Broadcast advertising and U.S. demand for alcoholic beverages.

1. Introduction

The debate over the advertising and promotion of alcoholic beverages is an important part of public policy toward alcohol. There are two general issues. The first issue is the competitive effect of advertising on market structure and thus the level of price relative to marginal cost. The second issue is the consumption effect of alcohol advertising on market demand and its possible role as a contributor to alcohol abuse and external costs of consumption, such as drunk driving, liver cirrhosis and other health effects, binge drinking, and underage drinking (Fisher 1993). Although the advertising-consumption issue has been debated for many years, there is a limited body of empirical research on the relationship between advertising and alcohol demand in the United States.(1)

A more specific issue is the possible effect of broadcast advertising on alcohol consumption. An especially strong view has been expressed by the policy watchdog organization Center for Science in the Public Interest (CSPI), which called for advertising bans and other restrictions on the messages and content of ads, especially those that might reach large numbers of youth or heavy drinkers (CSPI 1995). In 1983, CSPI and other groups petitioned the Federal Trade Commission (FTC) for a complete ban on broadcast advertising of alcohol (CSPI 1983), but the FTC ruled against the petition on the grounds that there was "no reliable basis to conclude that alcohol advertising significantly affects consumption, let alone abuse" (FTC 1985, p. 2). The implication is that advertising alters brand shares but has little or no effect on beverage or total alcohol consumption. The CSPI petition was followed by related efforts by other groups, including passage of the Alcoholic Beverage Labeling Act of 1988; congressional hearings on health warnings in alcohol advertising (U.S. Congress 1990); a second CSPI petition in 1992; and restrictive advertising bills introduced in the 101st and 102nd Congresses by Senators Gore, Thurmond, and Simon and Representatives Kennedy and Conyers. In addition, during 1991 and 1992, Surgeon General Novello spoke publicly on the possible dangers of alcohol advertising targeting youth, commissioning reports on youth and alcohol and the mass media and alcohol abuse. Finally, in June 1996, Seagram Distillers Company began airing local television ads for several of its brands. These ads broke the industry's 60-year voluntary ban on broadcast advertising by producers of distilled spirits. This action quickly drew criticism from various groups and individuals, including President Clinton, Representative Kennedy, and Chairman Hundt of the Federal Communications Commission. Some spokespersons called for additional restrictions or complete elimination of all forms of alcohol advertising, but none of the critics seemed to question the effectiveness of broadcast advertising of alcohol.

Does broadcast advertising affect alcohol consumption, including total consumption of pure alcohol (ethanol) and consumption of each of the three beverages? The objective of this paper is to provide an answer to this question. In addition, the paper will examine the possible interaction between demographic changes in the United States and alcohol advertising. Most previous empirical studies exclude demographic variables and use annual data on aggregate advertising expenditures and consumption for individual beverages (Lee and Tremblay 1992; Goel and Morey 1995), total consumption of ethanol (Duffy 1990; Nelson and Moran 1995), and systemwide demand for the three beverages (Selvanathan 1989; Duffy 1991; Nelson and Moran 1995). These studies fail to account for possible differences in advertising effects across media and substitution among media. A second shortcoming is the failure to account for seasonal patterns of alcohol consumption and advertising, and annual aggregation tends to obscure this source of variation (Clarke 1976; Bass and Leone 1983; Saffer 1993). Further, some advertisers are known to vary expenditures during the year to offset diminishing returns. Annual data on aggregate advertising fail to reflect important variation by media and season, and any resulting effects on total or beverage demand are obscured.

None of the previous empirical studies employs quarterly data, disaggregate advertising by media, demographic variables, and a systemwide approach to estimation.(2) This study attempts to fill this void by estimating a demand system for consumption of beer, wine, and spirits and total consumption of ethanol. Quarterly data for 1977:3-1994:4 are employed. Using the Rotterdam model of a differential demand system, the empirical model explains the growth rate of per capita consumption dependent on growth rates for beverage prices, real income, demographic variables, and real advertising by media and beverage. Real advertising expenditure is broken down into broadcast advertising messages (television, radio) and print advertising messages (magazines, newspapers, outdoor). Given strong separability of consumer preferences, the empirical analysis in the paper takes place in two stages. In the first stage, an analysis is conducted on the demand for the three beverages conditional on total alcohol consumption. Within this three-equation system, advertising serves to allocate a predetermined total consumption among the beverages, other things held constant. In the second stage, a composite demand function is estimated wherein the growth rate of total alcohol consumption per capita is the dependent variable and several different specifications of real advertising by beverage or media are included as explanatory variables. Both the conditional and the composite empirical relationships are important for public policy, the first for the controversy surrounding broadcast advertising and the second for the effects of all forms of advertising on total consumption. No other empirical study has addressed both of these concerns.

