Examining Truth and State-Sponsored Media Campaigns as a Means of Decreasing Youth Smoking and Related Disparities in the United States

Abstract

Introduction

To analyze the impact of Truth and state-sponsored anti-tobacco media campaigns on youth smoking in the United States, and their potential to reduce tobacco-related health disparities.

Aims and Methods

Our study included data from the 2000–2015 Monitoring the Future study, an annual nationally representative survey of youth in 8th (n = 201 913), 10th (n = 194 468), and 12th grades (n = 178 379). Our primary exposure was Gross Rating Points (GRPs) of Truth or state-sponsored anti-tobacco advertisements, from Nielsen Media Research. Modified Poisson regression was used to assess the impact of a respondent’s GRPs on smoking intentions, past 30-day smoking participation, and first and daily smoking initiation. Additive interactions with sex, parental education, college plans, and race/ethnicity were used to test for differential effects of campaign exposure on each outcome.

Results

Greater campaign exposure (80th vs. 20th GRP percentile) was associated with lower probabilities of smoking intentions among 8th graders, smoking participation among 8th and 12th graders, and initiation among 8th graders. Greater exposure was associated with a greater reduction in the likelihood of smoking participation among 10th and 12th grade males than females; 10th and 12th graders with parents of lower education versus those with a college degree; and 12th graders who did not definitely plan to go to college relative to those who did.

Conclusions

Media campaign exposure was associated with a lower likelihood of youth smoking behaviors. Associations were more pronounced for groups disproportionately affected by smoking, including youth of lower socioeconomic status. Media campaigns may be useful in reducing smoking disparities and improving health equity.

Implications

Few recent studies have investigated the impact of anti-tobacco media campaigns on youth smoking and their potential to reduce tobacco-related health disparities in the United States. We found media campaigns—specifically state-sponsored media campaigns—reduced the likelihood of several smoking outcomes among youth, with some evidence that they mitigate disparities for disproportionately affected groups.

Introduction

Nearly 90% of smokers start smoking prior to the age of 18.¹ If smoking patterns continue, an estimated 5.6 million youth under the age of 18 will die prematurely of illnesses related to smoking.¹ Anti-tobacco media campaigns have become a common strategy to reduce the burden of youth tobacco use in the United States, promoting anti-tobacco attitudes² and helping curb smoking among adolescents and young adults.^3–7 However, many studies that have contributed to this literature utilize data from the early 2000s,^3–6,8 and both the tobacco control landscape and nature of anti-tobacco media campaigns have shifted dramatically since then. Additionally, few studies examine the differential associations between media campaigns have and youth smoking across sex, race/ethnicity, and socioeconomic status (SES) groups, and the findings from those studies are inconsistent.^6,9,10 Our study aims to explore the effectiveness of youth anti-tobacco media campaigns and assess how the impact of campaigns differs by school grade, sex, race/ethnicity, and SES.

Anti-tobacco media campaigns have been launched by state governments and nonprofits such as the Truth Initiative to address tobacco use among youth. These ad campaigns have successfully fostered negative attitudes toward tobacco companies,^2,8 promoted awareness regarding the harms of tobacco products,^5,6,8,11,12 and reduced the likelihood of use and intent to use among youth in the United States,^3–7 through some recent work utilizing longitudinal data suggest that the short-term association between anti-tobacco ad campaigns and favorable smoking behaviors might not be sustained over-time.¹³ Importantly, empirical evidence, however, suggests that the effectiveness of media campaign advertisements in reducing overall youth smoking tendencies may vary by grade,⁴ race/ethnicity,^2,6 state-level tobacco control efforts,¹⁴ and specific media campaigns.^11,12 Still, findings are inconsistent and smoking disparities persist with regard to gender, SES, and race/ethnicity.¹⁵ Due to these disparities, it is critical to understand how anti-tobacco advertisements impact smoking in these vulnerable groups to determine their impact on health equity.

