Differential Responses to Cigarette Package Labeling Alternatives Among Adults Who Smoke: Results From a Randomized Trial

Abstract

Introduction

Little experimental research has evaluated whether the effects of cigarette package inserts with efficacy messages and/or pictorial health warning labels (PHWLs) differ across key subgroups of adults who smoke.

Aims and Methods

Adults who reported currently smoking (n = 367) were randomly assigned to one of four groups: Small text-only HWLs on pack sides (control); inserts with efficacy messages and small HWLs (inserts-only); PHWLs showing harms of smoking (PHWLs-only); both (inserts + PHWLs). Participants received a 14-day supply of cigarettes labeled to reflect their group. Every evening over 2 weeks, participants reported forgoing and stubbing out cigarettes before they finished smoking over the prior 24 hours, combined into a binary indicator of either behavior (eg, forgoing/stubbing). Separate mixed-effects logistic models were estimated to evaluate moderation of labeling group contrasts (ie, PHWLs vs not; inserts vs. not; inserts-only vs. inserts + PHWLs; PHWLs-only vs. inserts + PHWLs) by baseline covariates (self-efficacy to quit, intention to quit, education, health literacy, and time discounting), predicting day-level forgoing/stubbing.

Results

Education moderated PHWL effects, with PHWLs predicting more forgoing/stubbing only among those with low education (OR = 4.68, p < .001). Time discounting moderated insert effects, with inserts promoting forgoing/stubbing only among those with low time discounting (ie, lower impulsivity; OR = 4.35, p < .001).

Conclusions

Inserts with efficacy messages appear effective mostly among people with low time discounting, whereas PHWLs appear most effective among those with low education, suggesting their potential to address education-related disparities. Labeling strategies appeared equally effective across subgroups defined by self-efficacy to quit, quit intention, and health literacy. Combining inserts with PHWLs did not appear to mitigate moderation effects.

Implications

This randomized trial with adults who smoke suggests that cigarette packs with inserts describing cessation benefits and tips can promote cessation-related behaviors (ie, forgoing or stubbing out cigarettes) among those with low-time discounting (ie, low impulsivity). Alternative interventions may be needed for people with high-time discounting, as found in cessation trials. PHWLs appear most effective among those with low education, potentially addressing education-related disparities. No differential effects were found for those with different levels of self-efficacy to quit, quit intentions, or health literacy. Combining inserts and PHWLs may not be more effective than either alone.

Introduction

Tobacco product labeling is a foundational policy for communicating health information about tobacco use to current and potential future consumers. To maximize labeling effects among adults who smoke and to prevent initiation among youth, the World Health Organization Framework Convention on Tobacco Control recommends that countries adopt large, pictorial health warning labels (HWLs) that illustrate the harmful consequences of tobacco use.¹ Pictorial HWLs for cigarettes improve awareness of tobacco-related risks and promote cessation,^2,3 though the evidence is mixed regarding whether this policy is equally effective for all subgroups of adults who smoke. Canada’s innovative labeling policy not only includes pictorial HWLs but also messages inside packs—known as “inserts”—that target beliefs about cessation benefits (ie, response efficacy) and their confidence to quit (ie, self-efficacy). Studies suggest that these messages can promote efficacy beliefs and, potentially, quit attempts, and cessation.^4–6 This study aims to evaluate whether cigarette packages with pictorial HWLs, inserts with efficacy messages, or both are more or less likely to influence cessation-related behaviors among adults who smoke and have different levels of educational attainment, health literacy, intentions to quit, self-efficacy to quit, and time discounting tendencies.

Differential Effects of Fear-Arousing Pictorial HWLs

Most research on tobacco product labeling has focused on pictorial HWLs for cigarettes, with some evidence suggesting that they can mitigate education- and health literacy-related disparities. In general, pictures are more effective than text for communicating health information to people with relatively low health literacy.⁷ Some experimental evidence indicates that those with low health literacy⁸ and education⁹ were more likely than their counterparts to rate pictorial HWLs as more effective than text-only HWLs. Eye tracking research also indicates that adults who smoke and have lower literacy, but not lower education, viewed pictorial HWLs for a relatively longer period of time, suggesting greater engagement with or, perhaps, longer processing time for evaluating this kind of message.¹⁰ Furthermore, observational studies (eg, in Brazil,¹¹ Canada,¹² and some European countries¹³) that have implemented pictorial HWLs have also found that those with lower educational attainment reported stronger HWL responses. Nevertheless, some observational studies¹⁴ and experimental studies^10,15 showed null associations between education and pictorial HWL responses.

