Causal effect estimates of online e-cigarette marketing exposure on future e-cigarette harm perception and use Authors: Paul T. Harrell, Nicholas T. Williams, Kara E. Rudolph, Rachel Ayala Guzman, Ateeqa Ijaz, Nicole Rychagov, Aaron Null, Silvia S. Martins Published: 2026 ABSTRACT BackgroundGlobally, over 10 million youth use e-cigarettes; United States use, in particular, dramatically rose in the late 2010’s, exceeding year-over-year increases from any other substance in over four decades. This rise has been partially attributed to youth online e-cigarette marketing exposure, but has not been appropriately studied.MethodsWe examined youth data from the Population Assessment of Tobacco and Health (PATH) spanning the dramatic increase period (Waves 4, 4.5, and 5; 2016–2019). We estimated average risk differences (RD) for Wave 5 e-cigarette harmfulness perception and use comparing all youth versus no youth reporting past-month online e-cigarette marketing exposure in Waves 4 and 4.5. We used a doubly robust, nonparametric targeted minimum loss-based estimator (TMLE) to estimate RD, incorporating PATH survey weights. Initial analyses adjusted for demographics, mental health issues, and other forms of e-cigarette marketing. Subsequent analyses adjusted for frequency of social media use, other substance use, and tobacco (non-e-cigarette) use.ResultsInitial analyses estimated that online marketing was associated with a 9% decrease in e-cigarette harmfulness perception (RD=-0.09, 95% C=-0.12, −0.05), or a Risk Ratio (RR) of 0.85 (95% CI=0.80, 0.91), as well as a 4% increase in current e-cigarette use (RD=0.04, 95% CI=0.02, 0.06; RR=1.36, 95% CI=1.15, 1.62). However, after adjusting for additional potential time-varying confounding variables, point estimates were close to null with 95% confidence intervals spanning the null.DiscussionFrequency of social media use, other substance use, and/or tobacco use may be important confounding variables related to marketing and e-cigarette use that require further investigation. Introduction Youth trajectories of tobacco use are of critical importance in the future of public health. Tobacco dependence generally begins in youth; since 2014, electronic nicotine delivery systems (“e-cigarettes”) are the tobacco product most commonly used by youth in the United States (U.S.) (Jamal et al., 2017, Jamal et al., 2024). The World Health Organization estimates over 14 million youth age 13–15 currently use e-cigarettes (WHO, 2025). E-cigarette use rose dramatically in the U.S. from 2017 to 2018; the absolute increases in nicotine vaping prevalence among 12th and 10th-graders were the largest ever recorded by the Monitoring the Future study in their 44 years continuously tracking dozens of substances (Miech et al., 2019). In addition to changes in e-cigarette product design (Barrington-Trimis and Leventhal, 2018), this rise is often partially attributed to exposure to online e-cigarette marketing (Czaplicki et al., 2020, Kostygina et al., 2023), but this has not been systematically studied with appropriate methods. Marketing exposure has previously been linked to e-cigarette initiation (Donaldson et al., 2022, Vassey et al., 2022) likely by altering risk perceptions (Hendricks and Brandon, 2016). Risk perceptions are important at a population-level as they are modifiable by educational interventions and marketing exposure (Berry et al., 2017, Sheeran et al., 2014) unlike sociodemographic factors. Because marketing exposures and associated policy restrictions have the potential to reach a large audience, even limited levels of efficacy can have a large impact (Borzekowski and Cohen, 2013). We focus here on online marketing—as opposed to other marketing outlets—due to the relative novelty, high prevalence rates, and the potential for targeting by policymakers. Social media is often a source of online marketing exposure (Basch et al., 2021, Laestadius et al., 2019, Massey et al., 2021, Sun et al., 2023) and is used frequently among youth. Most U.S. adolescents (>90%) use at least one online platform and a third of adolescents report being online “almost constantly” (Anderson et al., 2023). Youth online e-cigarette marketing exposure is on the rise (Li et al., 2021). Tobacco brand and product promotions are common on platforms such as TikTok, Facebook, YouTube, Twitter, Instagram, and Snapchat (Basch et al., 2021, Laestadius et al., 2019, Massey et al., 2021, Sun et al., 2023) despite platform policies banning tobacco advertising (Fielding-Singh et al., 2021, Jackler et al., 2019, O'Brien et al., 2020). Online marketing to adolescents is recognized as an important area for policy regulations (Moreno, 2011). Globally, 22 countries completely ban advertising of e-cigarettes, with 53 additional countries with partial