Objective This study tests relationships between the volume of advertisements (ads; public service announcements (PSAs)) employed in state antismoking campaigns, use of different themes and stylistic features in these PSAs, and state youth smoking prevalence between 1999 and 2005.
Methods We merged commercially available data on televised antismoking PSAs that aired between 1998 and 2004 with data on state tobacco control activity to test the relationship between the volume and content of youth-targeted and general-targeted/adult-targeted antismoking PSAs on youth smoking prevalence, controlling for other tobacco control efforts. We use content analysis and ordinary least squares (OLS) regression to assess which thematic and stylistic features employed in state antismoking PSAs are associated with reduced smoking prevalence.
Results A 100-ad increase in the yearly volume of youth-targeted state antismoking PSAs was associated with a 0.1 percentage point decrease in state youth smoking rates in the following year. This relationship was driven by variation in state use of PSAs emphasising health consequences to self or others and anti-industry appeals. Controlling for appearances of these themes, use of graphic imagery and personal testimonials did not predict reduced smoking prevalence. Adult-targeted/general-targeted PSAs were not associated with youth smoking prevalence.
Conclusions Youth-targeted state antitobacco PSAs that emphasise the health consequences of smoking (to oneself and others) and contain anti-industry appeals are associated with reduced youth smoking rates. Future work should avoid typologies that do not account for the co-occurrence of thematic and stylistic content in antitobacco PSAs.
- Public policy
- Social marketing
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Cigarette smoking among US high school students decreased sharply between 1997 and 2003, but the rate of decline in smoking among teenagers has since slowed considerably. The US Surgeon General estimates that there would be 3 million fewer American smokers today if the success in reducing youth tobacco use observed between 1997 and 2003 had been sustained.1 There is strong evidence that comprehensive tobacco control programmes were a major contributor to the observed declines in youth smoking prevalence during this time period.2–4 Mass media campaigns are widely accepted as an integral part of these programmes,4 ,5 and numerous studies link campaign exposure to reduced smoking prevalence in the USA.5–8
A parallel body of work has sought to identify characteristics of effective antismoking advertisements (ads; public service announcements (PSAs)), emphasising the importance of both thematic content and stylistic features.9–14 Some of these studies have classified a small number of ads into mutually exclusive categories that combine information about their theme (eg, health consequences) and style (eg, graphic imagery; personal testimonials).9 ,10 Others have sought to separate out the effects of style and theme by coding larger samples of PSAs for the presence or absence of both types of design characteristics and gauging audience responses to them11–14 or holding the ad theme constant by focusing on the effects of different stylistic factors using messages from a single campaign.15 Most of these studies have focused on short-term (eg, recall, perceived effectiveness, attitudes, intentions) rather than long-term outcomes (eg, smoking behaviour or prevalence).9–13 ,15 In addition, many have implicitly assumed (through their use of mutually exclusive coding procedures) that different types of content do not coexist in the same message—in other words, that PSAs focus on either a health consequences or an anti-industry message, but not both.9–12
Understanding which PSA themes and stylistic elements are associated with declines in state youth smoking rates can contribute to the design of more effective antismoking messages. This study seeks to address many of the limitations of previous work on the topic. We utilise commercially available data on state-sponsored antitobacco television PSAs (those purchased by or airtime donated to a state tobacco control program, called ‘state PSAs’ in the rest of the paper) between 1998 and 2004 to examine the relationship between the overall volume of state PSAs, the use of different combinations of themes and stylistic features in these state PSAs, and state youth smoking prevalence from 1999 to 2005, while controlling for other state and national antismoking programmes and policies. We first use content analysis to identify the degree to which discrete themes and stylistic elements co-occur in youth-targeted and adult-targeted/general-targeted state PSAs appearing between 1998 and 2004. We then use ordinary least squares (OLS) regression models to test the relationship between the volume and content (both thematic and stylistic) of these state PSAs on youth smoking prevalence from 1999 to 2005, controlling for numerous potential confounders.
PSA appearance data and ad content coding
We first obtained a data set of all state-sponsored antitobacco PSAs appearing on television between 1998 and 2004 from TNS, an affiliate of Kantar Media. TNS provided information on the media market in which the state PSA appeared, the time and channel of airing, and the digital ad to permit content coding. The data set includes 1320 unique state PSAs that aired a total of 218 721 times during the time period. TNS also provided ad appearance data for cigarette ads in print media (including location of distribution based on readership data), smoking cessation product adverts in print media and television, and the number of TRUTH ads appearing on television in a particular market, channel and time.
