magazines: do beer, wine, and spirits ads target youth?
In 2003, the alcohol beverage industry spent more than $1.6 billion on advertising in measured media outlets, including $394 million on ads placed in magazines. Industry critics allege that these activities intentionally target adolescent audiences and thereby contribute importantly to social problems associated with underage alcohol consumption (Center on Alcohol Marketing and Youth [CAMY] 2005b; Center for Science in the Public Interest 2002). The favored regulatory approach has been to advocate a placement standard based on the youth audience expressed as a proportion of the total audience. For example, the CAMY (2002, 2005a) claims that advertisements in media outlets that reach audiences with more than 15% underage youth result in "overexposure" to alcohol ads. Two reports by the Federal Trade Commission (FTC 1999, 2003) advocate a placement threshold of 25% as a best practice response by alcohol companies, and a report by the National Research Council's Institute of Medicine (NRC 2004, p. 138) argues that the industry should move toward a 15% threshold for television and 25% for other media. Other organizations, such as the American Medical Association (2004), support a total statutory ban of alcohol advertising except for ads placed inside of retail and wholesale outlets.
Following the FTC's 1999 report, the major companies in the alcohol industry collectively altered their self-regulatory advertising codes and media buying guidelines: Beer Institute's Advertising and Marketing Code; Wine Institute's Code of Advertising Standards; and the Distilled Spirits Council of the United States (DISCUS), Code of Responsible Practices for Beverage Alcohol Advertising and Marketing. The Wine Institute amended its code in 2000 to adopt a 70% adult placement standard. In October 2003, the beer and spirits codes were amended to require that adults constitute at least 70% of the audience for TV, radio, and magazine advertisements, which represents an increase from the previous 50% adult standard (FTC 2003). Furthermore, the revised beer and spirits codes require that industry members conduct postplacement audits, including a third-party review system for controversial beer and spirits advertisements (DISCUS 2005).
Do alcohol advertisements target underage youth? The evidence on the affirmative side is based largely on a series of descriptive reports commissioned by an advocacy group, the CAMY (2002, 2005a). CAMY's studies measure the youth audience as a percent of the total audience for different alcohol brands and media outlets, which are aggregated to obtain measures of advertising exposure per capita for youth and adults. For magazines, CAMY's measures of gross rating points (GRPs) account for an advertisement's frequency and reach (audience composition), but fail to account for audience size. Because underage youth constitute about 15% of the total population, CAMY characterizes any audience containing more than 15% adolescents as youth-oriented. This designation is used regardless of other aspects of placement decisions, such as the number of adults in the audience or the number of adult readers per copy (FTC 2003, p. 32). Further, CAMY's studies are descriptive and based on the simplistic notion that targeting occurs whenever the 15% threshold is exceeded. Magazines such as Popular Mechanics and Sports Illustrated with 17% and 25% youth readership, respectively, are characterized as youth-oriented despite other features of the audience and magazine (CAMY 2002a). In addition to audience size, advertising content and costs are ignored by GRP-based measures.
Analytical evidence on youth exposure to alcohol advertisements in magazines is provided by two recent regression studies. Garfield et al. (2003) examined the occurrence (counts) of annual alcohol advertising placements for 35 major magazines that tracked youth readerships during 1997-2001. Ad counts that are 0 were apparently excluded. Using a Poisson model, they regressed the count of annual ads in each magazine on a set of demographic variables, including the number of youth readers (ages 12-19), number of young adults (ages 20-24), number of adults (ages 25+), number of male readers, number of black readers, number of low-income readers, and year dummy variables. Because popular magazines tend to have a large number of readers in all categories, many of these variables are highly correlated. In addition, the explanatory variables were measured for 1999 only and did not contain any temporal variation. Consequently, the results for youth readers can only capture cross-sectional differences in the ad counts. Garfield et al. (2003, p. 2428) concluded that magazine ads for beer and spirits were associated positively with adolescent readership, and at a minimum, indirect targeting of youth was occurring. In a second study, Nelson (2005) used the cumulative data in Garfield et al. to examine cross-sectional features of their model, collinearity among the readership demographics, zero counts, and overdispersion of the Poisson residuals (excess zero counts). Using alternative estimation methods and new variables for average age and income of adult readers, Nelson (2005) concluded that targeting of youth was not occurring for beer, wine, or distilled spirits.
The present study seeks to expand on the results in these two studies. First, the advertising count data cover a more recent time period of 2001-03, which includes the last year of beer and spirits advertising under the old 50% placement standard. Zero counts are included in the analysis. Second, the explanatory variables vary across magazines and over time and include the temporal variation that was missing in both previous studies. Third, the data set allows examination of explanatory variables that were ignored in previous studies, including measures of audience size, magazine sales outlets, and standardized costs of advertisements across magazines and time. In particular, the study demonstrates the importance of each magazine's real advertising cost per 1000 copies (CPM) in circulation as a variable affecting placements. This variable is the advertising industry's measure of magazine cost efficiency, and is available for standardized advertisements such as a full-page four-color ad (P4C). A cost variable (nonstandardized) was discarded as insignificant by Garfield et al. and was unavailable in Nelson. The empirical results for price help clarify some of the earlier findings, such as placement of ads in magazines with predominantly African American readers (e.g., Ebony, Jet, and Vibe magazines). Fourth, following the emphasis in the regulatory literature, the article focuses on the percent of youth in the audience as an explanatory variable, which was largely ignored by previous studies in favor of the absolute number of youth. As pointed out by Nelson, popular magazines tend to have a large number of readers in all age groups, which leads to collinearity in readership numbers as well as difficulties in formulating regulatory standards (see NRC 2004, p. 139). Fifth, following Nelson (2005), both Poisson and negative binomial count models are estimated and compared.
The remainder of the article is divided into five sections. Section II describes the model and selected aspects of the data. Section III presents the econometric results for Poisson and negative binomial regressions for total ad counts, including specification tests for overdispersion. Section IV considers the marginal importance of the explanatory variables, including audience size and price elasticities. Section V examines beer and spirits ads separately, which increases the number of zero counts in the analysis. Section VI contains the conclusions and discusses the policy implications of the study.