The remainder of the paper is divided into five sections. Section 2 provides a brief overview of recent trends in alcohol consumption and advertising. Section 3 develops the systemwide demand model. Section 4 reports empirical results by beverage, and section 5 reports results for the composite demand function for total alcohol. Section 6 presents the study's conclusions. The conclusion is that alcohol advertising, including broadcast advertising, has little or no effect on alcohol demand.

2. Background

Alcoholic Beverage Consumption and Advertising

During the past 20 years, important changes have taken place in alcoholic beverage consumption and advertising. Table 1 displays basic data on consumption and advertising by beverage. On a per capita basis, spirits consumption has been declining since 1975, beer since 1980, and wine since 1985. As a result, total consumption of pure alcohol has declined since 1980. In part, this is due to a shift in consumption toward beer and wine, which have lower ethanol contents.(3) In 1975, 46% of ethanol consumption was in the form of beer, followed by spirits, (42%) and wine (12%). In 1995, the breakdown was beer (57%), spirits (30%), and wine (13%). In contrast to consumption, nominal beverage prices have increased steadily over the past 20 years. While beverage prices rose at average annual rates of 2.5% to 3.1% during 19851995, the consumer price index increased on average by 3.5% per annum. Thus, the price of alcoholic beverages has declined relative to all other goods and services.

Table 1 also displays basic data on alcohol advertising. For beer, nominal and real advertising expenditures peaked in 1985, about five years after the peak for per capita consumption. The trend in beer advertising in the early 1980s can be traced in part to efforts by several leading brewers to win back market share lost to Miller Brewing during the 1970s. For wine, nominal and real advertising peaked in 1985 along with per capita consumption. The change after 1985 reflects the substantial decline in the demand for wine coolers, a product introduced in 1981. For spirits, nominal and real advertising peaked in 1980, about five years after the peak in per capita consumption. This trend reflects in part a belated attempt by spirits producers, especially Seagram, to win back some of the market share lost to beer and wine during the 1970s. In addition, there have been important changes in the use of different advertising media. Since 1980, beer and wine advertising have increasingly used broadcast media, especially television, whereas spirits advertising has focused more on print media relative to outdoor ads.

Table 1. Alcoholic Beverage Consumption and Advertising

                                                       ACGR (%)(a)
Beverage    1975     1980     1985     1990     1995   1985-1995

Fluid Gallons (Millions)

Beer      4598.6   5512.2   5677.3   5984.9   5881.0        0.35
Wine(b)    361.6    471.8    569.0    516.1    469.0       -1.91
Spirits    423.4    449.4    416.3    374.5    328.0       -2.36

Fluid Gallons per Capita'

Beer        29.3     32.1     31.1     31.2      29.2      -0.63
Wine         2.3      2.7      3.1      2.7       2.3      -2.94
Spirits      2.7      2.6      2.3      2.0       1.6      -3.56

Ethanol Gallons per Capita

Beer        1.32     1.44     1.40     1.40      1.31      -0.66
Wine        0.33     0.35     0.40     0.35      0.30      -2.84
Spirits     1.19     1.10     0.90     0.81      0.68      -2.76

CPI (1990 = 100)

Beer        51.4     68.6     86.4    100.0     110.9       2.53
Wine        57.2     78.3     87.8    100.0     117.3       2.94
Spirits     60.0     71.4     83.8    100.0     113.8       3.11

Nominal Advertising (Millions of Dollars)

Beer       139.6    419.6    770.0    659.3     720.0      -0.67
Wine        59.4    139.1    220.7    105.1      85.3      -9.07
Spirits    196.6    348.0    303.0    290.6     227.4      -2.83

Real Advertising (Millions of 1990 Dollars)(c)

Beer       309.5    650.5    916.7    659.3     626.1      -3.74
Wine       131.7    215.7    262.7    105.1      74.2     -11.88
Spirits    435.9    539.5    360.7    290.6     197.7      -5.84

Source: Jobson's Liquor Handbook and Impact (various issues).

a Annual compound growth rate.

b Resident population aged 16 and over (as of July 1).

c Deflator is the implicit price deflator for GDP.

In summary, beer advertising increased during 1980-1985, but consumption did not - real annual beer advertising expenditures increased by 41% from 1980 to 1985, yet total annual sales rose by only 3% and per capita annual consumption declined by 3%. Real advertising of spirits increased by 24% during 1975-1980, but total consumption rose by only 6% and per capita consumption fell by 4%. Real wine advertising fell by 60% from 1985 to 1995, whereas total and per capita consumption fell by 18% and 25%, respectively. Advertising of all beverages rose by 75% from 1975 to 1985, but average ethanol gallons per capita fell by 5%. Advertising of all beverages fell by 42% from 1985 to 1995, whereas per capita consumption of ethanol fell by 15%. Further, if advertising has a strong effect on consumption, producers might be expected to increase advertising to prevent or limit the decline in sales that began in the 1980s. Instead, nominal and real expenditures have fallen along with consumption, raising questions about the direction of causality in these data.(4) Thus, these data fail to reveal a consistent pattern or relationship between advertising and consumption. However, this has not prevented claims of a positive relationship between advertising and consumption (e.g., CSPI 1995) The objective of this paper is to estimate this relationship using better data and more rigorous methods.