advertising bans; however, at least 120 countries, including the United States, have no policies regarding e-cigarette advertising (WHO, 2021). The U.S. public is broadly in support of policies regulating online advertising of tobacco (Blake et al., 2022). Although the Master Settlement Agreement placed strong regulations on traditional tobacco cigarette marketing via traditional methods (i.e., print, television advertising), marketing of other forms of tobacco (e.g., e-cigarettes) or via alternative methods (e.g., internet, smartphone) are largely unregulated. There are regulation challenges, such as free speech issues (Lange et al., 2015), but there are nonetheless opportunities for innovation and changes. Online marketing may be particularly influential for a variety of reasons. Carah and Brodmekel (2021) identify five particular characteristics that differentiate online marketing from other outlets: participatory facilitation (“influencers” who seem like ordinary platform consumers but create content that embeds brands in ways brands themselves often cannot do), algorithmic coordination, context-specific promotions, real-world enmeshment, and purchasing integration (Carah and Brodmerkel, 2021). Online marketing exposure appears to increase odds of susceptibility to tobacco use, lifetime use, and current use (Donaldson et al., 2022, Vassey et al., 2022). However, prior longitudinal research assessing impact of marketing on tobacco use involved fixed effects approaches (Cavazos-Rehg et al., 2021, Donaldson et al., 2022, Perez et al., 2022, Soneji et al., 2018, Stanton et al., 2022, Vassey et al., 2022). This is a limited approach as it can lead to biased estimation when there is time-varying exposure-confounder feedback. Time-varying exposure-confounder feedback occurs when variables at some follow-up time confound future exposure-outcome relationships, but themselves are affected by prior exposures. Controlling for such variables in a standard way is inappropriate because there are different confounding variable sets depending on which exposure timepoint is considered. For example, in considering our current research question., mental health and other forms of tobacco marketing exposure (e.g., retail stores, television) may be affected by online marketing exposure and affect future instances of online marketing exposures and e-cigarette use (Daniel et al., 2013, Imai and Kim, 2019). Despite multiple findings of associations between online tobacco marketing exposure and youth e-cigarette susceptibility and use (e.g., Donaldson et al., 2022; Vassey et al., 2022), previous studies have not used approaches that appropriately adjust for time-varying confounders subject to exposure–confounder feedback. We aim to fill that gap here to reduce bias in estimation of the longitudinal effect of online marketing exposure. Research can help inform future policy by providing clearer evidence regarding the influence of online marketing on youth behavior. In the current paper, we address these concerns by appropriately adjusting for time-varying confounders and the exposure-confounder feedback that results (Díaz et al., 2023). Further, by adjusting for other forms of tobacco marketing exposure, we can help understand the unique influence of online marketing exposure as separate from other marketing outlets. We estimated the extent to which e-cigarette harm perception and use would differ had everyone been exposed to online e-cigarette marketing versus had no one been exposed to online e-cigarette marketing, adjusting for potential confounding variables. We hypothesized that exposure to online e-cigarette marketing would reduce reports of e-cigarette harmfulness perception and increase current e-cigarette use. Methods Data source and participants Study datasets were from the Wave 4 (December 1, 2016 – January 3, 2018), Wave 4.5 (December 1, 2017 – December 1, 2018), and Wave 5 (December 1, 2018 – November 30, 2019) youth Restricted Use Files of the Population Assessment of Tobacco and Health (PATH) Study. The PATH Study is a longitudinal cohort study which generates a nationally representative sample of non-institutionalized youth (aged 12–17 years) in the U.S. (Hyland et al., 2017). The PATH Study’s response rate at Wave 4 was 70.6% for youth. Retention rates for Wave 4.5 and Wave 5 among Wave 4 youth respondents were 89.1% and 83.5% respectively (USDHHS, 2024). The analyses of this study were performed on a subsample of youth (n = 8548) who completed all three survey waves (Wave 4, Wave 4.5, and Wave 5). Predictor variable: online e-cigarette marketing exposure Past month online e-cigarette marketing exposure (yes/no) was assessed with the question, “In the past 30 days, have you noticed e-cigarette or other electronic nicotine products being