II. MODEL AND DATA
Previous econometric studies of media placements analyzed advertising economies of scale at the brand or industry level (Bresnahan 1984; Seldon et al. 2000), intermedia choices at the brand and industry level (Fare et al. 2004; Seldon and Jung 1993; Silk et al. 2002), and the price of advertising (Depken 2004; Depken and Wilson 2001; Koschat and Putsis 2000, 2002). None of these studies estimated a demand function for media. The present study estimates a model of the demand for media space across magazines and time, conditional on reader demographics, magazine characteristics, and real price of a standardized advertisement.
Assume that advertisers' demand for media space is derived from consumers' demand for information about the existence and attributes of products and brands, including information that is persuasive in nature (Ehrlich and Fisher 1982). Assume also that the advertiser has solved the problem of media mix and must next decide on the choice of space in available magazines. Magazines can be described in terms of various characteristics of the readers (age, gender, race, income), characteristics of the magazine (subject matter, paid circulation, audience size, single-copy sales, number of issues), and the magazine's price for a standardized advertisement, for example, the cost of a P4C advertisement. Because advertising is provided jointly with the magazine's subject matter and magazines also contain numerous ads, there is considerable "clutter" or noise in the information process. A number of specialized services exist to collect, verify, and provide data about readers and magazines to both publishers and advertisers, which implies that the advertisers attempt to reduce the noise in the information process through placement or targeting decisions. Such data are typically proprietary but available to the public on a limited basis.
Specifying the demand function as a count model leads to the following equation for the expected number of occurrences (counts) of alcohol ads, [N.sub.it], placed in the ith magazine in year t:
(1) E([N.sub.it]) = exp([X'.sub.it][beta] + [Z'.sub.it][theta] + [alpha][P.sub.it] + [delta]ln(issue[s.sub.i])),
where X is a vector of reader demographics, Z is a vector of magazine characteristics, P is the real CPM for a P4C advertisement, and [beta], [theta] and a represent the coefficients. Holding incidence rates constant, weekly magazines have more annual alcohol advertisements than monthly magazines. Equation (1) treats the number of annual issues of each magazine as the "exposure" variable, which implies that the elasticity coefficient [delta] should be close to unity (Cameron and Trevedi 1998, p. 81). For count data, the Poisson model offers a number of advantages, but distribution plots suggested that the negative binomial might be more appropriate (Winkelmann 2003, p. 32). Following Nelson (2005), econometric results and tests are reported for both models. The price variable is identified by the existence of different real prices for a standardized advertisement, reflecting real changes over time for a given magazine and differences across magazines that reflect unobserved costs of supply that apply to all advertisers, including alcohol advertisers. Audience size is measured by readers per copy and is a measure of the marginal benefits of advertising that should be important for placement decisions. The main hypothesis in the article concerns the sign and significance of the variable for the percent of youth readers. The null hypothesis is that alcohol advertisers do not target youth, which means that the regression coefficient in (1) for youth readership should be insignificantly different from 0. According to Garfield et al. (2003, p. 2428), absent explicit evidence of intent, targeting occurs whenever a group is reached in a measurable or material manner. They argue that their significant results for a youth demographic variable demonstrate targeting of adolescent readers. The present article offers a test of the robustness of the conclusions in Garfield et al. The test also is consistent with U.S. Supreme Court interpretations of First Amendment protections provided to commercial speech under the so-called Central Hudson doctrine (Central Hudson Gas & Electric Corp. v. Public Service Commission of New York, 447 U.S. 557 ).. (1)
A. Variable Definitions and Data Sources
The sample consists of 28 major magazines for the time period 2001-03. Table 1 reports the data sources and definitions for the variables used in the regressions. Empirical results are reported in section III for four demographic variables (percent youth readers, adult median age, adult median real income, percent adult male readers) and five magazine characteristics (real CPM price, percent single-copy sales, adult readers per copy, square of readers per copy, annual issues). The annual number of issues for each magazine is a measure of exposure in the count model, that is, the expected ad count per year is the product of an incidence rate per issue and the level of exposure as measured by the annual number of issues. Because the number of alcohol ads can change for reasons external to the magazine market, some regressions also include year fixed-effects dummies. These variables control for changes in the prices of other media and changes in the total amount of alcohol advertising contained in other media. Due to possible discounting from published price lists, such as during the 2001 recession, the year dummies also permit a stronger test of the importance of price in placement decisions. According to the Statistical Abstract, total advertising revenues in magazines in 2003 were $18.3 billion, including $394 million in the alcohol category, or only 2.2% of the total. Hence, the explanatory variables, including the CPMs, should be exogenous to advertising choices by alcohol producers, reflecting decisions made by all advertisers.
Table 2 summarizes selected aspects of the data set, including the cumulative numbers of ads for 2001-03; percent of youth in each magazine's audience in 2003; rate base circulation used in the CPMs; real CPM-P4C in dollars; and the audience size. The number of alcohol ads for each magazine was drawn from CAMY (2005a, p. 20), which covers annual count data on alcohol ads in 124 consumer magazines for the years 2001-03. The ad counts reflect advertising placements for all three alcohol beverages; that is, ads are combined for beer, wine, and distilled spirits. Reflecting constraints on broadcast advertising, CAMY's magazine counts are dominated by distilled spirits advertisements. For 2003, CAMY (2005a, p. 5) reported 495 magazine ads for beer, 417 ads for wine, and 2,330 spirits ads (72% of the total). In 2003, total spending on magazine ads was beer, $70.6 million; wine, $52.9 million; and spirits, $271.0 million (69%). For the period 2001-03, the 28 magazines in my sample contained a total of 3,675 alcohol ads, including 652 beer ads, 118 wine ads, and 2,905 distilled spirits ads (79%). Hence, the dispersion of ads by beverage is reasonably representative of industry practices. Many magazines have very few youth readers, and CAMY (2005a, p. 10) selectively examines data for 21 magazines with a "disproportionately" high youth readership, that is, the percent of adolescent readers exceeds 15%. For purposes of the present study, the online report MRI+ Pocketpiece Magazine (Teen) tracks annual data on youth readerships for 28 magazines, including 14 of the magazines on CAMY's overexposed list.2 MRI defines youthful readers as ages 12-19, who make up about 13.7% of the total population. As shown in Table 2, the youth audience in 2003 ranged from 4.7% for Better Homes and Gardens to 33.3% for The Source magazine. The 28 magazines in the sample are a small fraction of the total number of magazines. Although exact comparisons are difficult, if anything, the sample is biased toward a positive relationship for youth readers. All of the sampled magazines accept alcohol ads, which is not the case for some popular youth-oriented magazines; for example, Seventeen, Teen, and YM do not accept alcohol ads. Four magazines in the sample are among the most widely read magazines among adolescents (People Weekly, Rolling Stone, Sports Illustrated, Vibe). The percent of youth readers equals or exceeds 20% for 14 of 28 magazines in the sample.