Consistent with expectations from behavior change theories,^16–18 among adults who smoke, those with greater self-efficacy and intentions to quit have generally reported stronger responses to fear arousing pictorial HWLs, whether in observational studies^19,20 or experiments.^9,21,22 To the best of our knowledge, however, no randomized controlled trials have directly addressed moderation of pictorial HWL effects by these key psychosocial characteristics (ie, self-efficacy and quit intentions).

Efficacy Messages and Cigarette Labeling

Behavior change theories, such as the extended parallel processing model¹⁸ and protection motivation theory,¹⁷ posit that fear-appeals will be most effective when message recipients believe the benefits of protective action (ie, response efficacy) and are confident they can engage in this action (ie, self-efficacy). Indeed, fear appeals generally produce desired effects independent of efficacy beliefs.^18,20,23 Furthermore, the inclusion of efficacy information in fear appeals can enhance message effects.²³ HWL messages that most countries have adopted—and that have been used in experimental research—minimally address efficacy beliefs by including information about cessation resources (eg, quitlines and websites), which is associated with increased resource use.^24–26 However, most adults who smoke quit without using these resources,^27,28 and some observational research indicate that the inclusion of this information on new pictorial HWLs in Australia did not change self-efficacy to quit.²⁹ Other research in Australia found that adding more efficacy information to existing pictorial HWLs did not produce the desired effects.³⁰

Canada is the only country in the world that requires inserts with health information in cigarette packaging. Canadian inserts contain eight rotating, elaborated efficacy messages that underscore cessation benefits and provide cessation tips, including tips that do not require using cessation resources.³¹ Post-implementation observational research in Canada found that among adults who smoke, those who read inserts were more likely to have a sustained smoking cessation attempt, with the effects partly mediated by self-efficacy to quit.^5,6 Two 2-week field trials of adults who smoke who were provided packs including Canadian-style inserts found both positive and null effects of insert exposure on self-efficacy to quit and response efficacy,^4,32 though one of these studies found that inserts promoted forgoing of cigarettes,³² an established predictor of cessation attempts. Hence, inserts with elaborated efficacy messages may promote efficacy beliefs and, potentially, cessation behaviors.

Differential Effects of Cigarette Label Efficacy Messages Among Adults Who Smoke

Research is somewhat mixed on differential responses to smoking-related efficacy messages by health literacy and education. Cross-sectional experiments among adults who smoke have found that health literacy was positively associated with the perceived effectiveness of response efficacy messages about cessation benefits,³³ as well as with cessation-related intentions after exposure to this kind of message.³⁴ Discrete choice experiments for Canadian-style inserts found that among adults who smoke, those with lower education were more likely to indicate that none of the efficacy messages they assessed were helpful or motivating to quit.³⁵ Nevertheless, those with lower education found efficacy messages that included cessation resource information to be more helpful and motivating to quit than those with higher education.³⁵ Adding to these mixed results, an observational study of Canadian adults who smoke found education was unassociated with frequency of reading pack inserts with efficacy messages.⁵

Adults who smoke and intend to quit appear generally more responsive to efficacy messages than those who do not intend to quit.^5,35,36 This is likely because, as predicted by theories like the Transtheoretical Model, efficacy information is more relevant to those who are in the “contemplation stage” of changing their behavior than to those who are not. For example, Canadian adults who smoke who intended to quit reported reading inserts with efficacy messages more frequently than those who did not intend to quit.⁵ Experimental studies have found that among adults who smoke, those who intended to quit evaluated efficacy messages as more effective³⁶ and were more likely to perceive at least some of these types of messages to be helpful and motivating to quit.³⁵ These latter studies also found that those with high self-efficacy were more likely to perceive messages as helpful and motivating.³⁵ Nevertheless, self-efficacy to quit was unassociated with the frequency of reading efficacy inserts among Canadian adults who smoke and were exposed to this labeling strategy.⁵ Hence, it is unclear whether efficacy messages work any better for those with greater self-efficacy to quit.