advertised in any of the following places? Choose all that apply.” Nine options were provided, including “on websites or social media sites.” Respondents who selected this option were considered exposed to online e-cigarette marketing. Missingness ranged from 1.6% to 2.3% across the 3 waves. Outcome variables: e-cigarette harm perception and use E-cigarette harm perception was measured across the waves (4, 4.5, and 5) and was assessed with the question, “How much do you think people harm themselves when they use e-cigarettes or other electronic nicotine products?” Responses were collapsed into “No harm,” “A little harm,” and “Some harm” versus “A lot of harm.” Respondents that reported “A lot of harm” were coded as 1; all others were coded as 0. Current e-cigarette use (Yes/No) was derived from the question measuring using e-cigarettes within the past 30 days. Missingness ranged from 0.2% to 0.9%. Individual time-invariant covariates Wave 4 individual time-invariant variables were included as covariates in our analysis. These included the following variables: age, gender (male/female), race/ethnicity (Hispanic, non-Hispanic white, non-Hispanic Black, non-Hispanic other), highest parental education (GED or less, high-school, some college but no degree, associate degree, bachelor’s degree and graduate degree), and household income (24,999 or less, 25,000–49,900, 50,000–99,999 and 100,000 or more). Missingness was 0.8% for education and 4.6% for income. There were no missing data on other time-invariant covariates. Individual time-varying covariates We conducted two sets of analyses involving two different sets of individual time-varying covariates. The first set involved psychological symptoms and marketing exposure outlets. Regarding psychological symptoms, the Global Appraisal of Individual Needs-Short Screener (GAIN-SS) items assessed externalizing and internalizing symptoms, identifying individuals at risk for mental health disorders using a continuous measure divided into three categories of symptom severity: no/low (0–1 symptoms), moderate (2–3 symptoms), and high (4 + symptoms), as used in prior research (Conway et al., 2018, McDonell et al., 2009). Other forms of e-cigarette marketing were assessed with the same question as online marketing (“In the past 30 days, have you noticed e-cigarette or other electronic nicotine products being advertised in any of the following places?”). These additional e-cigarette marketing outlets fell into the categories of retail stores, television, billboards, newspapers/magazines, radio, events, clubs/bars, and other. The next set of time-varying covariates assessed at each wave were social media use frequency, use of tobacco (non-e-cigarette) products, and use of other substances. Derived variables for frequency of social media use (4 levels: never/no account, less than once a day, once a day, and more than once a day) were created for each wave by combining PATH variables on if youth have a social media account and social media use frequency. Separate binary derived variables were created for marijuana, alcohol, and other substance use in the past 12 months for each wave. The other substance use variable included cocaine, crack, heroin, inhalants, solvents, hallucinogens, and stimulants like methamphetamine or speed as well as non-prescribed substances. A Wave 4 past 12-month tobacco product use variable was derived separately for cigarettes, cigars, pipes, hookahs, snus, smokeless tobacco, dissolvable tobacco, bidis and kreteks, following the same algorithm described in the PATH Study codebook for Wave 4.5 and Wave 5. Then, derived past 12-month total tobacco product use variables, excluding e-cigarettes, were created for each wave. There was no missing data for tobacco use variables. Missingness on other time-varying covariates ranged from 0.04% for use of other substances at Wave 5 to 3.0% for externalizing symptoms at Wave 5. Analytical plan Based on the PATH user’s guide (USDHHS, 2024), we employed Wave 5 special collection all-waves weights for the Wave 4 cohort to adjust for the complex survey design. Descriptive analysis was reported as frequencies, weighted percentages, and their corresponding 95% confidence intervals (CI). Pearson Chi-square tests were conducted to explore respondents' characteristics by online e-cigarette marketing exposure. A t test was used to compare mean ages. Missing data were imputed using multiple imputation by chained equations (MICE) with 20 imputations. We included a large set of predictors in our MICE model, which can help support our assumption that the data were missing at random. Estimates were combined across multiply imputed datasets using Rubin’s Rules (Rubin, 1976). We were interested in estimating to what extent would e-cigarette harm perception