Standardized prices for advertisements are available publicly from the SRDS Consumer Magazine Advertising Source. In Table 2, the CPMs range from $29.64 per thousand circulation for Jet magazine to $114.33 for Road & Track. The low prices for ads in black magazines explain some of the ad placements in this category. The last column in the Table 2 is the estimated number of adult readers per copy, which is an industry measure of audience size. This variable reflects the pass-along rate per copy, because it measures the number of adult readers compared to the paid circulation. Other things being equal, advertisers prefer magazines that reach a larger audience. For example, in 2003, Sports Illustrated had a paid circulation of 3.262 million and an estimated adult audience of 20.12 million. The average number of adult readers per copy is therefore 6.17. This size variable is calculated for each magazine for each year, and varies from 3.33 for Self magazine in 2001 to 15.8 for The Source in 2003. The square of readers per copy is used to capture nonlinearity in this variable.
B. Content Categories
Table 3 displays summary data on six magazine content categories: automobiles; black; men's style and sports; women's style; entertainment and music; and general and other magazines. In the next section, several model specifications are estimated that contain dummy variables for these content categories. This specification captures panel features of the data and allows additional tests of alcohol placement decisions that often are the subject of criticism, such as alcohol ads in conjunction with sports or automobiles. Using data for 2003, the automobile category has the highest average youth readership percentage, although the number of alcohol ads in this category is quite small. The entertainment and music category has the second highest youth percentage, the lowest mean adult age, and a large number of alcohol ads. This category also has a high pass-along rate as shown by the value of 8.8 for the mean number of adult readers per copy. Average adult reader income is lowest for black magazines and highest for the general and other category. In 2003, the average ad price was lowest for black magazines and highest for automobile magazines.
III. EMPIRICAL RESULTS
Table 4 displays the regression results for the Poisson and negative binomial models. The results for the Poisson model allow comparison with earlier studies by Garfield et al. (2003) and Nelson (2005). Three alternative specifications are estimated for each model. First, regressions (1), (4), and (5) omit the year dummies. These regressions include four demographics, three magazine characteristics, real price, and the log of the annual number of issues. Regression (5) constrains the exposure elasticity coefficient to a unitary value. Second, regressions (2) and (6) include the year dummies for 2002 and 2003. Third, regressions (3) and (7) report fixed-effects specifications for six magazine categories and three years. All of the reported standard errors are based on robust procedures in Stata 8.2. All the youth coefficients in Table 4 are insignificantly different from 0, regardless of the specification or model. Statistical tests reported indicate that the negative binomial is a better representation of the data, which confirms results reported in Nelson (2005).
A. Poisson Results
The Poisson results in regressions (1)-(3) fail to demonstrate that targeting of adolescents is taking place, although there is evidence that advertisers tend to favor young adult audiences. All of the youth coefficients are insignificantly different from 0, and have standard errors equal to or greater than the coefficient magnitudes. Among the demographic variables, alcohol placements are negatively associated with the median age of adult readers. Median adult income and percent adult male readers have positive coefficients, but neither variable is statistically significant. Among the magazine variables, positive effects are found for percent single-copy sales (newsstand sales) and adult readers per copy. In regression (2), the year dummies are not statistically significant. This result illustrates the weakness in the data and model used by Garfield et al. (2003). Regression (3) demonstrates that alcohol advertisers have the strongest preference for men's style and sports magazines, followed by entertainment and music magazines. In this specification, the exposure elasticity was small in magnitude and insignificantly different from 0. Consequently, the annual issues variable was omitted from regressions (3) and (7).
B. Negative Binomial Results
In Table 4, regressions (4)-(7) contain the results for the negative binomial count model. For count data, the negative binomial is the main alternative to the Poisson model. Count data may be better described by the negative binomial if there is occurrence dependence or unobserved heterogeneity across magazines (Winkelmann 2003, p. 22). The negative binomial model also relaxes the presumed equality of the mean and variance functions that underlies the Poisson model. The negative binomial results again fail to demonstrate that targeting of adolescents is taking place as all of the youth coefficients are insignificantly different from 0. In other respects, the negative binomial results parallel the Poisson results, although the number of significant regressors increases. In particular, the real CPM price is statistically significant and negative. The exposure elasticities also are closer to unity. Comparing regressions (4) and (5), the results do not change much when the exposure elasticity is constrained to unity. Overall, the negative binomial results demonstrate the statistical importance of the real CPM price, adult median age, percent single-copy sales, adult readers per copy, square of readers per copy, and the exposure variable. The year dummies are insignificant in regression (6). The results for the fixed-effects specification are similar, except that the coefficient on the automobile category is significantly negative in regression (7).
The positive results for readers per copy is of special importance, since it illustrates a criticism by the FTC (2003, p. 32) of CAMY's methodology. Presumably, advertisers are concerned about the composition and size of the audience. CAMY's methodology addresses only the composition of the audience and completely ignores its size. As pointed out by the FTC (2003, p. 33), the young adult population (ages 21-34) is about 50% larger than the underage youth population. Furthermore, alcohol consumption per capita by younger adults is greater than older adults, and brand loyalty increases with age. Hence, there are several business reasons for alcohol companies to advertise in magazines with large young adult audiences who have not yet formed strong brand preferences.