Quarterly Consumption and Advertising

The empirical analysis in this paper uses quarterly data on apparent consumption and advertising.(5) The consumption data are derived from monthly data on shipments of each beverage and are expressed in both fluid gallons and ethanol gallons per capita consumed of beer, wine, or spirits. In the empirical models, to remove purely seasonal effects, I used as dependent variables the growth rates of consumption using four-quarter differences of logarithms. Quarterly data on broadcast and print advertising also display seasonal variation during the year. To develop the data series on quarterly advertising, eight advertising media are analyzed: magazines, newspaper supplements, outdoor, network radio, and network, spot, syndicated, and cable television. The real data series are obtained by deflating expenditure on each media by its own price index, adjusted for media "reach" or audience potential. The amount of variation in the growth-rate data is substantial, suggesting that any relationship with consumption will not be obscured by temporal or media aggregation.(6)

Demographic Changes

It is well known that alcohol use declines with age, especially among the elderly, and that the proportion of abstainers increases with age for both sexes (Hilton and Clark 1987). More recently, U.S. drinking patterns have been affected by a decrease in the college-aged population that began in 1981 and an increase in the over-65 population as mortality rates continue to decline among the elderly. Thus, empirical studies of average per capita consumption need to include variables for both tails of the population age distribution. A failure to control for demographic factors runs the risk of falsely attributing a decline in consumption to a decline in real advertising expenditures or to other restrictive alcohol policies. The importance of these demographic changes is demonstrated by Nelson (1997), who reports that the net decline in ethanol consumption in the 1980s was due largely to an increase in the 65-plus population and a decline in the population in the 18-29 age-group, other things held constant. The present study extends this earlier work by adding several advertising variables and by focusing more on the composite demand function for total ethanol.

There is the additional policy concern that advertising, especially broadcast advertising, plays an important role in underage drinking. Data on average consumption per capita cannot address this concern directly. Nevertheless, several related concerns can be addressed. First, given that beer is the drink of choice among youth, does beer broadcast advertising have an especially strong influence on per capita consumption compared to print advertising? Second, does advertising reinforce the effects of demographic variables? That is, if advertising contributes to underage drinking, and advertising also affects overall consumption, do these effects combine to produce a complementary relationship among advertising, growth of the drinking age population, and per capita consumption? Third, advertising could affect average per capita consumption by causing individuals to drink rather than abstain, start drinking at an earlier age, drink on more occasions, and drink more on a given occasion. These effects might be offset in part by lifestyle and legal changes, and these types of changes will be captured as exogenous trends. Holding constant beverage prices, income, and trend variables, there do not seem to be other factors that should necessarily counteract the overall influence of advertising on consumption, except for the substitution that can take place among beverages and brands.

3. Theoretical Model Specification

Differential Demand Systems

Following Theil (1975/76, 1980), Clements (1987), and Batten (1993), write [p.sub.i], [q.sub.i] for the price and quantity of the ith commodity and let p = [[p.sub.i]] and q = [[q.sub.i]] denote the corresponding vectors. The consumer chooses [q.sub.1], ..., [q.sub.n] to maximize any utility function, u(q), subject to a linear budget constraint M = [Sigma] (pq). Ignoring advertising and other variables, the optimal quantities can be expressed in terms of the Marshallian demand functions [q.sub.i] = [f.sub.i](M, [p.sub.1], ..., p.), which can be rewritten in a growth-accounting form as

d ln [q.sub.i] = [[Eta].sub.i] (d ln M - [summation over J] [w.sub.j]d ln [p.sub.j]) + [summation over J] [[Epsilon].sub.ij]d ln [p.sub.j], i, j, ..., n, (1)

where [[Eta].sub.i] is the income elasticity of demand, [[Epsilon].sub.ij] is the Slutsky compensated price elasticity, and [w.sub.j] = ([p.sub.j][q.sub.j])/M is the budget share of the jth commodity, [w.sub.j] [greater than] 0 and [Sigma] [w.sub.j] = 1. The right-hand-side terms in Equation (1) are the income and substitution effects of price changes, where the term in parentheses is the growth rate of real income. The left-hand side is the logarithmic differential or percentage growth rate of real consumption of the ith commodity in the consumer's budget. The Rotterdam model is obtained by multiplying Equation (1) by [w.sub.i] to yield(7)

[Mathematical Expression Omitted], (2)

where [[Theta].sub.i] = [w.sub.i][[Eta].sub.i] = [Delta]([p.sub.i][q.sub.i]/[Delta]M is the ith marginal budget share, d(ln Q) = [Sigma] [w.sub.i]d(ln [q.sub.i]) is the Divisia volume index of the change in real income, and [[Pi].sub.] = [w.sub.i][[Epsilon].sub.ij] is the (i,j)th Slutsky coefficient. Empirically, the Rotterdam model treats [[Theta].sub.i] and [[Pi].sub.ij] as parameters.