This article examines the impact of Truth and state-sponsored televised media campaign exposure on cigarette smoking tendencies and health equity among youth in the United States. We do so by exploring the possibility of differential associations between media campaign exposure and youth smoking across school grade, sex, race/ethnicity, and SES (operationalized as parental education and college plans), using a nationally representative sample of youth over a longer period of time than previously studied, while also bringing in data reflecting more recent anti-tobacco campaign activities to effectively capture how media campaigns might have differed in effectiveness for individuals that are disproportionately impacted by tobacco.

Methods

Sample of Respondents

We analyzed repeated cross-sectional data from the annual Monitoring the Future (MTF) study, a nationally representative survey of 8th, 10th, and 12th graders in the United States, from 2000 to 2015 (MTF details provided elsewhere).¹⁶ A total of 259 919 8th graders, 247 341 10th graders, and 228 975 12th graders participated in MTF during this time period. Analytic sample sizes, reported for each analysis, varied by outcome and grade because some questions are included on only a subset of randomly distributed survey forms.

Adolescent Smoking Outcomes

We investigated four cigarette smoking outcomes. Smoking intentions (defined as the intent to smoke in next 5 years vs. definitely will not smoke) was assessed among never smokers, while past 30-day smoking participation (Y/N) (hereafter referred to as “smoking participation”) was assessed among all respondents, and first smoking initiation (defined as Y/N) was assessed among never smokers and youth who smoked their first cigarette within their current grade. Among all respondents, smoking participation (Y/N) was assessed. Among those who had not reported smoking daily prior to their current grade, initiation to daily smoking was assessed (Y/N).

County-Level Media Campaign Exposure

Our main exposure of interest was exposure to anti-tobacco television ad campaigns on monitored national networks and cable television across the United States. Ad ratings by month, year, and media market were provided by Nielsen Media Research.^17,18 There are 210 designated market areas (DMAs), or units comprised of counties that are used to measure television ratings by family audiences, known as households, and Nielsen tracks ratings of televised anti-tobacco advertisements from the 75 largest DMAs.¹⁷ Rankings to determine the top 75 DMAs relate to the number of households with televisions in a set media market. Media exposure in our study was measured using gross rating points (GRPs) from these top 75 DMAs, which were then converted to county-level measures to merge to respondent-level data. GRPs are measures of the total ad campaign volume shown to a target audience (any viewers in a specific area) and are calculated as the percentage of the target audience reached by the campaign multiplied by the average number of exposures. We divided GRP measures by 100 to create an exposure measure representing average views per person.^17,18 Primary regression models defined GRPs continuously and compared those at the 80th versus the 20th percentile of the distribution across years. We obtained monthly anti-smoking GRP exposure data for campaigns funded by state health departments (2000–2015) as well as the American Legacy Foundation’s Truth campaign (2000–2008). From the monthly data, we derived two measures: a 4-month depreciated sum and a 12-month non-depreciated sum of anti-tobacco ad exposures.^6,19 To keep consistent with previous literature, we used a depreciation value of 0.3 for previous months of GRP exposure.¹⁹ These measures were subsequently merged to students by the county the student’s school was located in, in addition to the year and month of survey administration. After students were linked with their appropriate exposure, we summed each student’s state and Truth campaign exposures to create a single combined measure for total exposure to anti-smoking media. We used the combined (state + Truth) 4-month depreciated sum exposure in our analyses for the short-term outcomes of smoking intentions in the next 5 years and past 30-day smoking participation. For the smoking initiation outcomes, we used the combined 12-month non-depreciated sum in consideration of the longer time period of possible exposure prior to initiation.¹⁸ Students were only linked to data if they went to school in a market for which media data was available. Hence, after merging media data approximately 78% of students were linked with 4-month media exposure from 2000 to 2015. Slightly fewer students (approximately 75%) had data on 12-month media exposure as no 12-month exposure values were calculated for the 2000 survey year. Respondents at the 20th percentile of our pooled media campaign exposure variable using 4-month depreciated sums received a scaled total of approximately 1.3 GRPs, relative to 14.2 GRPs for those at the 80th percentile. For 12-month non-depreciated sums, respondents at the 20th percentile were potentially exposed to approximately 10.6 GRPs, while those at the 80th were exposed to about 87.6 GRPs.