C. Specification Tests
A well-known feature of the Poisson model is the presumed equality of the conditional mean and variance functions (equidispersion). This restriction may not hold due to occurrence dependence, unobserved heterogeneity, or because the zero outcomes of the data-generating process are quantitatively different from the positive outcomes. Occurrence dependence or systematic contagion can reflect past advertising successes or perhaps a tendency by advertisers to focus on a few magazines during a given time period due to so-called pulsing behavior (see Winkelmann 2003, pp. 16-22). Unobserved heterogeneity can arise if different models apply to different magazines due to random contagion, and this is reflected in a different proportion of zeros in the sample. For example, it is not clear if a zero placement occurs because advertisers did not happen to use a particular magazine during the study period or because that magazine would rarely be chosen for alcohol ads (e.g., zero counts for ads in Better Homes and Gardens). Using the results in Table 4, several specification tests were conducted. First, the Poisson model is nested within the negative binomial model (Winkelmann 2003, p. 100). Using comparable results in Table 4, a likelihood ratio (LR) test strongly rejects the Poisson model in favor of the negative binomial model. The LR test statistics are 923.0, 912.0, and 1031, respectively. The critical value of the chi-square distribution with one degree of freedom is 50.9 at the 99% confidence level. Second, the overdispersion parameters in the negative binomial regressions are significantly positive (see Cameron and Trevedi 1998, p. 79). Third, formal tests for overdispersion due to Cameron and Trevedi (1990) and Wooldridge (1996) rejected the null for the three Poisson regressions. Overall, the results strongly favor the negative binomial model as the better representation of count data for alcohol advertisements in magazines. Combining the results in the present article with Nelson (2005), this result is robust for a variety of data and model specifications.
IV. INCIDENCE RATE RATIOS, MARGINAL EFFECTS, AND ELASTICITIES
To assess or gauge the importance of different explanatory variables for placement of alcohol ads in magazines, it is useful to report standardized coefficients. Various standardized coefficients exist for count models (see Cameron and Trivedi 1998, pp. 80-82; Winkelmann 2003, pp. 68-71). In the conditional expectation function given by equation (1), each regression coefficient is the (constant) proportionate change in the conditional mean due to a unit change in the explanatory variable. If the regressor is a dummy variable, the coefficient gives the approximate relative impact. The incidence rate ratio (IRR) is given by exp(coefficient), and is a common way of comparing relative impacts in count models. The IRRs show the relative change in advertising counts for each unit change in an explanatory variable. Marginal effects depend on the observation size, but an average marginal effect is found by multiplying the regression coefficient by the mean of the dependent variable. Last, average elasticity values are computed by multiplying the coefficient estimate by the mean of the respective explanatory variable (Cameron and Trivedi 1998, p. 82).
For regression (4), Table 5 reports the incidence rate ratios, average marginal effects, and average elasticities. The standardized coefficients support the conclusion that the size of the adult audience is the most important variable for placement decisions. The audience IRR is 3.614 and the elasticity is 8.511. The price elasticity also is substantial, -1.955, and illustrates the importance of treating placement decisions as a demand function rather than just a marketing ploy. Ranked by the average elasticities, the statistically important variables are adult readers per copy, adult median age, square of readers per copy, real CPM price, annual issues, and percent single-copy sales.
V. BEER AND SPIRITS ADVERTISING COUNTS
An extension of the analysis is an examination of advertising placements by beverage. The data by beverage are drawn from an online CAMY report. The author collected count data for beer and spirits ads for 2001-03 for the sample of 28 magazines. For wine, there are too few positive counts to warrant analysis (65 out of 84 possible observations are 0 or 77.4%). There are 38 zero counts for beer (45.2%) and 17 zero counts for spirits (20.2%). A larger proportion of zero counts suggests the analysis should consider more complicated empirical models. Given the findings in section III, results are reported for the negative binomial model and the zero-inflated negative binomial model. The latter model allows for separate treatment of zeros and strictly positive outcomes (Winkelmann 2003, p. 148).
Table 6 displays the results by beer and spirits. Adult median age is significantly negative in all of the regressions. Percent male readers is significantly positive in all six regressions, which is a change from Table 4. The CPM price of an advertisement is significantly negative for spirits but insignificant for beer ads. Possibly there are too few magazines in the sample with positive placements to capture this aspect of decision making by beer advertisers. Percent single-copy sales is significantly positive in the negative binomial model. Adult readers per copy is always significantly positive and the square of readers per copy is significantly negative. The log of the annual number of issues is significantly positive and close to unity in all regressions. The year dummies are not significant, which again illustrates the shortcomings of the model used by Garfield et al. (2003). The percent of youth readers is not significant in any of the regressions, regardless of the model or specification. The results fail to support the allegation that beer and spirits advertisers are targeting youth readers. Beer advertisers favor magazines with more young adults, male readers, and larger adult audiences, but not adolescents. Spirits producers favor magazines with more young adults, male readers, and larger adult audiences, but not adolescents. Spirits producers also favor magazines with lower costs per advertisement.
In a number of cases, the coefficients for beer and spirits are similar in magnitude, but the average marginal values depend on the mean of the respective dependent variable. For example, in the zero-inflated model, the beer and spirits coefficients are identical for adult median age and the average elasticities are -5.43 and -5.40, respectively, which suggests similar responses by beverage. However, the elasticities for adult readers per copy are 15.9 and 31.6 for beer and spirits, respectively. Hence, the results indicate that spirits producers advertise in magazines with a broader reach compared to beer producers. This outcome reflects the fact that magazines are the principal means of spirits advertising, given the long-standing voluntary ban of spirits ads on radio and television. Although this ban has been relaxed for cable TV, 70% of spirits ads in 2003 were in magazines compared to only 16% for broadcast media. (3) The comparable percentages for beer are 6.6% for magazines and 80% for television.
Advertisements for alcohol beverages appear in a variety of magazines, including those with adolescent readers. The empirical results in this article illustrate some of the factors that affect advertising placement decisions for a sample of 28 major magazines, including the size of the adult audience and the price charged for an ad placement. The results for audience size capture criticisms by the FTC (2003, p. 33) of the methodology used by CAMY. The results for price are new economic evidence that was ignored by past researchers. Considerable controversy exists regarding the placement of alcohol ads in magazines where the youth proportion of the audience is greater than 15%. The exact basis for this regulatory standard is difficult to discern, because several recent literature reviews fail to provide evidence that alcohol ads affect alcohol consumption in a material manner (Grube 2004; Nelson 2001, 2004; National Institute on Alcohol Abuse and Alchoholism 2000, p. 422; NRC 2004, p. 134). Using an improved data set and econometric methods, the results in the article fail to support claims that alcohol advertisers target underage youth. The empirical findings are contrary to the conclusions in Garfield et al. (2003) and also illustrate the shortcomings of the methodology used in a series of reports commissioned by the CAMY (2002, 2005a). Finally, policy makers in the alcohol area would be well advised to turn their attention to discussion of matters of importance for youthful drinking behaviors, rather than decisions made in the market for advertising space.