In the following empirical estimation, all differentials in Equation (2) are replaced by finite fourth differences, for example, D[x.sub.t] = ln([x.sub.t]) - ln([x.sub.t-4]), and the budget shares are replaced by arithmetic moving averages, [Mathematical Expression Omitted]. The addition of intercept terms reflects exogenous growth in each dependent variable due to autonomous trend factors. With the addition of error terms and time subscripts, the absolute price version of the Rotterdam model is given by

[Mathematical Expression Omitted], (3)

where [[Alpha].sub.i] is the intercept and each error term [e.sub.it] is assumed to be serially independent and normally distributed, with a zero mean and constant contemporaneous covariance matrix. To stay within the budget constraint, the trend and error terms in Equation (3) must sum to zero. Imposing the Engel and Slutsky adding-up conditions means that one demand equation can be omitted while the remaining equations are estimated jointly. The estimation method is full-information maximum likelihood (FIML) using the DEMMOD-3 software package (Barren et al. 1989). Appropriate tests are conducted for the theoretical restrictions imposed by homogeneity, symmetry, negativity, and complete preference independence.

Alcohol Advertising

The demand restrictions that can be placed on Equation (3) deal with the response of consumption to changes in prices and income. When advertising is added to a demand system, the only necessary restriction is the adding-up restriction; that is, to satisfy the budget constraint, advertising that increases the demand for one product must be offset by decreases in demand for one or more other products. Imposing adding-up, I first test the demand restrictions using an empirical model that includes total advertising by beverage and then augment this model with several media advertising variables. With two types of advertising and three beverages, there are a maximum of 12 coefficients to estimate. The advertising coefficients are estimated in an unrestricted form, except for the adding-up condition, and these variables are specified as four-quarter differences of logs of real advertising per capita. Although more formal specifications are possible, previously reported results for alcohol advertising using more restrictive specifications are similar to the results in the present paper. Because advertising may be endogenous (Berndt 1991), several specification tests also are conducted to examine this possible condition.(8)

Conditional and Composite Demand Functions

To reduce the number of parameters that must be estimated, assume that the utility function is additive over group utility functions. Given block independence, the n goods are divided into G [less than] n groups, written [S.sub.1], ..., [S.sub.G], such that each good belongs to only one group. The G composite demand functions specify the allocation of total expenditures to each group. This allocation depends on total real income and the prices of the group. The conditional (within-group) demand function for a good contains only the prices of the goods belonging to the same group, real expenditures on the group, and real advertising of each of the goods. Given these [TABULAR DATA FOR TABLE 2 OMITTED] restrictions, let total expenditures be divided into alcoholic beverages and all other goods (AOG). Let the three alcoholic beverages (beer, wine, spirits) constitute the gth group. Adding the advertising variables and using primes to denote the conditional parameters, Equation (3) can be rewritten as a conditional demand model for the three beverages given by

[Mathematical Expression Omitted], (4)

where [Mathematical Expression Omitted] is the conditional budget share of the ith beverage, [[Theta][prime].sub.i] is the conditional marginal share, [Q.sub.gt] is the alcohol group's Divisia volume index, [a.sub.jt] is the log of real advertising of the jth beverage, and [[Beta][prime].sub.ij] is the coefficient of advertising. To estimate Equation (4), I assumed the parameters [[Theta][prime].sub.i], [[Pi][prime].sub.ij], and [[Beta][prime].sub.ij] to be constant coefficients. Each associated elasticity is obtained by dividing the estimated coefficient by the budget share; thus, the elasticities vary inversely to the budget shares. Empirical results for Equation (4) are given in section 4.

Given this formulation, the aggregate total or composite demand for ethanol is expressed in terms of total expenditure Q, and the relative price of the alcohol group as

[Mathematical Expression Omitted], (5)

where [Mathematical Expression Omitted] is the budget share of the alcohol group; [[Theta].sub.g] is the marginal budget share; [Phi] is the income flexibility coefficient (Clements 1987), [Mathematical Expression Omitted] and [Mathematical Expression Omitted] are the Frisch price indexes for alcohol and all n goods, respectively; and [E.sub.gt] is the error term. The alcohol price index in Equation (5) can be approximated by its Divisia index, and [Mathematical Expression Omitted] can be approximated by a marginal-share-weighted average of the Divisia price index for alcohol and a price index for all other goods, [P.sub.ot]. Adding the advertising variables and collecting terms, the composite demand model is given by

[Mathematical Expression Omitted]. (6)

[TABULAR DATA FOR TABLE 3 OMITTED]

Section 5 reports empirical results for Equation (6), and several alternative specifications are considered for the advertising variables, including aggregates by beverage and media. The composite model also will be augmented with two demographic variables.(9) I also report selective results for variables for the minimum legal drinking age and for the unemployment rate.