We trained two independent coders to determine the target audience for each of the 1320 unique state PSAs. Coders first classified the state PSAs into one of three groups: youth targeted, adult targeted or general targeted. Youth-targeted ads were judged to have been directed at youth or teens, doing one or more of the following: aiming to deter youth from initiating smoking, discussing peer pressure to smoke, describing assistance for quitting directed at teens or youth, or featuring youth discussing the tobacco industry. Coders were highly reliable in these assessments (Cohen's κ=0.99; table 1); study principal investigators resolved any discrepancies. We split ads into two categories based on coded targeting: youth-targeted PSAs (n=362 unique PSAs, n=39 039 PSA appearances, 18% of all state PSAs) and adult-targeted or general-targeted PSAs (n=958 unique PSAs, n=177 889 state PSA appearances). Most youth-targeted PSAs appeared on the networks FOX (31%), WB (13%), ABC (12%) and NBC (11%), while adult-targeted/general-targeted PSAs were relatively equally distributed largely across the four major broadcast networks ABC (20%), NBC (19%), FOX (17%) and CBS (6%). State PSAs aired more frequently over time, with the most appearances occurring in 2004 (youth n=5207; adult/general n=46 655), 2002 (youth n=13 215; adult/general n=35 047), 2003 (youth n=5036; adult/general n=39 638), and 2001 (youth n=7886; adult/general n=29 515), and with the fewest total appearances in 1998 (youth n=1423; adult/general n=9056), 1999 (youth n=2878; adult/general n=9733), and 2000 (youth n=2302; adult/general n=8245).
The research team next coded for the presence or absence of any reference to specific themes in state PSAs, informed by previous work.16 ,17 We pretested these themes prior to the initiation of final content coding and then trained six independent coders to use the pretested instrument and codebook to content code the state PSAs for theme appearance and two stylistic features that have been the focus of previous work and recommended in the design of PSAs.10–12 Single state PSAs could feature multiple themes and styles (table 1).
The most prevalent message theme for youth-targeted PSAs were health consequences to self or others (73%), anti-industry appeals (63%), social consequences to self (44%), efficacy appeals (38%), behavioural directives (31%), and irrationality appeals (12%), with the rest of the themes appearing in less than 10% of the youth-targeted PSAs. One in five (21%) youth-targeted PSAs contained personal testimonies while 10% utilised graphic imagery. Adult-targeted/general-targeted PSAs often focused on health consequences to self or others (73%), efficacy appeals (62%), anti-industry appeals (30%), and irrationality appeals (23%), using either a personal testimonial (15%) or graphic imagery (16%) in about one in six state PSAs. Most state PSAs utilised multiple themes (youth targeted: 1 theme n=2792 appearances, 7% of state PSAs; 2 themes n=11 284 appearances, 29% of PSAs; 3 themes n=14 477 appearances, 37% of PSAs ; 4 themes n=6275 appearances, 16% of PSAs; 5+ themes n=4261 appearances, 11% of PSAs) (adult/general targeted: 1 theme n=45 496 appearances, 25% of PSAs; 2 themes n=69 913 appearances, 39% of PSAs; 3 themes n=44 671 appearances, 25% of PSAs ; 4 themes n=13 729 appearances, 8% of PSAs; 5+ themes n=4479 appearances, 3% of PSAs).
