The effect of advertising on tobacco and alcohol consumption.
Saffer, Henry
Researchers study the effects of tobacco and alcohol advertising
because the consumption of these substances is known to have potentially
adverse health consequences. Tobacco use results in illness in
proportion to its consumption, with about one-third of tobacco consumers
dying as a result of these illnesses. Alcohol is different in that about
nine out of 10 adults use alcohol in limited amounts with no adverse
outcomes. The other one in ten abuses alcohol, which results in a range
of negative health and social outcomes including an estimated 100,000
premature deaths per year.
There have been a number of empirical studies on the effects of
tobacco and alcohol advertising. The bulk of these studies indicate that
advertising does not increase tobacco and alcohol consumption. However,
many public health advocacy organizations do not accept these results.
An examination of the methods and data commonly used in empirical
studies provides an explanation for these divergent opinions. The key to
understanding the empirical problems lies in the advertising response
function and the type of data used to measure advertising.
The advertising response function explains the relationship between
consumption and advertising. A brand-level advertising response function
shows that the consumption of a specific brand increases at a decreasing
rate as advertising of that brand increases. That is, the response
function illustrates a diminishing marginal product of advertising. (1)
Ultimately, consumption is completely unresponsive to additional
advertising. The assumptions of the brand-level advertising response
function also can be applied to industry-level advertising. The industry
level includes all brands and products produced in an industry; for
example, the industry level for alcohol would include all brands and
variations of beer, wine, and spirits. The industry-level advertising
response function is assumed to be subject to diminishing marginal
product, as in the case of the brand-level function. The industry-level
response function is different from the brand-level response function,
though, in that advertising-induced sales must come at the expense of
sales of products from other industries. Increases in consumption come
from new consumers, often youths, or from increases by existing
consumers.
The industry-level response function can be defined by measuring
advertising with a time-series of national data. This function also can
be defined by measuring advertising with cross-sectional data from local
markets. The industry-level advertising response functions provide two
simple predictions: first, if advertising is measured at a high enough
level, there will be little or no consumption response; second, the
greater the variance in the advertising data, the greater the
probability of measuring the effect of advertising in the upward sloping
section of the response function.
Most prior studies of tobacco and alcohol advertising use annual or
quarterly national aggregate advertising expenditures as the measure of
advertising, probably because this type of data was, at one time, the
least expensive available. These time-series studies generally find that
advertising has no effect. The oligopolistic nature of the tobacco and
alcohol industries results in competition for market share with
advertising (and other marketing) rather than with price. Indeed, price
competition may set off a price war in which all firms will lose
revenue. Alternatively, the "share of voice"--that is, the
percent of industry-level advertising undertaken by one firm--is
directly proportional to the share of market. The advertising to sales
ratios for tobacco and alcohol companies are about 6 to 9 percent while
the average American firm has an advertising to sales ratio closer to 3
percent. Aggregate national advertising may well be in the range of
near-zero marginal product. The advertising response function predicts
that studies using national aggregate data are not likely to find much
effect of advertising, and the empirical work supports this prediction.
Local advertising, known as spot advertising, is a function of
local cost conditions, demographics, regulations, and other local
factors. As a result, local advertising varies more than aggregate
national advertising. Studies using cross sectional measures of
advertising generally find that is has positive effects; this is
consistent with measurement in the upward sloping portion of the
response function. A few prior studies used cross-sectional advertising
data measured at the individual or local level. These studies generally
found that advertising had positive effects. One possible explanation
for the results from the time-series studies is that the national-level
data, being more aggregated, has less variance and thus leads to
insignificant effects.
The one other common type of research on advertising is the study
of advertising bans. The effect of a ban on the use of one or more media
is substitution into the remaining non-banned media and into other
marketing techniques. This does not necessarily reduce advertising
expenditures. Bans can, however, lower the average product of a given
advertising budget. Advertising and other marketing expenditures may
increase to compensate for the loss of sales attributable to the
downward shift of the response function. If the bans are comprehensive
enough, they may reduce consumption. The empirical work finds some
evidence that bans do reduce consumption.
Counteradvertising, which is designed to reduce consumption, also
fits into the framework of a response function. The counteradvertising
response function slopes downward and is subject to diminishing marginal
product. The levels of counteradvertising that have been undertaken are
small in comparison to advertising. Thus it is likely that these
expenditures are in the falling portion of the counteradvertising
response function. The empirical work finds evidence that
counteradvertising does reduce consumption.
To summarize, the response function predicts that using time-series
aggregate national advertising data probably will lead to finding little
or no effect of advertising. Cross-sectional data measuring local
variations in advertising are more likely to fall in the upward sloping
portion of the advertising response function, and are more likely to
lead to finding a positive effect of advertising. Advertising bans, if
comprehensive enough, may lead to finding effects of advertising on
consumption too. With these predictions in mind I have completed seven
studies which use either cross-sectional advertising data, advertising
ban data, or cross-sectional counteradvertising data.
My most recent study, with Dhaval Dave, examines the effect of
alcohol advertising on alcohol consumption by adolescents. (2) We use
the Monitoring the Future (MTF) the National Longitudinal Survey of
Youth 1997 (NLSY97) datasets for the empirical work. These datasets are
augmented with alcohol advertising data, originating at the market
level, for five media. Use of both the MTF and the NLSY97 datasets
improves the empirical analysis because each has unique advantages. The
large sample size of the MTF makes it possible to estimate regressions
with race and gender-specific subsamples. The panel nature of the NLSY97
makes it possible to estimate individual fixed-effects models. In
addition, similar specifications can be estimated with both datasets.
