Assessing the Impact of Anti-Drug Advertising on Adolescent Drug Consumption:
Results from a Behavioral Economic Model
Lauren G. Block, Ph.D.
Baruch College
Vicki G. Morwitz, Ph.D.
New York University
William P. Putsis, Jr., Ph.D.
London Business School
Subrata K. Sen, Ph.D.
Yale University
FINAL REVISION
(Word Count abstract: 119)
(Word Count text: 2995)
(Word Count references: 690)
(Number of tables: 3)
Corresponding Author: All correspondence concerning this article should be sent to William P. Putsis at the London Business School, Regent's Park, London, NW1 4SA, United Kingdom, +44 (0) 207-706-6733 (telephone), (e-mail).
Acknowledgements
The authors thank Gordon Black and Edgar Adams for providing the data, and Scott Armstrong, Ravi Dhar, George Foltin, Eric Greenleaf, Edward Kaplan, Elisa Montaguti, Bob Shoemaker and three anonymous reviewers for their helpful comments on earlier drafts of the paper. We also thank Elisa Montaguti for computational assistance.
Assessing the Impact of Anti-Drug Advertising on Adolescent Drug Consumption:
Results from a Behavioral Economic Model
Objectives. This study examined whether adolescents’ recall of anti-drug advertising is associated with decreased probability of using illicit drugs, and reduced volume of drug use given use.
Methods. We developed a behavioral economic model of influences on drug consumption using survey data from a nationally representative sample of adolescents to determine the incremental impact of anti-drug advertising.
Results. The results provide evidence that recall of anti-drug advertising is associated with lower probability of marijuana and cocaine/crack use. Recall of anti-drug advertising was not associated with deciding how much marijuana or cocaine/crack to use. Results suggest that individuals predisposed to try marijuana are also predisposed to try cocaine/crack.
Conclusion. These results provide support for the effectiveness of anti-drug advertising programs.
This paper evaluates the effectiveness of the Partnership for a Drug-Free America’s (PDFA) national anti-drug advertisements. PDFA’s donated media makes it the second largest advertiser of a “single product” in the U.S. next to McDonalds.1
We analyze data from the first four years of annual tracking surveys (“Partnership Attitude Tracking Survey,” PATS) conducted by PDFA to independently test whether the commencement of the campaign was associated with a change in adolescents’ drug use. The first “wave” of PATS was initiated during February/March of 1987, three months before the first anti-drug messages were aired. Additional waves were conducted during February/March each year thereafter and measured respondents’ recall of PDFA advertisements. These waves form a “natural experiment,” where respondents during the first wave were not subjected to PDFA advertising, while respondents in subsequent waves were.
A preliminary examination of the PATS data reveals that the percent of respondents who reported marijuana or cocaine/crack use in the past twelve months decreased significantly over the years 1987-1990. Other sources of data corroborate this pattern (e.g., survey data from the University of Michigan's Institute of Social Research, and National Household Survey on Drug Abuse, DAWN, Drug Use Forecasting)2,3. While this overall pattern is consistent with the hypothesis that anti-drug advertising reduces drug consumption, such a simple analysis does not accommodate other potential explanations for changes in drug consumption over time. To adjust for these other factors, we use a detailed behavioral economic model that investigates the relationship between adolescents’ recall of exposure to anti-drug advertising and their probability of using marijuana, cocaine or crack, as well as their volume of drug use given that they are using drugs.
Methods
Data Sources
Data were obtained through multiple-site central location sampling (cf. Black et al. for a detailed report of the PATS methodology)4. Selected sites approximated a national probability sample using cluster sampling. Sites were selected to match the distribution of the population of the contiguous U.S. based on Census data, and were chosen along two dimensions: 1) regional, and 2) urban, suburban and rural distribution of the population. At each site, quotas were established for gender and race. The sample size of adolescents aged 13-17 (sampling locations in parentheses) for 1987-1990 was 797 (96), 1,031 (89), 870 (85), and 1,497 (99), respectively.
Questionnaires were self-administered by respondents in a private facility and returned anonymously in a blank envelope. This data collection method should result in increased willingness to reveal illicit or undesirable behaviors.5 Although evidence indicates that self-reported drug use has a high degree of reliability and validity6,7, we also conducted a detailed analysis of the impact of potential reporting biases on the results.