4. Empirical Results: Conditional Beverage Demand

To examine the effects of different specifications of the dependent variables, I estimated the conditional model using both beverage (fluid) gallons per capita and ethanol gallons per capita obtained from each beverage. Because the empirical results are similar, the discussion concentrates on the use of ethanol gallons per capita as dependent variables.(10) I first test for the possible endogeneity of advertising, aggregating across all media, and then examine the demand restrictions. The first set of results includes total advertising by beverage as explanatory variables in addition to the constant term, beverage prices, and conditional real income (measured by total consumption in fluid or ethanol units). Given this model, the Hausman-Wu specification test was used to examine the advertising variables for endogeneity in which the identifying variables are media-specific prices of advertising messages. This test was conducted on an individual equation basis and jointly for the system of equations using the residual values from the fitted advertising equations. Neither of these specification tests indicates that advertising might be endogenous at the beverage level. Furthermore, Table 2 shows the results of a likelihood ratio test comparing the conditional model with no advertising to specifications in which advertising is treated as exogenous or endogenous. This test also indicates that the "ads exogenous" model is the preferred specification for the demand system. Because it is often claimed that advertising has a stock, or carry-over, effect, I also estimated models in which the real advertising growth variables enter with one- and two-period lags. These variables were insignificant, as were tests for advertising-squared terms.

Table 2 also shows the tests for the restrictions on the income and price coefficients for each of the three different advertising specifications. Homogeneity and symmetry are never rejected, and complete preference independence cannot be rejected if advertising is exogenous or omitted. In the following discussion, this latter result is ignored in favor of a less restrictive model that imposes only the classical demand restrictions, including negativity. The advertising results are not affected by this choice of restrictions on prices and income.

The conditional model was then estimated by full-information maximum likelihood (additional details are contained in Nelson 1998). Two sets of empirical results were obtained for the advertising specification: (i) total real advertising by beverage and (ii) real advertising by media and beverage. Table 3 presents the conditional elasticities and standard errors; the elasticities are evaluated at the sample means of the budget shares. Using the total ads specification, all the conditional income (volume) elasticities are significantly positive: beer, 0.76; wine, 1.72; and spirits, 1.13. All six of the own-price elasticities are significantly negative. The conditional demand for wine is the most price-elastic demand (-0.67), followed by beer (-0.20) and spirits (-0.16). All the cross-price elasticities are positive, and five are statistically significant. The results using beverage (fluid) gallons are similar to the results using ethanol gallons. In the top half of Table 3, advertising by beverage is aggregated across all media. Statistically positive own-advertising elasticities are found for wine demand (0.07) but not for beer or spirits. The only statistically negative cross-advertising effects are beer (-0.04) and wine (-0.03) in the spirits demand function. Again, the results using beverage units and ethanol units are similar in all cases. The intercept terms, which represent the exogenous growth rates for the sample period, were significantly positive for beer (0.469 for fluid gallons, 0.478 for ethanol gallons), positive for wine (0.178, 0.174), and negative for spirits (-0.647, -0.652).

Next, total advertising was broken into print and broadcast advertising for beer and wine and print and outdoor advertising for spirits. These elasticity estimates are reported in the lower half of Table 3. Print advertising is significantly positive for beer demand measured in fluid gallons (0.01). Broadcast advertising of beer is insignificant for beer and positive for wine (0.08). Wine print advertising is negative for beer (-0.01) and positive for wine (0.03). Wine broadcast advertising is positive for wine (0.05) and negative for spirits (-0.03). Spirits print advertising is positive for beer (0.02) and negative for wine (-0.11). The own-demand effect of print or outdoor advertising of spirits is statistically insignificant in all cases.

The elasticities obtained using quarterly data on U.S. alcoholic beverage consumption and advertising are generally smaller than those obtained using annual data (see Nelson and Moran 1995). This finding is similar to Duffy's (1991) results using quarterly and annual data for the United Kingdom and confirms theoretical and empirical results reported by Bass and Leone (1983). Overall, the effects of advertising on beverage consumption are very modest and relatively unimportant in terms of magnitude. There is little or no evidence that beer and wine broadcast advertising alter drinking choices in a fundamental way, especially when contrasted to the significant effects of exogenous trends, prices, and real income. For example, consider the effect of a 50% increase in print and broadcast advertising of wine. This would increase the annual demand for wine by 4%, or an additional 0.1 gallons of wine per capita (0.013 gallons of ethanol per capita). Holding total consumption constant, there also would be a net reduction in the demand for spirits of 1.5% (-0.01 gallons of ethanol per capita) and little or no change in the demand for beer. These advertising effects are very small, indicating advertising's minor role in substitution among beverages and its apparently large role in determining brand shares within a beverage category. Because these results hold total consumption constant, the next question addressed is the effect of advertising on total ethanol consumption.