The dependent variable was annual (1999–2005) state youth (12–17 years) cigarette use rates as reported by Substance Abuse and Mental Health Services Administration (SAMHSA), Office of Applied Studies, National Household Survey on Drug Abuse (NHSDA) and defined as the per cent of youth in a state using cigarettes at least once in the past year.18
The independent variables for the study were the volume and content of state PSAs that, based on TNS data, appeared in media markets with majority coverage within a particular state in the preceding year (eg, 1998 PSA volume to predict 1999 smoking rates). We used a 1-year lag for all state PSA and state-level variables because reports of smoking in the NHSDA are retrospective (ie, span the previous year; see table 2).5 ,8
We used TNS data to account for the airing of commercial tobacco and antitobacco ads in a state (with a 1-year lag, 1998–2004), including the total: number of print ads for tobacco products per state/year; number of television ads for smoking cessation products (patch, pills, gum, etc) per state/year; number of print ads for smoking cessation products per state/year; and number of ads from the national TRUTH antismoking campaign per state/year (table 2).8
We also accounted for state-level factors used to predict youth smoking rates in previous work. We obtained data on state tobacco control policies and antismoking expenditures from the ImpacTeen Tobacco Control Policy and Prevalence Data, University of Illinois at Chicago, Institute for Health Research and Policy.19 These data provided information on annual: per capita tobacco control funding in the state; state excise tax per pack of cigarettes; an index (range 0–8) of the number of smoking bans in the state in that year (public school, private school, public transit, restaurant, recreational facility, health facility, childcare facility, private worksite facility); per cent of state gross domestic product (GDP) from tobacco production; and the extensiveness of state tobacco control youth access laws in that state (total Alciati score; see table 2).20 In addition, we controlled for several state-level population factors including the percentage of the state population under 18 years of age, percentage of adult population that is African-American, percentage of adult population that is Hispanic/Latino, and percentage of adult population that is unemployed.
We used OLS regression to estimate the relationship between state PSA appearances (in the previous state/year) and state youth smoking rates (in the current state/year), controlling for state/year variation in other tobacco-related ad activity (previous state/year) and the state tobacco control environment (previous state/year). We ran several different regression models. In the first model (labelled ‘model 1’ in the table), we estimated the relationship between the volume of state PSA appearances and youth smoking rates, controlling for potential confounders (other smoking-related ads and state-level variables), with separate models for each state PSA theme and style. In the second model (‘model 2’), we fit a model that included two state PSA variables: the overall volume of youth-targeted PSA appearances and the overall volume of adult-targeted/general-targeted PSA appearances, again controlling for potential confounders. In the third model (‘model 3’), we included all youth-targeted content variables (styles and themes) that were featured in at least 10% of youth-targeted PSA appearances in the same model (controlling for potential confounders). In the fourth model (‘model 4’), we included all adult-targeted/general-targeted PSA content variables (styles and themes) that appeared in at least 10%of state PSA appearances in the same model (controlling for potential confounders). Models 3 and 4 thus isolate the independent contributions of specific thematic and stylistic content on youth smoking prevalence by accounting for the co-occurrence of multiple themes and stylistic content in the same state PSA appearance. We tested for evidence of near-extreme multicollinearity in each model by requesting variance inflation factors (VIFs) for each variable in the model.i
OLS regression models predicting state–year youth smoking rates
Table 3 shows results from OLS regression models predicting state youth smoking rates by state PSA appearance volume, volume of other tobacco-related messaging, and other state-level characteristics. Models 1 and 2 reveal that a 100-ad increase in the yearly volume of state PSA appearances was associated with a 0.1 percentage point decrease in state youth smoking rates in the following year. Models also show that use of three state PSA content features were associated with decreased smoking prevalence: youth-targeted PSA appearances emphasising health consequences to self or others, those emphasising tobacco industry misdeeds, and those using normative appeals. Model 3 reveals that two of these content features, youth-targeted PSA appearances emphasising health consequences to self and others (B=−0.24) and using anti-industry appeals (B=−0.18), remained significant in multivariable models controlling for other ad themes and styles.ii Youth-targeted state PSA appearances featuring explicit behavioural directives were associated with increased state youth smoking prevalence. Many of the themes and styles included in model 3 were strongly correlated with one another (table 4); however, none of the VIFs in model 3 were above 7.5, indicating that the model coefficients are robust to controls for the appearance of other state PSA themes and styles. Notably, neither personal testimony nor graphic imagery was associated with reduced smoking prevalence in any of the youth-targeted state PSA models.
None of the adult-targeted/general-targeted state PSA variables, including the overall volume (in models 1 and 2) or theme-specific/style-specific volume (in models 1 and 4), were significant predictors of youth smoking prevalence (all variable VIFs <10). Furthermore, print cigarette ad volume, print cessation ad volume, televised smoking cessation ad volume, and TRUTH antismoking ad appearances were not associated with reduced youth smoking rates (models 1–4).