Respondent’s Sex, Race/Ethnicity, and SES

Sex was coded as male or female. Race/ethnicity was coded as non-Hispanic (NH) White, NH Black, Hispanic, NH Asian, and other NH (including multiple races). Parental education, a proxy for SES, encompassed the highest educational attainment reported for either parent and was grouped as less than high school, high school graduate, some college, or college graduate or higher. Twelfth graders’ college aspirations were defined as “definitely will,” “probably will,” and “probably/definitely won’t” plan to attend college. College plans were also used as proxy measures for SES, as more commonly used proxies such as parental education or income often impact students’ plans to attend college.^20,21

Respondent-Level Covariates

Other student characteristics obtained from MTF and included in all adjusted models were living arrangement (ie, neither mother nor father in household, lives with mother, lives with father, lives with both parents), high school program type (ie, college prep., general, vocational/technical, other/do not know), and mother’s employment status (ie, among 8th/10th graders: not employed, part time, full time; among 12th graders: none, sometimes, most of the time, all of the time). Also included were covariates to control for survey year, defined categorically, and four-category Census region (Midwest, Northeast, South, West).

State-Level Covariates

State-level covariates used in all models were unemployment rate, poverty rate, percent of Hispanic individuals, percent of Black individuals, percent college graduates among the population age 25 and older, and average sale price per pack of 20 cigarettes. Ecologic state-level covariates were used, rather than measures at the DMA or county-level, to keep consistent with much of the work that has been done analyzing anti-tobacco media campaigns.^{6, 18, 22, 23} Annual unemployment and poverty rates by the state were acquired from the University of Kentucky’s Center for Poverty Research.²⁴ Data for annual Hispanic and Black state populations were gathered from the Survey of Epidemiology and End Results population estimates.²⁵ The percentage of the state population with a college degree was obtained from the U.S. Census Bureau’s American Community Survey (ACS) for years 2005–2015.^26,27 Because data from the American Community Survey were unavailable for 2000–2004, state college attainment was linearly interpolated using estimates from the 2000 census and the 2005 American Community Survey 1-year estimates. Cigarette pack sale price was obtained from the Tax Burden on Tobacco²⁸ and adjusted to 2016 dollar values using the Gross Domestic Product Implicit Price Deflator.²⁹ Because average cigarette pack sale prices are recorded on November 1 of each year, we averaged prices between adjacent years to approximate the price for students during the spring survey period.³⁰

Primary Statistical Analyses

We conducted modified Poisson regression models stratified by grade to examine the association between anti-smoking media exposure and each of the four primary smoking behavior outcomes.³¹ Modified Poisson regression models entail the use of robust standard errors as Poisson models have been found to give incorrect standard errors when applied to binary outcomes.³¹ These models replace a logit link with a log link in the generalized linear model; the resulting coefficients can be interpreted as a log relative risk rather than a log relative odds. It also allows for the computation of predicted marginal probabilities for our sample population (average marginal effects, AME).³² Main effect models were first estimated for each outcome by grade. For each outcome and grade, we estimated three additional models, each of which contained a single interaction term between media exposure and either sex, race/ethnicity, or parental educational attainment. Furthermore, an additional interaction was examined between media exposure and college plans among 12th graders. We tested the statistical significance of the additive interactions using AME and we adjusted for multiple testing using a false discovery rate of 5%.³³ All analytic decisions with respect to tests for effect modification were specified a priori.

Multiple Imputation

Data were multiply imputed to correct for missing values (ranging from 2.2% to 12.0% for variables), using sequential regression imputation under the missing at random assumption. Five datasets were imputed separately by grade using imputation models with all variables displayed in Table 1 as well as age, school type, hours worked per week, weekly earnings from a job, weekly earnings from allowances or other sources, ever smoked, grade of smoking a cigarette for the first time, grade of starting daily smoking, ≥5 drinks in a row in the last 2 weeks, marijuana use in the last 30 days, and year of survey administration.