American Medical Association. "AMA Policy Consideration: Labeling, Advertising, and Promotion of Alcoholic Beverages." Report of the Council on Scientific Affairs (A-04), AMA, Chicago, 2004; available online at www.alcoholpolicysolutions.net/policy_ama_alcohol.htm.
Bresnahan, T. F. "The Demand for Advertising by Medium: Implications for the Economies of Scale in Advertising," in Empirical Approaches to Consumer Protection Economics, edited by P. M. Ippolito and D. T. Scheffman (Washington, DC: Federal Trade Commission, 1984), pp. 135-63.
Cameron, A. C., and P. K. Trivedi. "Regression-Based Tests for Overdispersion in the Poisson Model." Journal of Econometrics, 46, 1990, 347-64.
______. Regression Analysis of Count Data (Cambridge: Cambridge University Press, 1998).
CAMY (Center on Alcohol Marketing and Youth). Overexposed: Youth a Target of Alcohol Advertising in Magazines (Washington, DC: CAMY, 2002); available online at camy.org/research/mag0902.
______. Youth Overexposed: Alcohol Advertising in Magazines, 2001 to 2003 (Washington, DC: CAMY, 2005a); available online at camy.org/research/mag0405.
______. Alcohol Marketing and Youth (Washington, DC: CAMY, 2005b); available online at camy.org/factsheets/index.php?FactsheetID=24.
CSPI (Center for Science in the Public Interest). "Alcohol Advertising: Are Our Kids Being Targeted?" CSPI, Washington, DC, 2001; available online at www.cspinet.org/booze/iss_youth.htm.
Depken, C. A. "Audience Characteristics and the Price of Advertising in a Circulation Industry: Evidence from US Magazines." Information Economics and Policy, 16, 2004, 179-96.
Depken, C. A., and D. P. Wilson. "The Value of Advertising in a Magazine Bundle," in Advances in Applied Micro-Economics: Advertising and Differentiated Products, edited by M. R. Baye and J. P. Nelson (Amsterdam: JAI Press, 2001), vol. 10, pp. 109-28.
DISCUS (Distilled Spirits Council of the United States). Semi-Annual Code Report: Code of Responsible Practices for Beverage Alcohol Advertising and Marketing (Washington, DC: DISCUS, March 2005); available online at www.discus.org/pdf/SemiAnnualCode.pdf.
Ehrlich, I., and L. Fisher. "The Derived Demand for Advertising: A Theoretical and Empirical Investigation." American Economic Review, 72, 1982, 366-88.
Fare, R., S. Grosskopf, B. J. Seldon, and V. J. Tremblay. "Advertising Efficiency and the Choice of Media Mix: A Case of Beer." International Journal of Industrial Organization, 22, 2004, 503-22.
FTC (Federal Trade Commission). 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, 1999); available online at www.ftc.gov/reports/alcohol/alcoholreport.htm.
______. Alcohol Marketing and Advertising--A Report to Congress (Washington, DC: FTC, 2003); available online at www.ftc.gov/os/2003/09/alcohol08report.pdf.
Garfield, C. F., P, J. Chung, and P. J. Rathouz. "Alcohol Advertising in Magazines and Adolescent Readership." Journal of the American Medical Association, 289, May 14, 2003, 2424-29.
Grube, J. W. "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, 2004), pp. 597-624.
Koschat, M. A., and W. P. Putsis. "Who Wants You When You're Old and Poor? Exploring the Economics of Media Pricing." Journal of Media Economics, 13, 2004, 215-32.
______. "Audience Characteristics and Bundling: A Hedonic Analysis of Magazine Advertising Rates." Journal of Marketing Research, 39, 2002, 262-73.
National Institute on Alcohol Abuse and Alcoholism. "Alcohol Advertising: What Are the Effects?," in 10th Special Report to the U.S. Congress on Alcohol and Health (Washington, DC: NIAAA, 2000), pp. 412-26.
National Research Council, Institute of Medicine. Reducing Underage Drinking: A Collective Responsibility (Washington, DC: National Academies Press, 2004).
Nelson, J. P. "Alcohol Advertising and Advertising Bans: A Survey of Research Methods, Results, and Policy Implications," in Advances in Applied Micro-Economics: Advertising and Differentiated Products, edited by M. R. Baye and J. P. Nelson (Amsterdam: JAI Press, 2001), vol. 10, pp. 239-95.
______. "Advertising Bans in the United States," in EH.Net Encyclopedia, edited by R. Whaples (2004), pp. 1-29; online document available at www.eh.net/encyclopedia/contents/Nelson.AdBans.php.
______. "Advertising, Alcohol, and Youth." Regulation: The Cato Review of Business and Government, 28, 2005, 40-47.
Seldon, B. J., and C. Jung. "Derived Demand for Advertising Messages and Substitutability among the Media." Quarterly Review of Economics and Finance, 33, 1993, 71-86.
Seldon, B. J., R. T. Jewell, and D. M. O'Brien. "Media Substitutability and Economies of Scale in Advertising." International Journal of Industrial Organization, 18, 2000, 1153-80.
Silk, A. J., L. R. Klein, and E. R. Berndt. "Inter-Media Substitutability and Market Demand by National Advertisers." Review of Industrial Organization, 20, 2002, 323-48.
Winkelmann, R. Econometric Analysis of Count Data, 4th ed. (Berlin: Springer, 2003).
Wooldridge, J. M. "Quasi-Likelihood Methods for Count Data," in Handbook of Applied Econometrics, edited by M. H. Pesaran and P. Schmidt (Oxford: Black-well, 1997), vol. 2, pp. 352-406.
JON P. NELSON*
*The author thanks Patrick Anderson, Kenneth Elzinga, Robert Feinberg, Everett Peterson, Robert Michaels, and three anonymous referees for helpful comments on earlier drafts. The usual caveats apply. The author has consulted with a law firm that represents companies in the alcohol industry. The topic and content of the article were prepared independently by the author, and the article was not reviewed by the law firm or other interested parties prior to submission for publication.