5. Empirical Results: Composite Alcohol Demand

The second empirical task is to estimate the composite demand for ethanol, conditional on the composite price of alcohol, the price of all other goods, real personal consumption expenditures, two demographic variables, and several alternative measures of real advertising. The composite model is estimated subject to the constraint of complete preference independence. [TABULAR DATA FOR TABLE 4 OMITTED]This means that the price and income coefficients are proportional to each other and that the model has one relative price. As with the analysis of beverage demand, I first tested for possible endogeneity of advertising, but this hypothesis was again not supported by a Hausman-Wu test. Given this result, Table 4 displays the empirical results for five different specifications of the composite model. The first regression omits advertising. The total demand for ethanol depends significantly on the real price of ethanol and the two demographic variables. Total demand for ethanol is not affected significantly by real income as measured by total personal consumption expenditures. However, the demographic variables are significant in all cases except one. The negative demand elasticity for older consumers is larger in absolute value than the positive elasticity for younger consumers. Thus, per capita alcohol consumption in the United States can be expected to decline due to a "graying" effect.

The second regression adds six measures of advertising, but none of the coefficients is close to being statistically significance. Aggregating by media (regression 3), by beverages (regression 4), or by both media and beverages (regression 5) does not alter this basic result. None of these regressions passes a likelihood ratio test for joint significance of the advertising variables. Although not statistically significant, the advertising elasticities are about 0.02. An elasticity of 0.02 implies that it would take a 50% increase in all types and forms of advertising to raise the annual demand for ethanol by 1%, or an additional 0.0256 gallons (3.28 ounces) of ethanol per capita. This is approximately equivalent to the ethanol contained in an additional six-pack of beer per year (72 fluid ounces x 0.045 = 3.24 ounces of ethanol). Although none of the advertising coefficients is significantly different from zero, some tentative points can be made. First, the results for aggregate broadcast advertising are just as weak as those for print advertising. Second, there is no evidence of a reinforcement effect between advertising and the demographic variable for youth. When advertising variables are added to the model, the youth variable loses significance in the second and third regressions. Thus, the two variables do not have a double impact on aggregate consumption of ethanol. Third, the coefficient for print advertising of beer appears with a negative sign, suggesting that beer advertising may lead to lower total ethanol consumption as beverage switching occurs. These results also are consistent with the results of Duffy (1990, 1991) using quarterly data for the United Kingdom, and the results for the United States suggest even weaker effects of advertising on alcohol demand.

6. Conclusions

Using quarterly data, this study estimates a differential demand system for consumption of beer, wine, spirits, and total alcohol (ethanol). The study uses disaggregated advertising expenditures by media and media-specific price indexes to obtain deflated series on real advertising per capita. The major conclusions for total ethanol consumption are, first, that alcohol advertising does not have a statistically significant effect on total alcohol consumption and, second, that broadcast advertising is not a significant factor leading to increased consumption of alcohol and does not appear to be different than print advertising despite different rates of use.

At the beverage level, holding total alcohol consumption constant, broadcast and print advertising play minor roles in determining consumption by beverage. First, outdoor advertising of spirits is always insignificant in the demand functions. Second, wine broadcast advertising has a small negative effect on beer consumption and a small positive effect on wine consumption. Third, beer broadcast advertising has a small positive effect on wine consumption. Broadcast advertising of beer does not have a statistically significant effect on the own-demand for beer, suggesting that broadcast advertising affects brand shares and not beverage demand. Despite some major differences in method and data, the results in this study are consistent with the results reported in a number of prior studies, although few of these studies are for the United States. The lack of a major role for alcohol advertising has been noted in several earlier literature reviews (FTC 1985; Smart 1988; Fisher 1993). This result has now been confirmed using quarterly data on different media and beverages.

A preliminary version of this paper was presented at the Federal Trade Commission, Bureau of Economics Seminar Series, May 1997. I wish to thank, without implicating, Mike Baye, John Moran, Mark Roberts, and two anonymous referees for their helpful comments on earlier drafts.

1 Following the repeal of Prohibition, hearings were held before the Senate Committee on Interstate Commerce to prohibit the advertising of alcoholic beverages by radio (U.S. Congress 1939). During the period 1947-1958, numerous hearings were held by the House and Senate Committees on Interstate and Foreign Commerce on proposed legislation to ban print and broadcast advertising of alcohol (U.S. Congress 1952). More recent congressional discussions include hearings on media images of alcohol (1976), labeling of alcoholic beverages (1978-1979), beer and wine advertising (1985). alcohol abuse (1985, 1988-1989), alcohol warning labels (1988), health warnings in advertisements (1990), underage drinking (1991), alcohol labeling and marketing (1992), and alcohol beverage advertising (1992-1993).