Table 1.

Weighted Descriptive Statistics for all 8th, 10th, and 12th Graders in 30-Day Smoking Participation Sample, Monitoring the Future, 2000–2015

	Sample for 30-day smoking participation
	Grade 8	Grade 10	Grade 12
Variables	Wt. %	Wt. %	Wt. %
Sex
Female	51.2%	51.0%	51.3%
Male	48.8%	49.0%	48.7%
Race/ethnicity
Non-Hispanic White	52.5%	59.3%	59.7%
Non-Hispanic Black	13.3%	13.0%	11.0%
Hispanic	17.9%	13.9%	16.0%
Non-Hispanic Asian	5.0%	4.5%	4.9%
Non-Hispanic Other	11.3%	9.4%	8.4%
Education, parents’ highest
Less than high school	9.2%	7.8%	8.8%
High school	19.2%	18.7%	19.3%
Some college	16.0%	17.9%	20.7%
College or greater	55.5%	55.6%	51.1%
College plans (grade 12)
No, probably/definitely			17.4%
Yes, probably			22.8%
Yes, definitely			59.8%
Living arrangement
Neither mother or father in household	3.8%	3.9%	5.8%
Lives with father	3.7%	3.9%	4.8%
Lives with mother	18.4%	18.1%	21.1%
Lives with father and mother	74.0%	74.1%	68.4%
Employment, mother’s current (grade 8/10)
Not employed	21.3%	20.7%
Part time	19.6%	17.1%
Full time	59.1%	62.2%
Employment, mother’s current (grade 12)
None			14.6%
Sometimes			19.5%
Most of time			18.1%
All the time			47.7%
High school program
College prep.	34.8%	48.8%	52.2%
General	17.6%	24.5%	33.5%
Vocational/technical	4.8%	5.0%	5.1%
Other/do not know	42.8%	21.6%	9.2%
Census region
Northeast	18.8%	24.1%	20.9%
Midwest	21.1%	24.9%	22.7%
South	37.2%	28.3%	32.0%
West	23.0%	22.7%	24.4%
Smoking participation (past 30 days)
No	92.7%	87.2%	79.9%
Yes	7.3%	12.8%	20.1%
State cigarette price (mean $ (SE), range)	5.5 (1.3), 3.4–10.5	5.5 (1.3), 3.4–10.5	5.6 (1.4), 3.6–10.5
Media campaign	8.4 (8.1), 0–49.4	8.3 (8.0), 0–46.2	8.4 (8.1), 0–46.2
State unemployment (mean % (SE), range)	6.3 (2.0), 2.3–13.7	6.5 (2.1), 2.4–13.7	6.4 (2.1), 2.3–13.7
State poverty (mean % (SE), range)	13.3 (2.8), 4.5–23.1	13.0 (2.9), 5.4–23.1	13.2 (2.7), 5.4–22.2
State % Black (mean % (SE), range)	12.3 (7.1), 0.6–59.5	12.3 (7.1), 0.8–48.6	12.0 (6.8), 0.5–59.5
State % Hispanic (mean % (SE), range)	16.6 (12.6), 1.4–47.8	14.9 (12.4), 0.7–48.2	17.0 (12.8), 1.1–48.2
State % college grad (age 25+) (mean % (SE), range)	27.9 (4.8), 17.1–56.7	28.0 (4.6), 14.8–55.0	28.4 (4.5), 17.5–56.7
Unweighted N	201 913	194 468	178 379

Results shown are using imputed data (m = 5)

SE = standard error; wt = weighted.

Sensitivity Analyses

We conducted several sensitivity analyses. First, we conducted complete case analyses for our main results. We also analyzed different functional forms of campaign exposure (quadratic, cubic, square root), to test for non-linearities in associations. Third, we tested whether there were differential effects of media campaign exposure on the four outcome variables over time using interactions between the sex, race/ethnicity, parental education, and college plans with campaign exposure on the multiplicative scale. Fourth, we ran additional regression models in which GRPs were defined continuously and scaled by the SD of GRPs within grades, rather than comparing those at the 80th and 20th percentiles. We used this same SD-scaling approach to measure the impacts of Truth (2000–2008) and state-sponsored ad campaigns (2000–2015), separately.