Theory and Key Constructs
We begin with an individual-level behavioral economic model of drug use, focusing on the impact of advertising. This well-established economic framework provides the rigorous link between the underlying theory and the statistical model needed to estimate individual behavior.8,9,10 We then rely on health behavior theory to select the specific variables used within this empirical specification. The measures used in the analysis represent the predominant benefits and costs to drug use identified by major health behavior theories.11,12,13 Since factor analyses indicated that all the multi-item measures described below load on one factor, items were averaged.
Measures of Drug Consumption. We analyzed marijuana separately from cocaine/crack use since reasons for using drugs differ for specific drugs14. We combined cocaine and crack into one category because 92% of respondents reported using both with equal frequency. Respondents indicated how often in the past 12 months they had used each drug by checking one of seven alternatives: 1=None, 2=Once, 3=2-3, 4=4-9, 5=10-19, 6=20-39, and 7=40+. We used these responses to determine both the percent of respondents who reported using each drug in the past 12 months (MJ USE, CC USE: 0=none, 1=any use in the past 12 months) and the volume of use for those reporting they use each drug. For users of both drugs, we divide their volume of use (MJ VOLUME, CC VOLUME) at the median and consider those below the median light (coded as 0, representing 1 to 9 times) and those above the median heavy users (coded as 1, representing 10 to 40+ times).
Perceived susceptibility. The more adolescents perceive themselves to be susceptible to the negative consequences of drug abuse, the less likely they are to use drugs15. Perceived susceptibility (SUSCEPTMJ, a=.86; SUSCEPTCC, a=.94) was obtained by asking respondents to rate three items (four-point scale) indicating the degree to which people risk harming themselves by using drugs (physically or in other ways), where low scores correspond to no risk .
Perceived severity. The more adolescents perceive the consequences of drug abuse to be severe (SEVERE, a=.88), the less likely they are to use drugs13. Respondents rated four items on a four-point scale (where low scores correspond to no fear at all) indicating the degree they would fear the consequences of getting caught with drugs.
Attitudes Towards Drugs. The more favorable teens’ attitudes toward drugs, the higher their likelihood of using drugs15,16,17. Attitude toward drugs (ATTDRUG, a=.89) was measured by asking respondents to indicate their level of agreement (using a 5-point scale) with fourteen items describing benefits of drug use (high scores indicate unfavorable attitudes towards drugs).
Attitudes Toward Drug Users. A positive attitude toward drug users presents a benefit to drug use as evidenced by national14 and regional17 surveys. Attitude toward drug users (USERMJ, a=.80; USERCC, a=.82) was measured by having respondents indicate whether each of 27 personality characteristics would describe a marijuana, cocaine, or crack user (high scores indicate an unfavorable attitude toward drug users).
Peer Pressure. Drug use is influenced by social norms and peer pressures.18,19 FRIENDMJ and FRIENDCC reflect peer pressure by querying respondents on two items using five-point scales: the number of their friends who use each drug occasionally at parties or social events, and how many of their close friends get “stoned” or “high” on each drug once a week or more (low scores correspond to no close friends).
Drug Availability. The supply or availability of drugs (AVAILMJ, AVAILCC) is also a significant factor in drug use. Respondents rated the difficulty they would have in obtaining each drug on a single-item five-point scale (low scores correspond to very difficult).
Addictive Properties of Drugs. Past drug usage accounts for a significant degree of variability in subsequent drug consumption16,20,21. Prior addiction (HOOKEDMJ, HOOKEDCC with 1=yes and 0=no) was measured by asking respondents whether they ever thought they were hooked on marijuana, cocaine or crack.
Anti-drug Advertising. Recall (ADVERT, a=.81) was measured by asking respondents to read a short description of each ad and indicate how often they had seen each on a three-point scale (where low scores correspond to not at all). All six advertisements were aired nationally, with no known differences in frequency or reach for the intended teenage audience.
Demographic Covariates. Three covariates in our model control for individual heterogeneity. Respondents indicated their gender (0=female, 1=male), race (1=white, 0=other), and whether they lived in an urban or rural area (1=city or suburb of a city, 2=town/village or rural area).
Statistical Methodology22
“Stage 1” – The Decision to Use Marijuana and/or Cocaine/Crack. The probability that a respondent reported using marijuana (MJ USE) and cocaine/crack (CC USE) over the last 12 months is expressed in standard “Probit” formulation23 as a function of both the attributes of the individual (e.g., demographics) and their perceptions of drug use itself (e.g., perceived severity).