2 Previous studies using quarterly data include Duffy (1990, 1991) for the United Kingdom and Franke and Wilcox (1987) for the United States. The systemwide studies by Duffy do not break down advertising by media. Franke and Wilcox employ a break down of advertising but do not use a systemwide approach or include price variables or demographic variables in their study.

3 The ethanol data attempt to account for changes in the average ethanol content of each beverage aggregate due to substitution among beverage-level brands and the introduction of new products (e.g., light beer, wine coolers); see Nelson (1998).

4 Early macroeconomic studies suggested that causality tends to run from consumption to advertising, but more recent work suggests that causality can run in both directions (Jung and Seldon 1995). However, Berndt (1991) argues that reverse causality is more likely to occur when using annual data.

5 Data sources and variable construction methods are described in more detail in Nelson (1998), which is available on request. The basic data sources are (i) apparent quarterly beverage consumption - Beer Institute, Brewer's Almanac, Jobson's Wine Marketing Handbook and Jobson's Liquor Handbook; (ii) conversion to ethanol gallon - NIAAA (1985, 1996) and Kling (1989); (iii) beverage expenditures - yearly expenditure series from the consumption sources and the Bureau of Labor Statistics. CPI Detailed Report (quarterly values by beverage); (iv) income and population - Department of Commerce, Survey of Current Business, and Bureau of the Census, Current Population Reports; and (v) real advertising expenditures - Leading National Advertisers, LNA Class/Brand QTR $ and McCann-Erickson media price indexes (CPM version) from Advertising Age and Media Decisions.

6 Saffer (1993) argues that the relationship between advertising and alcohol consumption has been obscured by the use of annual data, and thus "alcohol advertising expenditures probably should be measured in monthly or quarterly intervals" (p. 129). See also Bass and Leone (1983) and Clarke (1976).

7 Theoretical development of the Rotterdam model, including responses to the McFadden critique, are found in Barnett (1979), Selvanathan (1991), and Theil (1975/76, 1980). For recent surveys, see Clements, Selvanathan, and Selvanathan (1996) and Selvanathan and Clements (1995).

8 Coulson, Moran, and Nelson (1998) examine the lag structure for advertising using cointegration and error correction models for each beverage. Their results are in substantial agreement with the present paper. In particular, long-run own-advertising elasticities are similar in magnitude for beer and spirits (0.027 to 0.085) and slightly larger for wine (0.156 to 0.189).

9 Block independence implies that groups of goods are preference independent and that the own-price coefficient for the group is proportional to its income elasticity. Equation (6) thus contains only one deflated price. Further, Thiel's (1975/76, 1980) theory of rational random behavior implies that the disturbances [E.sub.gt] and [e[prime].sup.it] are independent, and the conditional income term in Equation (4) is thus predetermined.

10 I also estimated the conditional and the composite models including quarterly unemployment as an explanatory variable. However, this variable was never close to significant and had very small elasticities. I also tested for the possibility of a structural shift after 1981 (due to new products, etc.), but this effect was not found. Finally, I constructed a weighted-average variable for the minimum legal drinking age. This variable was insignificant and positive in the beer and wine demand equations and negative and significant for spirits. However, the constant term for spirits was reduced by almost the same magnitude, suggesting that the drinking-age variable is capturing other long-term changes in drinking patterns, demographic changes, and so on.

References

Barnett, William A. 1979. Theoretical foundations for the Rotterdam model. Review of Economic Studies 46:109-30.

Barten, Anton P. 1993. Consumer allocation models: Choice of functional form. Empirical Economics 18:129-58.

Barten, Anton P., et al. 1989. Users' guide to DEMMOD-3. Leuven, Belgium: Katholieke Universiteit Leuven.

Bass, Frank M., and Robert P. Leone. 1983. Temporal aggregation, the data interval bias, and empirical estimation of bimonthly relations from annual data. Management Science 29:1-11.

Berndt, Ernst R. 1991. The practice of econometrics. Reading, MA: Addison-Wesley.

Center for Science in the Public Interest. 1983. Omnibus petition for regulation of unfair and deceptive alcoholic beverage advertising and marketing practices. Docket No. 209-46 before the Federal Trade Commission, Center for Science in the Public Interest, Washington, DC.

Center for Science in the Public Interest. 1995. Double dip: The simultaneous decline of alcohol advertising and alcohol problems in the United States. Washington, DC: Center for Science in the Public Interest.

Clarke, Darral G. 1976. Econometric measurement of the duration of advertising effect on sales. Journal of Marketing Research 13:345-57.

Clements, Kenneth W. 1987. Alternative approaches to consumption theory [and] The demand for groups of goods and conditional demand. In Applied demand analysis: Results for system-wide approaches, edited by Henri Theil and Kenneth W. Clements. Cambridge, MA: Ballinger, pp. 1-35, 163-84.