Software, Survey Design, and Ethics Approval

All analyses were conducted using Stata version 15.0 and accounted for MTF’s complex survey design using strata, school cluster, and individual sample weights.³⁴ IVEware 0.3 was used for multiple imputation of missing data.³⁵ The empirical (sandwich) variance estimators also allow for overdispersion of the Poisson regression model. This study was deemed exempt from review by an Institutional Review Board due to the use of de-identified secondary data.

Results

Descriptive Statistics

Table 1 provides descriptive statistics for the 8th, 10th, and 12th graders in the past 30-day smoking participation sample between 2000 and 2015, for those with corresponding GRP data. The smoking participation sample was comprised of 574 760 students (201 913 8th graders, 194 468 10th graders, and 178 379 12th graders). The percentage of respondents that reported past 30-day smoking participation increased with grade (7.3%, 12.8%, 20.1% for 8th, 10th, and 12th graders, respectively), and the mean 4-month depreciated campaign exposure was between 8.3 and 8.4 GRPs. Demographic breakdowns of the smoking intentions, first initiation, and daily initiation samples are located in Supplementary Table 1.

Main Effects Analysis

Adjusted main effects analyses among 8th graders found respondents at the 80th percentile of GRP exposure (hereafter, GRPs) had a 1.9% lower probability of smoking intentions (AME −0.019, 95% CI: −0.032 to −0.007) relative to those at the 20th percentile (Table 2). Further, greater GRPs was associated with a 1.1% lower probability of smoking in the past 30 days (AME −0.011, 95% CI: −0.016 to −0.007), and a 0.4% lower probability of smoking a cigarette for the first time in the current grade (AME −0.004, 95% CI: −0.007 to −0.001). No significant association was found between GRPs and respondents starting daily smoking (AME −0.002, 95% CI: −0.004 to 0.001).

Table 2.

Average Marginal Effects Change in Media Exposure (State + Truth) From 20th to 80th Percentile of Media Distribution, Monitoring the Future, 2000–2015

	8th Graders		10th Graders		12th Graders
	AME (95% CI)a	p	AME (95% CI)a	p	AME (95% CI)a	p
Smoking intentions 5 years
(4-month depreciated)	−0.019 (−0.032 to -0.007)	.002	−0.003 (−0.012 to 0.006)	.534	−0.005 (−0.015 to 0.005)	.288
N	52 961		44 109		33 433
Smoking participation
(4-month depreciated)	−0.011 (−0.016 to -0.007)	<.001	−0.006 (−0.012 to 0.001)	.056	−0.011 (−0.021 to -0.001)	.031
N	201 913		194 468		178 379
Daily smoking initiation
(12-month non-depreciated)	−0.002 (−0.004 to 0.001)	.084	−0.001 (−0.004 to 0.002)	.432	−0.002 (−0.006 to 0.002)	.437
N	184 883		171 700		73 782
First cigarette initiation
(12-month non-depreciated)	−0.004 (−0.007, −0.001)	.003	−0.002 (−0.005,0.002)	.370	−0.007 (−0.014,-0.001)	.023
N	157 817		133 466		51 140

Results are using imputed data (m = 5). Boldface p-value indicates statistically significant AMEs (p < .05).

AME = average marginal effects; CI = confidence interval.

^aEach AME is estimated from a single model with media campaign exposure as the independent variable. All models control for baseline covariates shown in Table 1 for each grade as well as a year indicator.

Among 10th graders, adjusted main effects analyses showed no significant relationship between GRPs and smoking intentions, past 30-day smoking, smoking the first cigarette in the respondent’s current grade, or initiating daily smoking in the respondent’s current grade.