Nelson: Professor Emeritus, Department of Economics, 608 Kern Graduate Building, Pennsylvania State University, University Park, PA 16803. Phone 1-814-865-0130, Fax 1-814-863-4775, E-mail email@example.com
CAMY: Center on Alcohol Marketing and Youth
CPM: Cost per Thousand Circulation
DISCUS: Distilled Spirits Council of the United States
FTC: Federal Trade Commission
GRP: Gross Rating Points
IRR: Incidence Rate Ratio LR: Likelihood Ratio
NRC: National Research Council
P4C: Full-Page Four-Color Ad
1. See Nelson (2001, 2004) for discussion of alcohol advertising and the First Amendment. The third prong of the Central Hudson test requires that the government censor must demonstrate that an advertising ban or regulation will directly and materially advance a substantial government interest.
2. CAMY's list of 124 magazines includes many magazines that do not have large youth audiences, such as Bon Appetit, Forbes, and The New Yorker. Restricting the sample to magazines that allow alcohol ads and are read by adolescents would appear to bias the results toward rejecting the null. Tables 2 and 3 can be used to judge the dispersion of magazines by youth readership, circulation, and subject content.
3. In June 1996, now-defunct distiller Joseph E. Seagram aired TV advertisements for Crown Royal Canadian Whiskey on an NBC-TV affiliate in Corpus Christi, TX, thereby breaking the industry's long-standing voluntary ban of broadcast advertising. Reactions to the ads included a full-page protest in the August 2 edition of the New York Times and a "Just Say No" bill introduced in Congress by Rep. Joseph Kennedy (D-MA). President Bill Clinton asked the industry to go back to the ban, calling Seagram's TV ads "simply irresponsible." He also requested an investigation by the Federal Communications Commission. In late 2001, NBC-TV announced its intention to allow alcohol ads after 9 p.m. on programs such as Saturday Night Live, and the controversy shifted focus onto the network. NBC eventually reinstated its voluntary ban to realign itself with the other three major networks. Since that date, public health groups have continued to pressure the industry and the FTC to prevent spirits ads from appearing on network TV. As a consequence, distilled spirits advertising continues to be dominated by print advertisements.
TABLE 1 Variable Description and Data Sources Definition, Units, and Variable Summary Statistics Data Source Ad Count Number of alcohol ads for CAMY 2005 2001, 2002, & 2003, Magazine Study including 11 zero observations (13.1% of total observations). Sample mean (SD) = 43.8 (52.2); median = 16.0. Covers all three beverages. Beer Ad Count Number of beer ads for 2001, CAMY Web site, 2002, and 2003, including interactive data 38 zero observations (45.2% tool of total observations). Sample mean (SD) = 7.8 (12.7); median = 1.0. Beer includes favored malt beverages. Spirits Ad Count Number of spirits ads for CAMY Web site, 2001, 2002, and 2003, interactive data including 11 zero tool observations (20.2% of total observations). Sample mean (SD) = 34.6 (41.0); median = 14.0. Spirits include liqueurs and cordials. Percent Youth of Teen audience (ages 12-19 MRI Magazine Report Audience yrs.) divided by total (Teen) audience (teens + adults). Mean (SD) = 17.1% (5.5). Range: 4.3 to 31.2%. Adult Median Age Median age of adult readers. MRI Magazine Report Mean (SD) = 34.8 yrs. (5.8). Range: 23.2 to 46.4 yrs. Adult Median Real Median household income of MRI Magazine Report Income adult readers. Expressed in thousands of real dollars using the CPI (2000 = 100). Mean (SD) = $54.8 (8.9). Range: $33.3 to $71.9. Percent Adult Male Adult male audience divided MRI Magazine Report Readers by total adult audience (adult men + adult women). Mean (SD) = 49.8% (29.2). Range: 6.3 to 91.9%. Real CPM Price of a Real CPM price of a P4C ad. SRDS Magazine P4C Ad Magazine's cost for a P4C Advertising Source ad divided by its rate base circulation (in thousands) for each year. Expressed in real dollars using the PPI (2000 = 100). Mean (SD) = $63.1 (18.5). Range: $26.1 to $100.6 per P4C ad. Percent Single-Copy Percent of circulation SRDS Magazine Sales accounted for by single-copy Advertising Source sales at newsstands. Mean (SD) = 23.5% (19.3). Range: 2.6 to 80.1%. Adult Readers per Total adult audience divided MRI Magazine Report Copy by circulation. Square of this variable is used to pick-up other nonlinearity. Mean (SD) = 6.62 (2.3). Range: 11.1 to 248.7 readers per copy. Magazine Category Five dummy variables for Author constructed magazines by category. Omitted category is general and other. See Table 3 for categories. Year Dummies 2002 dummy = 1 if year is Author constructed 2002; 2003 dummy = 1 if year is 2003. Annual No. of Log of annual no. of issues MRI Circulation Issues for "exposure" differences Report (no time across magazines. Unlogged variation) mean (SD) = 21.3 (16.0). Range: 12 to 53 Notes: All data for 2001-2003 from MRI Magazine Report, MRI Magazine Report (Teen), and MRI Circulation Report accessed at