We consider three versions of this formulation, each with a slightly different assumption about the relationship between the cocaine/crack and marijuana use decisions. First, we estimate the marijuana and cocaine/crack equations independently, assuming that the decision to try the two drugs is independent. However, empirical research suggests that the process may be sequential, i.e., first one tries marijuana and then cocaine/crack.17, 24 The common syndrome theory 25 suggests that individuals have a “predisposition” to use drugs that manifests itself first in marijuana use. Alternatively, something about the experience of using marijuana could lead people to use harder drugs, such as cocaine/crack. This has been referred to as a “gateway” or “stepping stone” theory.26,27 These three alternatives result in different statistical specifications, allowing us to test the alternative hypotheses using the available data.
“Stage 2” – The Volume Decision. In addition to the “use” choice, we also investigate the decision regarding how much to use (the “volume” decision) given that the individual has reported using marijuana or cocaine/crack. Although the decision regarding how much to use is a continuous one, data limitations (data reported categorically, too few observations in key cells)22 force us to categorize individuals into “light” versus “heavy” users.
This produces a classic sequential choice decision: an individual uses the drug and then, based upon his/her experience and additional information (e.g., anti-drug advertising), he/she decides whether or not to use the drug again.22, 23 Accordingly, for each drug, we first estimate the Stage 1 probability equations and then estimate the probability of being a light versus heavy user conditional on prior use. Thus, each second stage equation is estimated employing a dichotomous dependent variable that takes on a value of one if the respondent is a “heavy” user (zero if he/she is a light user), using only those who have previously used drugs.
“Stage 3” – Evaluating Advertising Effectiveness. The first “wave” of PATS (conducted prior to the initiation of anti-drug advertising) provides us with the data necessary to assess the determinants of drug use in the absence of PDFA advertising (the “control” in our natural experiment). We then can assess the significance of recall of PDFA advertising on use and volume decisions via a series of “treatment” groups consisting of each of the subsequent waves that were exposed to PDFA advertising.
We began by estimating the three sets of probability of use equations (“independent,” “gateway” and “predisposition”) using the Wave 1 data for marijuana and cocaine/crack, respectively. Second, based upon the best fitting of these equations, the second stage regressions on the probability of being a light versus heavy user are estimated, also using the Wave 1 data. This provides us with a detailed analysis of the factors influencing the decision to use and the volume of use for each drug prior to the commencement of PDFA advertising.
Now, one way of assessing the impact of PDFA advertising would be to repeat the Stage 1 and Stage 2 probability equations for Waves 2, 3 and 4, including an additional variable (ADVERT) capturing respondent recall for PDFA advertising. The problem with this approach is that advertising recall may be related to an individual’s prior drug use behavior. For example, a heavy drug user may tune out anti-drug advertising. Accordingly, in measuring the impact of PDFA advertising in Waves 2, 3 and 4, we must control for the endogeneity of any advertising recall measure by adjusting for the probability of use at the individual level. Fortunately, we have a ready-made estimate of an individual’s probability of use from the Wave 1 control group probability equations.
Specifically, using the estimated coefficients from the Wave 1 control group, we predict the probability of use for each individual in Wave 2 (the first wave exposed to PDFA advertising). This provides us with estimates of the probability of use in Wave 2 in the absence of PDFA advertising (since the parameter estimates were generated by control group relationships). YMJ (2) and YCC (2) denote the predicted probability of marijuana and cocaine/crack use in the absence of PDFA advertising in Wave 2.
The probability of using drugs in Wave 2 is expressed as a function of two variables: the probability of using in the absence of PDFA advertising and recall of PDFA advertising (ADVERT). The coefficient on the ADVERT variable provides a test of the impact of PDFA advertising on the two probability of use equations for respondents in Wave 2.
This process was repeated for marijuana and cocaine/crack use in Waves 3 and 4. The same methodology was then employed on the set of users based upon the Stage 2 analysis on the volume of marijuana and cocaine/crack use by existing users.
Since the three-stage methodology uses the results from the Wave 1 “control” group data as the basis for the subsequent analysis, many of the statistical problems associated with self-reported survey data are alleviated. For example, “social desirability bias” is the tendency of respondents to give responses they think are more socially desirable. However, a detailed analysis of this potential reporting bias suggests22 that it only serves to strengthen the results by a) lowering the estimated marginal impact of anti-drug advertising on drug use, and b) inflating coefficient standard errors,28 thereby increasing the likelihood of concluding that advertising has no effect. Thus, our results represent a conservative estimate of the impact of anti-drug advertising.