Among 12th graders, adjusted main effects analyses found that respondents at the 80th percentile of GRPs had a 1.1% lower probability of smoking in the past 30 days relative to those at the 20th percentile (AME −0.011, 95% CI: −0.021 to −0.001), and a 0.7% lower probability of smoking a cigarette for the first time in 12th grade, relative to those at the 20th percentile (AME −0.007, 95% CI: −0.014 to −0.001). No significant associations were found between greater GRPs and smoking intentions or daily smoking initiation.

Effect Modification Analysis

After applying the Benjamini-Hochberg correction for multiple testing, we observed statistically significant interactions between GRPs and sex, and between GRPs and parental education for 30-day smoking among both 10th and 12th graders (Supplementary Table 2). Specifically, greater GRPs were associated with a lower probability of smoking in the past 30 days among males but not females in both grades (Supplementary Figures 1 and 2). The association between GRPs and smoking in the past 30 days for 10th grade students was most pronounced among students whose parental education level was less than high school compared to those whose parental education level was high school graduate or higher (Figure 1). There was no evidence of associations between GRPs and smoking participation for any other parental education levels among 10th graders. In contrast, for 12th graders, we found that greater GRPs were more strongly associated with a lower probability of smoking participation among students who had at least one parent with some college education as well as students with parents having less than high school education. Among students whose highest parental education level was either high school or college or higher, GRPs were minimally associated with smoking participation (Figure 2).

Figure 1.

Differential association of anti-smoking media campaign exposure on 30-day smoking participation among 10th graders, by parental education, monitoring the future, 2000–2015. Results shown are using imputed data (m = 5).

Figure 2.

Differential association of anti-smoking media campaign exposure on 30-day smoking participation among 12th graders, by parental education, monitoring the future, 2000–2015. Results shown are using imputed data (m = 5).

In addition, we observed a statistically significant interaction between GRPs and college plans for 30-day smoking among 12th graders. We found that greater GRPs were more strongly associated with a lower probability of smoking participation among students who did not plan to attend college than students who planned to attend college (Supplementary Figure 3). We found no other interactions by sex, race/ethnicity, parental education, or plans to attend college for any of the other outcomes in any of the other grades after adjusting for multiple testing.

Sensitivity Analysis

In sensitivity analyses, we tested whether results using complete case data were consistent with results derived from multiple-imputed data. We found that the results were similar in terms of statistical significance, as well as magnitude and direction. Sensitivity analyses scaling by the SD of media campaign exposure, as well as regression models analyzing the impact of state-sponsored campaigns only, were similar in terms of significance, direction, and magnitude to our primary results comparing those at the 80th versus the 20th percentile of Truth and state-sponsored ad exposure (Supplementary Tables 3 and 4). The relationship between overall exposure to the Truth campaign (2000–2008) and smoking outcomes was largely null, though higher exposure to the Truth campaign was associated with a higher probability of daily smoking initiation among 10th graders (Supplementary Table 5).

We also tested for different functional forms of campaign exposure using multiple-imputed data. We did not find any higher-order effects of GRPs except for past 30-day smoking participation among 8th graders and smoking intentions among 12th graders. Among 8th graders, GRPs were associated with past 30-day smoking participation in a quadratic form (Supplementary Figure 4). Specifically, while the association roughly mimicked the linear trend for respondents between the 0 and 80th percentiles, the impact appeared to have been larger in reducing the likelihood of smoking participation among respondents in the 100th percentile. Further, there appeared to be a quadratic association between GRPs and smoking intentions among 12th graders, GRPs, with the association largely resembling the linear trend for respondents in lower percentiles, but respondents in the 100th percentile seeing a lower likelihood of smoking intentions than the linear test indicated (Supplementary Figure 5). Nevertheless, these findings were likely due to the relatively small sample of respondents at the upper ends of the exposure distribution.

We also investigated if there were differential associations between our exposure and the four outcome variables over time. We found that time modified the association between GRPs and 30-day smoking among 12th graders. Specifically, respondents at the 80th percentile of the GRP distribution had a lower probability of smoking in the past 30 days than respondents at the 20th percentile in 2008 (Supplementary Figure 6). Time interactions for all other outcomes and grades were not significant after correcting for multiple testing.