and use at wave 5 differ had everyone been exposed to online e-cigarette marketing at waves 4 and 4.5 versus had no one been exposed at waves 4 and 4.5. This quantity can be interpreted as a causal effect assuming sequential exchangeability (i.e., no unobserved confounders), positivity and consistency, which we discuss further in the Supplementary Material. We used a doubly robust, nonparametric targeted minimum loss-based estimator (TMLE) (Díaz et al., 2023) to estimate the risk differences (RD), incorporating PATH survey weights. See Fig. 1 for an illustration of how this approach differs from traditional fixed effects approaches (Rudolph et al., 2020). This estimator is expected to be unbiased if either the treatment models or the outcome models are correctly specified. The TMLE was crossfit with 5-folds. Nuisance parameters were fit using the following ensemble (Van der Laan et al., 2007) of machine learning algorithms: an intercept-only model, a generalized linear model, multivariate adaptive regression splines (Friedman, 1991), random forest (Wright and Ziegler, 2017), and gradient boosting (Chen and Guestrin, 2016). We also report results on the multiplicative scale as a risk ratio (RR). Variance was estimated using the sample variance of the efficient influence function.Fig. 1Directed acyclic graph (DAG) with paths estimated in fixed effects analysis highlighted in blue. The paths estimated by our current approach included both blue and orange paths. A indicates exposure (online marketing); W indicates time-varying covariates in Waves 4 and 4.5. Y indicates e-cigarette use in Wave 5. Directed acyclic graph (DAG) with paths estimated in fixed effects analysis highlighted in blue. The paths estimated by our current approach included both blue and orange paths. A indicates exposure (online marketing); W indicates time-varying covariates in Waves 4 and 4.5. Y indicates e-cigarette use in Wave 5. All analyses were performed using R version 4.3.2. Baseline comparisons were conducted with the survey package (Lumley, 2004), multiple imputation used the mice package (van Buuren and Groothuis-Oudshoorn, 2011), and effect estimation used the lmtp package (Williams and Díaz, 2023). Results Youth characteristics are reported in Table 1. On average, participants were 13.5 years of age at Wave 4, 51.1% male, 52.6% Non-Hispanic White, 13.2% Non-Hispanic Black, 10.2% Non-Hispanic Other Races, and 24.0% Hispanic. Youth who reported online e-cigarette marketing exposure (OEME) at Wave 4 were older, more likely to be female, and with higher parental educational attainment. Supplemental Table 1 shows OEME was more common among those with high internalizing and externalizing symptoms across the 3 waves.Table 1Participant Baseline Stable Covariates by W4 Online E-cigarette Marketing Exposure among Youth aged 12–17 from the Population Assessment of Tobacco and Health Study W4, W4.5, and W5 (N = 8548)a.Online E-cigarette Marketing Exposure at W4All RespondentsExposedNot-Exposedp-valueN%95% CIn%95% CIn%95% CIBaseline Stable CovariatesAge (in years) Mean854813.513.50 – 13.55174713.713.66 – 13.77660213.513.46 – 13.52< 0.001Biological Sex Male444651.1%50.00 – 52.2780419.1%17.85 – 20.41352180.87%79.59 – 77.90< 0.001 Female410248.9%47.73 – 50.0094323.5%22.09 – 24.91308176.49%75.08 – 77.90Parental Educationb GED or less125012.4%11.69 – 13.0921119.00%16.52 – 21.4898381.00%78.52 – 83.48< 0.001 High school138714.9%14.16 – 15.7526219.75%17.45 – 22.05107680.25%77.94 – 82.55 Some college137815.5%14.67 – 16.2934225.64%23.11 – 28.16100974.36%71.84 – 76.88 Associate degree115313.4%12.69 – 14.2525623.32%20.62 – 26.0286676.68%73.98 – 79.37 Bachelor’s degree171122.0%20.09 – 23.0335721.08%19.02 – 23.14133878.92%76.86 – 80.98 Graduate degree160121.7%20.76 – 22.7330519.47%17.40 – 21.55127780.53%78.45 – 82.60Race/Ethnicityc Hispanic260724.0%23.13 – 24.9251120.51%18.83 – 22.19202379.49%77.81 – 81.170.356 NH-Black117013.2%12.41 – 13.9323020.69%18.10 – 23.2789779.31%76.72 – 81.90 NH-White391452.6%51.47 – 53.7383122.06%20.67 – 23.44302277.94%76.56 – 79.31 NH-Others85710.2%9.54 – 10.9817519.78%16.82 – 22.7466080.22%77.26 – 83.17Household Income $24,999 or less202721.6%20.65 – 22.4839920.97%18.99 – 22.94154979.03%77.06 – 80.000.671 $25,000 - $49,999190421.7%20.76 – 22.6438721.64%19.58 – 23.68146978.36%76.31 – 80.42 $50,000 - $99,999202825.9%24.85 – 26.9144622.46%20.49 – 24.42154577.54%75.57 – 79.51 $100,000 or more219530.9%29.79 – 32.0145421.00%19.19 – 22.82171979.00%77.18 – 80.81Notes. Significant findings in bold. Time-varying covariates provided in Supplemental Tables 1 and 2.aWeighted Percentages are reported, CI: Confidence Interval, bold: p < 0.05. P-value for testing group differences.bHS: High school, GED: General equivalency degree.cNH: Non-Hispanic. Table 1Participant