Turning to state-level variables, each additional dollar per pack of cigarette excise taxes was associated with a 1.93–1.97 percentage point decline in youth smoking rates; a 1 percentage point increase in state GDP from tobacco farming was linked to a 6.20–6.45 percentage point increase in state youth smoking prevalence. Location bans, per capita tobacco control funding, and youth access laws were not associated with declines in youth smoking rates, although several demographic factors (% youth population, % African-American, % Hispanic) did predict youth smoking rate.
This study offers new evidence that state-sponsored youth-targeted antismoking PSAs are significantly associated with declines in youth smoking observed in the late 1990s and early 2000s in the USA. More importantly, we identify two specific thematic characteristics of youth-targeted state antismoking PSAs that are independently associated with reduced youth smoking rates (although they often co-occur): youth-targeted PSAs emphasising health consequences to self and others, and youth-targeted PSAs emphasising deceptive or unethical tobacco industry behaviour. We also find some evidence that youth-targeted PSAs featuring explicit behavioural directives (‘don't smoke’) are linked to increased state smoking prevalence. We find no evidence that adult-targeted or general-targeted PSAs are associated with youth smoking rates, nor does specific stylistic content (PSAs using personal testimonials or graphic imagery) predict youth prevalence over and above thematic content.
This pattern of findings complicates results from previous studies which have tended to collapse ads into discrete typologies—for example, ads using graphic imagery to convey health effects versus ads describing the behaviour of the tobacco industry.9–15 We find that many of these features tend to co-occur in the same message. While studies that have used typologies have reached conclusions that are similar to those observed here (that PSAs should emphasise health consequences to self and others OR anti-industry themes),9–15 our findings suggest that there may be benefit to emphasising both message themes in the same state PSA (since multivariable model results suggest an additive effect of each theme). This approach has been the hallmark of the TRUTH antismoking campaign, which previous studies have found to be effective in reducing smoking prevalence (although we did not; more on this point below).8
We were surprised that ads utilising stylistic elements of personal testimonials or graphic imagery were not associated with reduced smoking prevalence. We do not suggest that campaigns should cease the use of these stylistic feature, as there is good evidence that personal testimonials and graphic images can draw attention to youth antismoking messages.9 ,10 ,14 ,15 We do suggest, however, that ads emphasising the health consequences of smoking or emphasising poor tobacco industry behaviour may not require the use of graphic images or personal testimonials to be effective at reducing smoking rates. Future work should continue to untangle the effects of antismoking ads with strong health consequences messages and the effects of ads with graphic images (which usually function to convey those consequences).
Findings also echo concerns raised in previous work about the potential negative consequences of using explicit behavioural directives in youth-targeted antismoking ads. Philip Morris’ ‘Think, Don't Smoke’ campaign was criticised for utilising this approach in their so-called antismoking campaign from the early 2000s.21 ,22 Asserting independence is an important part of adolescents’ cognitive and social development, and messages that explicitly threaten their personal freedoms to choose by directing behaviour (“do this, don't do this”) are unlikely to be productive and, as suggested here, may backfire.23
Contrary to previous work, we found no evidence that exposure to TRUTH antismoking ads was associated with declines in youth smoking.8 Our study was designed to examine state-level PSA ad volume on state youth smoking prevalence, whereas the TRUTH campaign was a national effort that was not restricted or targeted by the state. We suspect that limited state-level variation in TRUTH ad exposure may have reduced our chances of detecting any such effects.
Turning to state tobacco control variables, our finding that state excise taxes were associated with reduced state youth smoking prevalence echoes previous work, as does our finding that youth access laws were not associated with these declines.2–4 Contrary to previous work, however, we found no significant association between state tobacco control funding and youth smoking rates.2–4 At the same time, previous studies that have found evidence for effects of state funding on youth smoking have not accounted for media campaign exposure in the same model. Since media campaigns most likely represent the largest expenditure in state tobacco control programmes,24 we suspect that this finding can be explained by the fact that we measured and accounted for the effects of antismoking ad exposure in our study.