Discussion

We found that higher levels of Truth and state-sponsored anti-tobacco GRPs were associated with a lower likelihood of smoking intentions among 8th graders, which is consistent with the majority of studies analyzing the impact of the Truth,^2,8,10 and state campaigns^5,6 on smoking intentions among youth. Regarding smoking participation, we found higher levels of GRPs to be associated with a lower probability of past 30-day smoking participation among 8th and 12th graders, which is consistent with other studies evaluating exposure to Truth,^4,36,37 and state-sponsored campaigns.^5,6,37–41 Finally, main effects models showed higher levels of GRPs were associated with lower likelihoods of initiating smoking among 8th and 12th graders, though no meaningful differences were found with respect to daily smoking in the current grade. No cross-sectional studies to date have measured the impact of Truth or state campaign exposure on smoking initiation, though one longitudinal study found high levels of exposure to the Food and Drug Administration’s “The Real Cost” campaign to be associated with a lower probability of initiating smoking.⁹ Our findings, in addition to the existing cross-sectional literature, suggest there is potential for Truth and state-sponsored anti-tobacco media campaigns to help reduce the likelihood of smoking participation and first smoking initiation, though longitudinal work has suggested the observed impact might not be sustained as youth transition into adulthood.¹³

We also found that media campaign exposure was associated with a reduced likelihood of past 30-day smoking participation among 10th and 12th grade males, but not females. One prior study compared the impact of state-sponsored media campaigns on smoking participation by sex, and found no differences in the association with current smoking.⁶ With respect to all other outcomes, we found no evidence for effect modification of campaign exposure by sex. Only one study has addressed sex-related effect modification by smoking intentions, and also found no evidence of a differential impact on smoking intentions by sex.¹⁰ Furthermore, no studies have examined sex-related effect modification with respect to first cigarette and daily smoking initiation. Overall, our study suggests that campaign exposure may differ in effectiveness by sex and may have a greater impact on reducing the likelihood of smoking participation of males.

Regarding race/ethnicity, we did not observe a differential impact of media campaign exposure with relation to any of the four assessed outcomes. The existing literature base, while sparse, generally supports this finding.^2,6,10 Still, some studies using older campaign data found that media campaigns were associated with lower smoking intentions among Black youth,^2,6 though this could be an artifact of campaign messaging changes over time, as more recent studies have shown no evidence for effect modification by race/ethnicity, at least with respect to the Truth campaign.¹⁰ Future research should evaluate the health equity effects of campaigns that target specific identity groups, such as the Food and Drug Administration’s Fresh Empire campaign.

Finally, in examining differences in the associations between media campaign exposure and smoking outcomes by SES, we found that Truth and state media campaigns had a greater association with a reduced likelihood of past 30-day smoking participation among 10th and 12th graders whose parents had lower educational attainment compared to individuals with higher educational attainment. We also found that media campaigns were associated with a greater reduction in the likelihood of smoking participation among 12th graders who did not definitely plan to attend college compared to students who definitely did. We found no evidence for effect modification with respect to any other outcomes or markers of SES. Only one other study analyzed Truth campaign exposure in relation to SES, and found no differences in smoking intentions by parental education.¹⁰

Overall, the impact of Truth and state-sponsored anti-tobacco campaigns on smoking intentions, first smoking initiation, and daily smoking initiation, does not vary across sociodemographic groups. That said, the association between media campaign exposure and lower likelihoods of smoking outcomes among youth seem to be driven by state-sponsored ad campaigns. Importantly, our findings with respect to past 30-day smoking participation suggest media campaigns could help reduce tobacco-related health disparities with respect to SES, as media campaign exposure was more strongly associated with a lower likelihood of smoking participation among 10th and 12th graders with parents of lower education (compared to respondents with parents of higher educational attainment) and 12th graders that did not definitely plan to go to college (relative to those that did), given previous work has shown individuals of lower SES are generally more adversely impacted by tobacco across the lifespan.⁴² Our study indicates that Truth and state-sponsored anti-tobacco media campaigns may have had greater impact on reducing the likelihood of smoking participation among youth from lower SES households, which would in turn improve tobacco-related health disparities and downstream health equity.