Baseline Stable Covariates by W4 Online E-cigarette Marketing Exposure among Youth aged 12–17 from the Population Assessment of Tobacco and Health Study W4, W4.5, and W5 (N = 8548)a.Online E-cigarette Marketing Exposure at W4All RespondentsExposedNot-Exposedp-valueN%95% CIn%95% CIn%95% CIBaseline Stable CovariatesAge (in years) Mean854813.513.50 – 13.55174713.713.66 – 13.77660213.513.46 – 13.52< 0.001Biological Sex Male444651.1%50.00 – 52.2780419.1%17.85 – 20.41352180.87%79.59 – 77.90< 0.001 Female410248.9%47.73 – 50.0094323.5%22.09 – 24.91308176.49%75.08 – 77.90Parental Educationb GED or less125012.4%11.69 – 13.0921119.00%16.52 – 21.4898381.00%78.52 – 83.48< 0.001 High school138714.9%14.16 – 15.7526219.75%17.45 – 22.05107680.25%77.94 – 82.55 Some college137815.5%14.67 – 16.2934225.64%23.11 – 28.16100974.36%71.84 – 76.88 Associate degree115313.4%12.69 – 14.2525623.32%20.62 – 26.0286676.68%73.98 – 79.37 Bachelor’s degree171122.0%20.09 – 23.0335721.08%19.02 – 23.14133878.92%76.86 – 80.98 Graduate degree160121.7%20.76 – 22.7330519.47%17.40 – 21.55127780.53%78.45 – 82.60Race/Ethnicityc Hispanic260724.0%23.13 – 24.9251120.51%18.83 – 22.19202379.49%77.81 – 81.170.356 NH-Black117013.2%12.41 – 13.9323020.69%18.10 – 23.2789779.31%76.72 – 81.90 NH-White391452.6%51.47 – 53.7383122.06%20.67 – 23.44302277.94%76.56 – 79.31 NH-Others85710.2%9.54 – 10.9817519.78%16.82 – 22.7466080.22%77.26 – 83.17Household Income $24,999 or less202721.6%20.65 – 22.4839920.97%18.99 – 22.94154979.03%77.06 – 80.000.671 $25,000 - $49,999190421.7%20.76 – 22.6438721.64%19.58 – 23.68146978.36%76.31 – 80.42 $50,000 - $99,999202825.9%24.85 – 26.9144622.46%20.49 – 24.42154577.54%75.57 – 79.51 $100,000 or more219530.9%29.79 – 32.0145421.00%19.19 – 22.82171979.00%77.18 – 80.81Notes. Significant findings in bold. Time-varying covariates provided in Supplemental Tables 1 and 2.aWeighted Percentages are reported, CI: Confidence Interval, bold: p < 0.05. P-value for testing group differences.bHS: High school, GED: General equivalency degree.cNH: Non-Hispanic. Participant Baseline Stable Covariates by W4 Online E-cigarette Marketing Exposure among Youth aged 12–17 from the Population Assessment of Tobacco and Health Study W4, W4.5, and W5 (N = 8548)a. Notes. Significant findings in bold. Time-varying covariates provided in Supplemental Tables 1 and 2. Weighted Percentages are reported, CI: Confidence Interval, bold: p < 0.05. P-value for testing group differences. HS: High school, GED: General equivalency degree. NH: Non-Hispanic. Few (4.86%) at baseline reported using tobacco products other than e-cigarettes; use increased in Wave 4.5 (7.52%) and Wave 5 (10.69%) with OEME more common among youth who have used tobacco products. Marijuana use was rare in Wave 4 (5.83%), but increased in Wave 5 (16.87%). Similarly, alcohol use increased from about one-fifth (18.52%) in Wave 4 to about three-tenths at Wave 5 (31.30%) and “other” use increased from 5.70% in Wave 4 to 8.62% in Wave 5. Few (5.70%) reported past year use of “other” substances (ranging from cocaine to non-prescribed Adderall) in Wave 4; this increased in Wave 4.5 (8.13%) and Wave 5 (8.62%). OEME was more common among youth with self-reported marijuana, alcohol, and other substance across all three waves. Most participants reported high e-cigarette harm perception at all 3 Waves (58.0%, 55.2%, and 56.8% respectively). Those who reported high e-cigarette harm perception were less likely to report OEME at all 3 waves. Few participants (2.4%) reported current e-cigarette use at Wave 4, but this increased both at Wave 4.5 (6.4%) and at Wave 5 (12.3%). Those who reported current e-cigarette use were more likely to report OEME at all 3 waves. More than half (53.1%) reported using social media more than once a day at baseline; this increased in Waves 4.5 (61.3%) and Wave 5 (66.3%). OEME increased as frequency of social media use among youth increased, from 9.99% among Wave 4 youth with no account to one-fourth of Wave 4 youth (26.43%) who use social media more than once a day. Similar patterns were present at Waves 4.5 and 5. About one-fifth (21.3%) of youth participants reported OEME in Wave 4. As shown in Supplemental Table 2, this was the third most commonly reported form of e-cigarette marketing exposure at Wave 4, much less than retail outlets (51.2%) and somewhat less than television (23.4%). Exposure to e-cigarette marketing via the remaining outlets were as follows: billboards (19.7%), newspapers/magazines (15.2%), radio (8.7%), events/festivals (4.2%), nightclubs/bars (3.5%), and other (1.0%). At all 3 waves, those who were exposed to any form of e-cigarette marketing were significantly more likely to also report OEME, with the exception of “other”. OEME increased slightly at Wave 4.5 (22.8%) and Wave 5 (23.7%), replacing