www.mriplus.com/cgi-bin/WebObjects/mriplus.woa. CAMY 2005b, p. 20; CAMY Web site, interactive data tool for beverage data, "Magazine Alcohol Ads: Your Child and You," at camy.org. Standard Rate and Data Service, SRDS Consumer Magazine Advertising Source (Des Plaines, IL: SRDS), July issue. TABLE 2 Sample of Magazines and Selected Data for 2003 Rate Base Alcohol Ads, % Youth of Circulation Magazine (Circ. Rank) 2001-03 Audience (000s) Allure (88) 23 30.5 900 Better Homes and Gardens (5) 3 4.7 7600 Car & Driver (61) 23 22.5 1350 Cosmopolitan (19) 261 19.1 2600 ESPN The Magazine (40) 341 29.2 1500 Ebony (37) 84 18.6 1700 Entertainment Weekly (38) 332 16.7 1500 Fitness (58) 3 20.5 1100 Glamour (29) 39 18.1 2200 Hot Rod (na) 2 26.4 700 In Style (42) 291 18.9 1500 Jet (100) 102 21.3 900 Maxim (25) 453 15.0 2500 Motor Trend (67) 4 25.8 1250 Newsweek (17) 23 8.6 3100 People (12) 78 12.5 3250 Popular Mechanics (72) 47 16.3 1200 Popular Science (53) 30 19.9 1450 Road & Track (na) 12 21.4 750 Rolling Stone (66) 446 24.7 1250 Self (62) 35 19.0 1200 Shape (45) 5 15.8 1500 Spin (na) 185 29.3 525 Sports Illustrated (16) 463 20.5 3150 The Source (na) 7 33.3 475 Time (10) 51 9.8 4000 Vibe (na) 193 28.7 800 Vogue (75) 139 18.1 1100 Ad Cost per Adult Readers Magazine (Circ. Rank) Thous. Circulation per Copy Allure (88) $82.25 4.43 Better Homes and Gardens (5) 42.16 4.97 Car & Driver (61) 105.36 7.49 Cosmopolitan (19) 62.22 6.29 ESPN The Magazine (40) 90.00 6.21 Ebony (37) 34.08 5.62 Entertainment Weekly (38) 68.71 5.26 Fitness (58) 76.34 4.16 Glamour (29) 55.35 5.67 Hot Rod (na) 100.95 9.90 In Style (42) 64.33 5.16 Jet (100) 29.64 8.32 Maxim (25) 66.80 5.46 Motor Trend (67) 101.71 5.27 Newsweek (17) 61.77 6.08 People (12) 52.62 10.16 Popular Mechanics (72) 78.22 7.90 Popular Science (53) 58.92 5.12 Road & Track (na) 114.33 7.27 Rolling Stone (66) 88.66 7.90 Self (62) 73.15 3.78 Shape (45) 60.45 3.54 Spin (na) 94.17 4.77 Sports Illustrated (16) 71.75 6.17 The Source (na) 66.63 15.77 Time (10) 53.00 5.25 Vibe (na) 99.01 8.00 Vogue (75) 81.02 8.75 Notes: See Table 1 for description of variables and data sources. Youth ages 12-19 are 13.7% of the U.S. population ages 12 and older. Circulation rank in 2003 is from MPA data on the top 100 magazines; www.magazine.org/circulation. TABLE 3 Magazine Categories and Average Data for 2003 Total Alcohol Mean % Mean Adult Magazine Category Ads, 2003 Youth, 2003 Age (yrs) Automobiles 12 19.3 35.5 Black 137 18.5 34.2 Men's style and sports 345 17.6 32.0 Women's style 186 15.6 33.5 Entertainment and music 270 18.6 30.9 General and other 74 13.5 39.8 All magazines 1024 16.5 35.3 Mean Adult Mean Real Mean Adult Magazine Category Income ($000) P4C Price (a) Readers per Copy Automobiles 61.8 92.95 7.5 Black 38.1 47.75 7.3 Men's style and sports 64.4 67.06 5.9 Women's style 63.3 57.86 6.5 Entertainment and music 54.8 65.28 8.8 General and other 66.0 57.34 5.0 All magazines 59.8 63.93 6.6 Automobiles: Car & Driver, Hot Rod, Motor Trend, Road & Track. Black: Ebony, Jet, Vibe. Men's Style & Sports: ESPN, Maxim, Sports Illustrated. Women's Style: Cosmopolitan, Glamour, In Style, Vogue. Entertainment & Music: Entertainment Weekly, People, Rolling Stone, Spin, The Source. General & Other: Allure, Better Homes & Gardens, Fitness, Newsweek, Popular Mechanics, Popular Science, Self, Shape, Time. (a) Price is measured in constant 2000 dollars for a P4C ad per 1000 circulation. TABLE 4 Count Data Regressions for All Beverages Poisson Model Variable (1) (2) (3) Constant -1.853 (2.28) -1.711 (2.42) 2.368 (0.467)* % Youth (ages 0.029 (0.027) 0.024 (0.032) -0.008 (0.027) 12-19) Adult median -0.184 (0.027)* -0.185 (0.028)* -- age Adult median 0.022 (0.016) 0.023 (0.017) -- real income % Adult male 0.011 (0.008) 0.011 (0.008) -- readers Real CPM -0.015 (0.010) -0.014 (0.010) -- price of P4C ad % Single-copy 0.028 (0.013)* 0.027 (0.013)* -- sales Adult readers 1.608 (0.290)* 1.634 (0.294)* -- per copy Square of -0.115 (0.017)* -0.117 (0.017)* -- adult readers per copy Year 2002 -- -0.067 (0.115) -0.162 (0.065)* dummy Year 2003 -- -0.149 (0.131) -0.344 (0.087)* dummy Auto category -- -- -0.816 (0.501) dummy Black -- -- 1.689 (0.363)* category dummy Men's style -- -- 2.879 (0.270)* category dummy Women's style -- -- 2.038 (0.375)* category dummy Entertainment -- -- 2.198 (0.422)* and music category dummy Log of annual 1.529 (0.417)* 1.503 (0.426)* Not incl. no. of issues Log -818.8 -812.9 -863.2 likelihood Alpha -- -- -- dispersion parameter (SE) Negative Binomial Model Variable (4) (5) Constant 0.701 (2.28) 1.172 (2.23) % Youth (ages 0.052 (0.042) 0.052 (0.042) 12-19) Adult median -0.166 (0.035)* -0.176 (0.032)* age Adult median 0.015 (0.027) 0.026 (0.022) real income % Adult male 0.009 (0.008) 0.009 (0.007) readers Real CPM -0.031 (0.018)** -0.036 (0.018)* price of P4C ad % Single-copy 0.023 (0.014)** 0.017 (0.010)** sales Adult readers 1.285 (0.366)* 1.348 (0.338)* per copy Square of -0.087 (0.017)* -0.088 (0.016)* adult readers per copy Year 2002 -- -- dummy Year 2003 -- -- dummy Auto