Limitations

One limitation of this paper is that the data are cross-sectional, and thus can provide no evidence regarding causal effects. Cross-sectional studies are susceptible to reverse-causation bias as the temporal sequence of exposure and outcome cannot be assessed. However, our study used GRPs, an exogenous measure of media campaign exposure, which could be preferable to a respondent’s ability to recall having seen an ad, which is subject to recall bias.⁴³ Still, using GRPs as our exposure is also limiting, as it measures the number of advertisements aired in a specific area and not how many advertisements a specific respondent saw – so it is possible we are overestimating an individual’s true exposure. Conversely, it is possible that respondents were exposed to other anti-tobacco media campaigns that were not measured in this study due to data limitations (eg, Truth’s “Become an EX” [EX] campaign and the Food and Drug Administration’s “The Real Cost” campaign had considerable arms outside of television-based advertisement campaigns explored here)^44,45 or the focus on topics outside of youth cigarette smoking (eg, the Centers for Disease Control and Prevention’s adult-focused “Tips from Former Smokers” [Tips] campaign and the Truth “FinishIt” campaign that was heavily centered around e-cigarettes),^44,46 which could mean we are underestimating a respondent’s true exposure to anti-tobacco media campaigns. We attempted to combat the potential for unmeasured confounding by controlling for individual and state-level sociodemographic covariates, as well as changes in the tobacco control landscape, such as changes in smoke-free policy coverage at the county-level, and cigarette taxes at the state-level.

Second, data were utilized only from the 75 largest DMAs, and thus we are missing exposure data for 22% of MTF youth surveyed where 4-month depreciated sums were used, and 25% of MTF youth surveyed where 12-month non-depreciated sums were used. Given the varying sociodemographic composition and smoking outcome distribution between respondents that were and were not included (Supplementary Tables 9–12), this could potentially impact our results. Still, we are confident in our use of data from the 75 largest DMAs, as there is considerable precedent for using these approaches in analyzing anti-tobacco media campaigns.^{5,17,22,23,47} This approach, combined with the large, nationally representative samples from the MTF dataset yield results that are at least valid for the majority of the country, if not the entire United States.

Finally, given the overwhelming focus of Truth and state-sponsored ad campaigns from 2000 to 2015 on smoking, we chose to focus exclusively on cigarette smoking rather than on the use of other nicotine or tobacco products, such as e-cigarettes. Still, many recent media campaigns, such as The Real Cost and Truth’s FinishIT, have focused a considerable amount of resources on other types of tobacco products.^44,45 Future campaign evaluations should be focused more on e-cigarettes, and the feasibility of advertising on platforms other than television, such as social media or streaming services.

In conclusion, Our study found greater exposure to Truth and state-sponsored media campaigns was associated with lower past 30-day smoking participation among 8th and 12th graders, first cigarette initiation among 8th graders, and 5-year intention to smoke among 8th graders. We also found that media campaigns were more effective in reducing the likelihood of past 30-day smoking participation among 10th and 12th grade males relative to females, 10th and 12th graders with parents of lower educational attainment relative to those with a college degree, and 12th graders who did not plan on attending college compared to those that definitely did. Because males and individuals of lower SES are disproportionately impacted by tobacco, media campaigns may be effective tools to help reduce tobacco-related health disparities and improve long-term health equity.

Supplementary Material

A Contributorship Form detailing each author’s specific involvement with this content, as well as any supplementary data, are available online at https://academic.oup.com/ntr.

Funding

Research reported in this publication was supported by the National Cancer Institute of the National Institutes of Health (grant numbers R37CA214787 and P30-CA-46592) and the National Institute on Drug Abuse of the National Institutes of Health (grant number R01DA001411). The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Declaration of Interests

The authors declare that there are no conflicts of interest.