television (18.3% at Wave 4.5) as the second most commonly reported e-cigarette marketing exposure outlet. As shown on Table 2, adjusting for the smaller set of time-invariant and time-varying covariates, the average difference in probability of perceiving e-cigarettes as harmful had everyone been exposed to online e-cigarette marketing versus had no one been exposed was −0.09 (95% CI = −0.12 – −0.05). We also measured this association on the multiplicative scale, estimating a risk ratio (RR) of 0.85 (95% CI = 0.80 – 0.91). The average difference in probability of current e-cigarette use if everyone was exposed to online e-cigarette marketing exposure versus if no one was exposed to e-cigarette marketing exposure was 0.04 (95% CI = 0.02 – 0.06) with a corresponding RR of 1.36 (95% CI = 1.15 – 1.62).Table 2Risk Difference calculations and Risk Ratios comparing Wave 4 and 4.5 Online E-cigarette Marketing Exposure (A=1) versus no Wave 4 and 4.5 Online E-cigarette Marketing Exposure (A=0) (N = 8548).Initial ModelsaExpanded ModelsbWave 4 and 4.5 Online E-cigarette Marketing Exposure (A=1)Wave 4 and 4.5 Online E-cigarette Marketing Exposure (A=1)ψˆSE95% CIψˆSE95% CIWave 5 Harm Perception0.490.020.46–0.520.550.050.46–0.65Wave 5 Current Use0.140.010.12–0.170.110.010.10–0.12No Wave 4 and 4.5 OnlineE-cigarette Marketing Exposure (A=0)No Wave 4 and 4.5 OnlineE-cigarette Marketing Exposure (A=0)Wave 5 Harm Perception0.580.010.56–0.600.570.010.55–0.59Wave 5 Current Use0.110.010.10–0.120.110.020.06–0.15Risk DifferencesRisk DifferencesWave 5 Harm Perception-0.090.02-0.12 to -0.05-0.020.05-0.11–0.08Wave 5 Current Use0.040.010.02–0.060.000.02-0.05–0.04Risk RatiosRisk RatiosWave 5 Harm Perception0.850.030.80–0.910.970.080.82–1.14Wave 5 Current Use1.360.091.15–1.620.970.220.63–1.49Note. Significant findings in bold.aModels include time-varying covariates involving psychological symptoms and other marketing exposure outlets.bModels include time-varying covariates from Model 1, as well as social media use frequency, tobacco (non-e-cigarette) use, and other substance use. Table 2Risk Difference calculations and Risk Ratios comparing Wave 4 and 4.5 Online E-cigarette Marketing Exposure (A=1) versus no Wave 4 and 4.5 Online E-cigarette Marketing Exposure (A=0) (N = 8548).Initial ModelsaExpanded ModelsbWave 4 and 4.5 Online E-cigarette Marketing Exposure (A=1)Wave 4 and 4.5 Online E-cigarette Marketing Exposure (A=1)ψˆSE95% CIψˆSE95% CIWave 5 Harm Perception0.490.020.46–0.520.550.050.46–0.65Wave 5 Current Use0.140.010.12–0.170.110.010.10–0.12No Wave 4 and 4.5 OnlineE-cigarette Marketing Exposure (A=0)No Wave 4 and 4.5 OnlineE-cigarette Marketing Exposure (A=0)Wave 5 Harm Perception0.580.010.56–0.600.570.010.55–0.59Wave 5 Current Use0.110.010.10–0.120.110.020.06–0.15Risk DifferencesRisk DifferencesWave 5 Harm Perception-0.090.02-0.12 to -0.05-0.020.05-0.11–0.08Wave 5 Current Use0.040.010.02–0.060.000.02-0.05–0.04Risk RatiosRisk RatiosWave 5 Harm Perception0.850.030.80–0.910.970.080.82–1.14Wave 5 Current Use1.360.091.15–1.620.970.220.63–1.49Note. Significant findings in bold.aModels include time-varying covariates involving psychological symptoms and other marketing exposure outlets.bModels include time-varying covariates from Model 1, as well as social media use frequency, tobacco (non-e-cigarette) use, and other substance use. Risk Difference calculations and Risk Ratios comparing Wave 4 and 4.5 Online E-cigarette Marketing Exposure (A=1) versus no Wave 4 and 4.5 Online E-cigarette Marketing Exposure (A=0) (N = 8548). Note. Significant findings in bold. Models include time-varying covariates involving psychological symptoms and other marketing exposure outlets. Models include time-varying covariates from Model 1, as well as social media use frequency, tobacco (non-e-cigarette) use, and other substance use. However, as shown on Table 2, when adjusting for the larger set of covariates, the average difference in probability of perceiving e-cigarettes as harmful had everyone been exposed to online e-cigarette marketing versus had no one been exposed was −0.02 (95% CI = −0.11 – 0.08) or RR of 0.97 (95% CI = 0.82 – 1.14), indicating a null relationship between e-cigarette marketing exposure and perceptions of e-cigarettes as harmful. The relationship between e-cigarette marketing exposure and current e-cigarette use was also estimated to be null: 0.00 (95% CI = −0.05 – 0.04) with a corresponding RR of 0.97 (95% CI = 0.63 – 1.49), again indicating uncertainty regarding the direction of the relationship. Discussion Over ten million youth worldwide use e-cigarettes, with exposure to e-cigarette advertising associated with intentions to use in multiple countries (Pettigrew et al., 2023). Given how youth are increasingly exposed to online e-cigarette marketing