Clements, Kenneth W., E. Antony Selvanathan, and Saroja Selvanathan. 1996. Applied demand analysis: A survey. Economic Record 72:63-81.

Coulson, N. Edward, John R. Moran, and Jon P. Nelson. 1998. The long-run demand for alcoholic beverages: A cointegration analysis. Unpublished paper, Pennsylvania State University.

Duffy, Martyn H. 1990. Advertising and alcoholic drink demand in the UK: Some further Rotterdam model estimates. International Journal of Advertising 9:247-57.

Duffy, Martyn H. 1991. Advertising and the consumption of tobacco and alcoholic drink: A system-wide analysis. Scottish Journal of Political Economy 38:369-85.

Federal Trade Commission. 1985. Recommendations of the staff of the Federal Trade Commission: Omnibus petition for regulation of unfair and deceptive alcoholic beverage advertising and marketing practices. Docket No. 209-46, Federal Trade Commission, Washington, DC.

Fisher, Joseph C. 1993. Advertising, alcohol consumption, and abuse: A worldwide survey. Westport, CT: Greenwood Press.

Franke, George R., and Gary B. Wilcox. 1987. Alcoholic beverage advertising and consumption in the United States, 1964-1984. Journal of Advertising 16:22-30.

Goel, Rajeev K., and Matthew J. Morey. 1995. The interdependence of cigarette and liquor demand. Southern Economic Journal 62:451-9.

Hilton, Michael E., and Walter B. Clark. 1987. Changes in American drinking patterns, 1967-1984. Journal of Studies on Alcohol 54:515-22.

Jung, Chulho, and Barry J. Seldon. 1995. The macroeconomic relationship between advertising and consumption. Southern Economic Journal 61:577 87.

Kling, William. 1989. Measurement of ethanol consumed in distilled spirits. Journal of Studies on Alcohol 50:456-60.

Lee, Byunglak, and Victor J. Tremblay. 1992. Advertising and the U.S. market demand for beer. Applied Economics 24: 69-76.

National Institute on Alcohol Abuse and Alcoholism. 1985. U.S. apparent consumption of alcoholic beverages based on state sales, taxation or receipt data. Washington, DC: National Institute on Alcohol Abuse and Alcoholism.

National Institute on Alcohol Abuse and Alcoholism. 1996. Apparent per capita alcohol consumption: National, state, and regional trends, 1977-94. Washington, DC: National Institute on Alcohol Abuse and Alcoholism.

Nelson, Jon P. 1997. Economic and demographic factors in U.S. alcohol demand: A growth-accounting analysis. Empirical Economics 22:83-102.

Nelson, Jon P. 1998. Broadcast advertising and U.S. demand for alcoholic beverages: Systemwide estimates with quarterly data. Unpublished paper, Pennsylvania State University.

Nelson, Jon P., and John R. Moran. 1995. Advertising and U.S. alcoholic beverage demand: System-wide estimates. Applied Economics 27:1225-36.

Saffer, Henry. 1993. Advertising under the influence. In Economics and the prevention of alcohol-related problems, edited by Michael E. Hilton and Gregory Bloss. Washington, DC: National Institute on Alcohol Abuse and Alcoholism, pp. 125-40.

Selvanathan, E. Antony. 1989. Advertising and alcohol demand in the UK: Further results. International Journal of Advertising 8:181-88.

Selvanathan, E. Antony. 1991. Further results on aggregation of differential demand systems. Review of Economic Studies 58:799-805.

Selvanathan, E. Antony, and Kenneth W. Clements. 1995. Recent developments in applied demand analysis: Alcohol, advertising and global consumption. Berlin: Springer-Verlag.

Smart, Reginald G. 1988. Does alcohol advertising affect overall consumption? A review of empirical studies. Journal of Studies on Alcohol 49:314-23.

Theil, Henri. 1975/76. Theory and measurement of consumer demand. Volumes 1 and 2. Amsterdam: North-Holland.

Theil, Henri. 1980. The system-wide approach to microeconomics. Chicago: University of Chicago Press.

U.S. Congress, House Committee on Energy and Commerce. 1990. Health warnings on alcoholic beverage advertisements. Hearings before the Subcommittee on Transportation and Hazardous Materials, 100th Congress, 1st Session. Washington, DC: U.S. Government Printing Office.

U.S. Congress, Senate Committee on Interstate Commerce. 1939. To prohibit the advertising of alcoholic beverages by radio. Hearings before the Subcommittee on S. 517, 76th Congress, 1st Session. Washington, DC: U.S. Government Printing Office.

U.S. Congress, Senate Committee on Interstate and Foreign Commerce, 1952. Liquor advertising over radio and television. Hearings on S. 2444, 88th Congress, 2nd Session. Washington, DC: U.S. Government Printing Office.