category -- -- dummy Black -- -- category dummy Men's style -- -- category dummy Women's style -- -- category dummy Entertainment -- -- and music category dummy Log of annual 1.175 (0.326)* 1.000 no. of issues Log -357.1 -357.2 likelihood Alpha 0.917 (0.263)* 0.921 (0.268)* dispersion parameter (SE) Negative Binomial Model Variable (6) (7) Constant 0.607 (2.45) 2.184 (0.585)* % Youth (ages 0.053 (0.043) 0.003 (0.030) 12-19) Adult median -0.166 (0.036)* -- age Adult median 0.015 (0.027) -- real income % Adult male 0.009 (0.008) -- readers Real CPM -0.031 (0.018)** -- price of P4C ad % Single-copy 0.023 (0.014)** -- sales Adult readers 1.293 (0.363)* -- per copy Square of -0.087 (0.016)* -- adult readers per copy Year 2002 0.023 (0.122) -0.145 (0.103) dummy Year 2003 0.059 (0.134) -0.228 (0.103)* dummy Auto category -- -0.914 (0.467)* dummy Black -- 1.621 (0.334)* category dummy Men's style -- 2.816 (0.241)* category dummy Women's style -- 1.983 (0.357)* category dummy Entertainment -- 2.108 (0.391)* and music category dummy Log of annual 1.178 (0.333)* Not incl. no. of issues Log -357.0 -347.9 likelihood Alpha 0.917 (0.262)* 0.700 (0.299)* dispersion parameter (SE) Notes: Dependent variable is count of alcohol advertisements in each of 28 magazines for 2001, 2002, and 2003, including 11 zero observations. Estimates obtained using Stata 8.2. Robust SEs in parentheses; one and two asterisks indicate that the z-statistic is equal to or greater than 1.96 and 1.64, respectively. TABLE 5 Incidence Rate Ratios, Marginal Effects, and Elasticities Ave. Ave. Variable IRR (z-stat) Marginal Elasticity % Youth readers 1.053 (1.24) 2.28 0.887 Adult median age 0.847 (4.77) -7.27 -5.783 Adult median income 1.015 (0.53) 0.66 0.821 %Adult male readers 1.009 (1.21) 0.39 0.448 CPM-P4C price (real) 0.969 (1.70) -1.36 -1.955 % Single-copy sales 1.024 (1.64) 1.01 0.538 Adult readers per copy 3.614 (3.51) 56.28 8.511 Sq. readers per copy 0.917 (5.24) -3.81 -4.283 Annual no. of issues -- -- 1.175 TABLE 6 Count Data Regressions for Beer and Spirits Negative Binomial Model Variable Beer Beer Constant -2.383 (3.20) -3.458 (3.22) % Youth (ages 12-19) 0.028 (0.054) 0.054 (0.055) Adult median age -0.274 (0.047)* -0.268 (0.046)* Adult median real income 0.029 (0.031) 0.033 (0.031) % Adult male Readers 0.021 (0.009)* 0.023 (0.008)* Real CPM price of P4C ad -0.011 (0.019) -0.016 (0.018) % Single-copy sales 0.030 (0.017)** 0.031 (0.016)** Adult readers per copy 1.517 (0.575)* 1.565 (0.564)* Square of adult readers per copy -0.118 (0.035)* -0.121 (0.034)* Year 2002 dummy -- 0.328 (0.357) Year 2003 dummy -- 0.465 (0.366) Log of annual no. of issues 1.859 (0.416)* 1.862 (0.413)* Log likelihood -186.9 -186.1 Alpha dispersion parameter (SE) 1.017 (0.303)* 0.960 (0.296)* Negative Binomial Model Variable Spirits Spirits Constant -0.266 (2.40) -0.255 (2.49) % Youth (ages 12-19) 0.067 (0.043) 0.067 (0.044) Adult median age -0.147 (0.038)* -0.147 (0.038)* Adult median real income 0.001 (0.028) 0.001 (0.028) % Adult male Readers 0.013 (0.006)* 0.013 (0.006)* Real CPM price of P4C ad -0.030 (0.015)* -0.030 (0.016)** % Single-copy sales 0.033 (0.016)* 0.033 (0.016)* Adult readers per copy 1.240 (0.292)* 1.239 (0.295)* Square of adult readers per copy -0.087 (0.015)* -0.087 (0.015)* Year 2002 dummy -- -0.002 (0.296) Year 2003 dummy -- -0.007 (0.302) Log of annual no. of issues 1.318 (0.314)* 1.318 (0.315)* Log likelihood -337.9 -337.9 Alpha dispersion parameter (SE) 1.073 (0.198)* 1.073 (0.198)* Zero-Inflated Negative Binomial Variable Beer Spirits Constant -6.387 (2.07)* 3.078 (1.72) % Youth (ages 12-19) -0.005 (0.028) 0.031 (0.029) Adult median age -0.156 (0.026)* -0.155 (0.027)* Adult median real income 0.046 (0.017)* 0.009 (0.019) % Adult male Readers 0.015 (0.005)* 0.009 (0.004)* Real CPM price of P4C ad 0.010 (0.012) -0.029 (0.011)* % Single-copy sales 0.012 (0.010) 0.015 (0.010) Adult readers per copy 2.051 (0.503)* 0.914 (0.223)* Square of adult readers per copy -0.149 (0.034)* -0.062 (0.012)* Year 2002 dummy -- -- Year 2003 dummy -- -- Log of annual no. of issues 1.024 (0.273)* 0.889 (0.217)* Log likelihood -130.9 -288.3 Alpha dispersion parameter (SE) 0.135 (0.054)* 0.391 (0.073)* Notes: Dependent variable is count of advertisements by beverage in each of 28 magazines for 2001, 2002, and 2003. Estimates obtained using Stata 8.2. Robust SEs in parentheses; one and two asterisks indicate that the z-statistic is equal to or greater than 1.96 and 1.64, respectively. Zero counts are 38 for beer and 17 for spirits. Wine ads are excluded due to the large number of zero counts. For specification of the zero-inflated negative binomial model, see Winkelmann (2003).
|Printer friendly Cite/link Email Feedback|
|Author:||Nelson, Jon P.|
|Publication:||Contemporary Economic Policy|
|Date:||Jul 1, 2006|
|Previous Article:||Hungary's eurozone entry date: what do the markets think and what if they change their minds?|
|Next Article:||Advertising restrictions and cigarette smoking: evidence from myopic and rational addiction models.|