especially in countries that do not have bans or restrictions on online tobacco advertising (Ylitormanen et al., 2025), it is imperative to examine the impact of online marketing in these countries. Therefore, we estimated causal effects among a U.S. nationally representative longitudinal sample of youth. We found that in our baseline wave from 2016 to early 2018 about one-fifth reported online e-cigarette marketing exposure, which was the third most commonly reported form of e-cigarette marketing exposure. However, by the final wave spanning late 2018–2019, online marketing was second only to retail outlets as the most commonly reported form of e-cigarette marketing and was reported by close to one-quarter of youth. Consistent with our expectations, in our initial analysis, we found significant relationships between 2016-2018 online marketing exposure and reduced e-cigarette harm perception and increased e-cigarette use at the 2018–2019 2-year follow-up in causal inference models adjusting for multiple static and time-varying confounders. However, we surprisingly did not find evidence that online e-cigarette marketing had a significant independent association with e-cigarette harm perception and current e-cigarette use after controlling for frequency of social media use, tobacco (non-e-cigarette) use, and other substance use. We adjusted for multiple potential static and time-varying confounding variables of the relationship between online e-cigarette marketing, harm perception, and use. Most prior longitudinal research assessing the influence of marketing on tobacco use involved fixed-effects approaches (Cavazos-Rehg et al., 2021, Donaldson et al., 2022, Perez et al., 2022, Soneji et al., 2018, Stanton et al., 2022, Vassey et al., 2022), which can lead to biased estimation in the presence of time-varying confounders, i.e., mental health and other forms of tobacco marketing exposure (e.g., retail stores, television, billboards, newspapers/magazines) that may be affected by prior online marketing exposure and affect future instances of online marketing exposures and e-cigarette use (Daniel et al., 2013, Imai and Kim, 2019). In contrast, our analytic approach appropriately adjusts for such time-varying confounders affected by prior exposure. Though other studies have identified a direct relationship between online e-cigarette marketing and e-cigarette use with fixed-effects approaches (Cavazos-Rehg et al., 2021, Donaldson et al., 2022, Perez et al., 2022, Soneji et al., 2018, Stanton et al., 2022, Vassey et al., 2022), our analyses suggest that the relationship between online e-cigarette marketing and e-cigarette perceptions and behaviors may reflect underlying differences in factors such as frequency of social media use or other substance use. These variables may function as predisposing factors that increase both exposure to marketing and the likelihood of e-cigarette use. However, it is also possible that some of these variables lie partially on the causal pathway between marketing exposure and e-cigarette behaviors, and adjusting for them may attenuate the estimated associations. Online youth exposure to social media substance use advertising is a major focus for policy change. Although we lack clarity on the causal pathways, countries that have banned or restricted online tobacco advertising have a lower prevalence of youth e-cigarette use compared to countries without these policies in place (Cho et al., 2022; Ylitormanen et al., 2025). In the U.S., the public is broadly in support of policies regulating online advertising of tobacco (Blake et al., 2022). The U.S. Federal Trade Commission (FTC) has authority to sanction unfair and/or deceptive marketing claims. This authority could reasonably include sanctioning marketers who target youth for whom the use of the product is illegal (FTC, 2022a, Jernigan, 2022). The Food and Drug Administration (FDA) Center for Tobacco Products (CTP) has the authority to assess Civil Money Penalties (CMP) for violations of the Federal Food, Drug, and Cosmetic Act relating to tobacco products. For example, the FTC and FDA have sent warning letters to companies who did not include health warnings on social media promotions (FTC, 2022b). Further, FDA has filed CMP against 146 physical (“brick and mortar”) retailers and 46 online retailers for selling tobacco products that lack the required marketing authorization (FDA, 2025). Further work clarifying online social marketing causal pathways could be helpful. For example, future studies using mediation analyses should examine pathways between e-cigarette marketing exposure, e-cigarette harm perception, and e-cigarette use. There are some notable limitations to this research. Results should generalize to the U.S. youth population, but may not be