Since the datasets are independent, the basically consistent findings
increase the confidence in all the results. These results indicate that
blacks consume alcohol less than whites, and this cannot be explained
with the included variables as well as it is for whites. A comparison of
male and female regressions shows that price and advertising effects are
generally larger for females. Models that control for individual
heterogeneity result in larger advertising effects, implying that the
MTF results may understate the effect of alcohol advertising. The
results based on the NLSY97 suggest that a ban on all local alcohol
advertising , which is about one third of all advertising, might reduce
adolescent monthly drinking from about 25 percent to about 21 percent.
For binge drinking, the reduction might be from about 12 percent to
about 7 percent.
An earlier cross-sectional paper examined the effect of alcohol
advertising on motor vehicle fatalities. (3) The data used were
quarterly aggregates for the largest Metropolitan Statistical Areas for
four years. The data indicate that the effect of a ban on broadcast
alcohol advertising would be a reduction of about 2000 highway
fatalities per year. The data also indicate that the elimination of the
tax deductibility of alcohol advertising could reduce alcohol
advertising by about 15 percent, reduce motor vehicle fatalities by
about 1300 deaths per year, and raise about $300 million a year in new
tax revenue.
I also have published two studies on alcohol advertising bans. The
first uses a pooled time series from 17 countries for the period 1970 to
1983. (4) The empirical measures of alcohol abuse are alcohol
consumption, liver cirrhosis mortality rates, and highway fatality
rates. The results show that countries with bans on alcohol advertising
generally have lower levels of alcohol abuse, in particular, the results
indicate that countries with bans on spirits advertising have about 16
percent lower alcohol consumption than countries with no bans and that
countries with bans on beer and wine advertising as well have about 11
per cent lower alcohol consumption than countries with bans on spirits
advertising only. A second study of alcohol advertising bans, with
Dhaval Dave, followed up on the first by using a simultaneous equations
system that treats both alcohol consumption and alcohol advertising bans
as endogenous. (5) This study also updated the dataset with data from 20
countries over 26 years. The primary conclusions of this study are that
alcohol advertising bans decrease alcohol consumption and that alcohol
consumption has a positive effect on the legislation of advertising
bans. The results indicate that an increase of one ban could reduce
alcohol consumption by 5 to 8 percent. Furthermore, recent exogenous decreases in alcohol consumption will decrease the probability of
enactment of new bans and undermine the continuance of existing bans.
Canada, Denmark, New Zealand, and Finland recently have rescinded
alcohol advertising bans. Alcohol consumption in these countries may
increase, or decrease at a slower rate, than would have occurred had
advertising bans remained in place.
I have conducted two studies of tobacco advertising bans as well.
The first, with Frank Chaloupka, uses data from 22 OECD countries over
20 years. (6) We estimate the models with a full set of country and year
fixed effects, along with other time-varying covariates including
tobacco price, income, and the unemployment rate. The effects of the ban
tend to be smaller in the models that include these additional
independent variables. The primary conclusion of this research is that a
comprehensive set of tobacco advertising bans can reduce tobacco
consumption and that a limited set of advertising bans will have little
or no effect. A second study of tobacco advertising bans used data from
102 countries. (7) Since no consistent price or income data are
available for all of these countries, the models only use advertising
bans, dichotomous country, and dichotomous year indicators as
independent variables. Again, the conclusion is that a comprehensive set
of tobacco advertising bans can reduce tobacco consumption and that a
limited set of advertising bans will have little or no effect.
Finally, I am involved currently in a project with Melanie
Wakefield, Chaloupka, and others to examine the effect of tobacco
counteradvertising on youth smoking. This study uses data from Nielsen
Media Research (NMR) on the 75 largest media markets in the United
States between 1998 and 2002. These data were merged with the Monitoring
the Future data. The results show that among eighth, tenth, and twelfth
graders in the 19982000 MTF, exposure to tobacco industry-sponsored or
pharmaceutical company advertising for cessation aids were either
unrelated to, or increased, the probability of smoking, in contrast,
higher exposure to advertisements that were part of a state-sponsored
tobacco control media campaign was significantly associated with lower
levels of smoking.
In conclusion, the theory of an industry advertising response
function is supported by the empirical results from my own prior studies
and reconciles the contrary findings from other prior studies based on
aggregated time-series data. Taken together, these empirical studies
suggest that time series advertising data for alcohol and tobacco are
not appropriate for measuring the effect of advertising. However,
further studies using cross-sectional data are also likely to find
positive effects of advertising; studies of advertising bans will find
effects if they are comprehensive bans; and studies of
counteradvertising are likely to find that counteradvertising reduces
consumption.
(1) At low levels of advertising, increasing marginal product is
also possible.
(2) H. Saffer and D. Dave, "Alcohol Advertising and Alcohol
Consumption by Adolescents," NBER Working Paper No. 9676, May 2003.
(3) H. Saffer, "Alcohol Advertising and Motor Vehicle
Fatalities, Review of Economics and Statistics, 79 (3) (August 1997).
(4) H. Saffer, "Alcohol Advertising Bans and Alcohol Abuse: An
International Perspective," Journal of Health Economics, 10 (1991).
(5) H. Saffer and D. Dave, "Alcohol Consumption and Alcohol
Advertising Bans," Applied Economics, 34 (11) (July 2002).
(6) H. Saffer and F. Chaloupka, "The Effect of Tobacco
Advertising Bans On Tobacco Consumption," Journal of Health
Economics, (19) (2000).
(7) H. Saffer, "The Control of Tobacco Advertising and
Promotion" in Tobacco Control Policies in Developing Countries, P.
Jha and I: Chaloupka, eds., New York: Oxford University Press, 2000.
Henry Saffer, Saffer is a Research Associate in the NBER's
Program on Health Economics and a Professor of economics at Kean
University. His profile appears later in this issue.