We measured state PSA volume at the state level, but PSAs are typically purchased by media markets which do not strictly adhere to state boundaries. While most media markets are located within a particular state, some markets extend across state borders, meaning that our estimates of volume of PSAs aired may under-represent or over-represent the volume of exposure in cities that reside in a media market centred in another state. In addition, our measure of state PSA volume (number of PSAs appearing in a state) also does not account for variations in the size of the audience for a particular state PSA—state PSAs airing at 3:00 are likely to have a smaller audience than PSAs airing at 21:00, yet our models treat these PSAs equivalently. At the same time, we assume that state media campaign planners sought to maximise exposure within state boundaries given the resources available to them, and that the volume of these resources is highly associated with overall tobacco control funding in a state. Thus, the fact that we accounted for per capita state tobacco control expenditures is likely to account for much of the difference in resources available to generate widespread exposure to antismoking PSAs appearing in a state. To test this hypothesis, we ran additional models interacting overall state PSA volume with per capita funding for tobacco control programmes. Results indicate that the main effect of state PSA volume remains significant, although the magnitude of the marginal effect is reduced slightly. This indicates that the effects of airing additional youth-targeted state PSAs on youth smoking prevalence in highly funded states are slightly lower than the effects of additional youth-targeted PSAs in states with less funding for tobacco control programmes. Although we can only speculate, this may reflect a diminished marginal return per state PSA appearance in states where there is other prominent tobacco control activity underway.
We only examined two stylistic features, personal testimonials and graphic imagery, focusing on those that have been the focus of considerable previous research. Future research should explore the potential impact of other stylistic features in shaping ad response. Finally, our study is limited to youth smoking rates, not adult prevalence or quit rates. Other themes and styles may be effective among adult audiences, and our results do not address these outcomes.
State antitobacco youth-targeted PSAs that emphasise health consequences to self and others, as well as negative tobacco industry behaviour, appear more effective at reducing youth smoking rates than those that utilise other themes. Future research should consider avoiding PSA typologies that do not adequately capture the variety of thematic and stylistic content that commonly appear in state antitobacco PSAs.
What this paper adds
Previous research has sought to identify thematic and stylistic characteristics of effective antismoking public service announcements (PSAs), but this work has often focused on short-term outcomes and has examined a limited set of combinations of style and substance.
This study uses commercially available data on state-sponsored antismoking PSAs that aired in the USA between 1998 and 2004 to assess the independent contributions of various thematic and stylistic features that predict lower youth smoking prevalence between 1999 and 2005, controlling for the use of multiple themes and styles per PSA and other tobacco control efforts in the state.
We find that state-sponsored PSAs emphasising health consequences to self or others, as well as those emphasising tobacco industry misdeeds, were significantly associated with reduced youth smoking prevalence.
These results lead us to recommend that researchers avoid using typologies that oversimplify the variety of thematic and stylistic content that commonly appears in youth-targeted, state-sponsored antitobacco PSAs.
The authors would like to thank the Directors of the PhADS Laboratory at Cornell University for the use of their advertising database, and the undergraduate research assistants who worked on this project: Harlan Pittell, John Cantor, Katherine Strausser, Alice Cope and Julie Spaulding.
Contributors JN conceptualised the study, contributed to the interpretation of the data, and led the writing of the manuscript. RA contributed the time of the research assistants on this project, conceptualised the study, supervised the statistical analysis, and helped write the manuscript. SB conceptualised the study, contributed to the interpretation of the data, and helped write the manuscript. TS conducted the statistical analysis and contributed to the interpretation of the data.
Funding This paper benefited from the databases generated by two NIH grants: 5RO1CA094020 An Economic Study of Three decades of Smoking Cessation; and 5R01CA113407-4 Smoking Cessation and Advertising: An Econometric Study.
Competing interests None.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement The state smoking prevalence data are publicly available. The commercial marketing data are proprietary and thus are not able to be shared broadly, as per contractual agreements with Kantar Media.
↵i We originally created separate categories for health consequences to self and consequences to others. However, these variables were very highly correlated and introduced significant problems of near-extreme multicollinearity (VIFs >20) into the models. We thus combined these two variables in to a single content category. We also tried including all content categories, including those found in less than 10% of ads, in models 3 and 4; doing so also introduced multicollinearity issues (VIFs >15) so we removed rarely occurring PSA content from the models.
↵ii We conducted a series of sensitivity analyses to assess whether the statistically coefficient for normative appeals in model 1 was robust to model specification. In short, normative appeals were not statistically significant in multivariable models that controlled for other ad themes and styles. Since including rarely occurring content introduced multicollinearity issues (high VIFs) we did not include this variable in our final multivariable models.