relevant for other countries or cohorts. It is possible that data missingness is not at random, which would violate assumptions for our imputation procedures. Since analyses only included participants who completed all waves, there is potential for sample bias due to conditioning on a post-exposure variable. However, the majority of participants completed all three waves, so we anticipate relatively low bias potential. Covariates included in the initial and expanded models may have lacked important confounders, such as peer substance use. Other factors unable to be controlled (e.g., measurement error, residual confounding) may have impacted findings. Further, there are various concerns related to the potential for overadjustment when multiple variables are included. For example, variables such as social media and tobacco use may precede and/or partially mediate the exposure and thus could arguably be inappropriate, or misleading, to include as covariates. There are potential issues with the consistency assumption for causal inference as there is some ambiguity in exposure and outcome measurement. Online e-cigarette marketing exposure was assessed by a question about observing such advertising in the past 30 days on “websites or social media sites.” There may be wide variation in this level of exposure. For example, there are a variety of different online platforms (e.g., website, Instagram, Snapchat) and potential for different “doses” (e.g., once a month up to multiple times per day). Notably, even among participants who indicated no social media account or never using social media, some still reported exposure to online e-cigarette marketing, likely due to exposure through websites. Additionally, although the majority of online advertising outlets can arguably be considered websites or social media sites, there are some outlets that may or may not be considered by youth to be included in these two broad categories. These include emails, online streaming services such as YouTube, and online gaming services such as Twitch. Future surveys should consider addressing these potentially disparate categories. In addition, self-reported recall of marketing exposure may be subject to recall bias or other reporting errors. In terms of outcomes, e-cigarette harm perception may vary on a spectrum and may include multiple factors. Further, although current e-cigarette use is often considered as any use in the past 30 days, this is also a measure that includes a wide level of variability and could include incidental or casual use in addition to regular use and/or nicotine dependence. Conclusions There is a long history of research connecting adolescent tobacco marketing with changes in attitudes and behavior. Despite this, continuing concerns regarding the possibility of confounding factors persist. Our research using a U.S. sample suggests the potential for some merit to these concerns as we no longer found significant impacts of online e-cigarette marketing exposure after controlling for frequency of social media use, tobacco (non-e-cigarette) use, and other substance use. The present study expanded on prior analyses to address potential time-varying confounders using state-of-the-art methods to more accurately estimate the long-term implications of online e-cigarette marketing. Our findings suggest that frequency of online use and use of other substances are important to consider when addressing potential time-varying confounding. CRediT authorship contribution statement Ateeqa Ijaz: Writing – original draft, Formal analysis, Data curation. Rachel Ayala Guzman: Writing – review & editing. Kara E. Rudolph: Writing – review & editing, Methodology, Funding acquisition, Conceptualization, Supervision. Nicholas T. Williams: Writing – review & editing, Software, Methodology, Funding acquisition, Formal analysis, Conceptualization. Silvia S. Martins: Writing – review & editing, Writing – original draft, Supervision, Project administration, Funding acquisition, Conceptualization. Aaron Null: Investigation. Nicole Rychagov: Writing – original draft, Formal analysis, Data curation. Paul T. Harrell: Writing – review & editing, Writing – original draft, Supervision, Project administration, Investigation, Funding acquisition, Conceptualization. Funding This project was supported by grant number R21CA287457 from the 10.13039/100000054National Cancer Institute and Food and Drug Administration (10.13039/100000038FDA) Center for Tobacco Products (10.13039/100010628CTP). A portion of PTH’s time was supported by the Virginia Foundation for Healthy Youth (VFHY). The content is solely the responsibility of the authors and does not necessarily represent the official views of the NIH, FDA, or VFHY. Declaration of Competing Interest No conflicts of interest are noted.