Time discounting is an economic indicator of impulsivity, with higher time discounting predicting a lower likelihood of successful cessation.³⁷ Adults who smoke and have relatively high time discounting tendencies value immediate rewards (eg, stress relief) more than long-term rewards (eg, reduced risk of getting cancer) and have been found to be less likely to quit³⁸ and more likely to relapse.³⁹ Furthermore, greater time discounting was inversely associated with successful abstinence in cessation interventions that involve contingency management and cognitive behavioral therapy approaches.³⁹ To the best of our knowledge, time discounting has not been studied in relation to cigarette labeling effects; however, pictorial HWL messages that focus on long-term health risks may be less effective among people with high time discounting because they place relatively less value on long-term rewards. Similarly, response efficacy messages about the longer-term benefits of cessation may be less effective for these people. Furthermore, cessation interventions that include strategies to enhance self-efficacy to quit have been relatively less effective for people with high time discounting.³⁹ Hence, we expect that those with relatively high time discounting will be less responsive to efficacy messages on inserts.

Study Aims

We conducted a randomized controlled trial among U.S. adults who smoke using ecological momentary assessment (EMA) methods to evaluate the effects of contrasting cigarette labeling policy configurations. Participants were randomly assigned to one of four groups: control (text-only) HWL only; pictorial HWL only; insert only; insert; and pictorial HWLs. The main effects have been reported elsewhere.³² Briefly, participants exposed to inserts with efficacy messages or to pictorial HWLs were more likely to engage in forgoing cigarettes they would normally smoke or stubbing out cigarettes before finishing them, behaviors that predict cessation attempts.⁴⁰ Relative to the control group, labeling effects on psychosocial constructs were medium to large, but mostly not statistically significant.

This paper focuses on the moderation of labeling effects on the behavioral outcome of forgoing or stubbing out cigarettes, for which main effects were found. Pre-registered hypotheses (NCT04075682) for some moderating variables were specific to the contrasts between experimental groups (eg, stronger pictorial HWL effects among those with high self-efficacy; stronger insert effects among those with higher education, higher health literacy, and lower time discounting); however, the review of theory and empirical studies described above support assessment of moderation by all focal variables for these labeling contrasts, even those who were not pre-registered. Theory and empirical evidence suggest that the combination of inserts with efficacy messages and pictorial HWLs should be more effective than either alone; by extension, we hypothesized that moderation effects would be stronger for participants exposed to pictorial HWLs only or inserts only compared to those exposed to both inserts and pictorial HWLs because exposure to both message types would dampen defensive responding or perceived irrelevance that we expect would differ across the subgroups that moderation variables define. Given that this is not a cessation trial and most participants were not motivated to quit, the 2-week period of our study is too short to assess cessation attempts. We focus on forgoing or stubbing out cigarettes because these “microindicators of concern” increase when new HWLs are implemented,⁴¹ are promoted by pictorial HWLs in randomized trials,⁴² and consistently predict subsequent cessation attempts in observational studies with a range of follow-up periods.⁴³

Material and Methods

Study Design

Our randomized controlled trial involved a 2X2 between-subject design (ie, inserts with efficacy messages vs no inserts; large pictorial HWLs vs small text-only HWLs) wherein adults who smoke received a 14-day supply of their preferred cigarette brand variety with packs modified using labels that reflected their experimental condition. We used EMA over 15 days during which participants answered brief surveys approximately 4–5 times a day and on cessation-related behaviors each evening (see Evening Survey, below) on a study-provided smartphone.

Stimuli

Two insert messages about cessation benefits and two containing cessation tips (Figure 1) were developed based on prior research,^35,36 with phrasing for primary school reading levels (range = 4.6–5th grade). Inserts were printed on 5 × 8.5 cm glossy cards and placed inside packs between the external packaging and the foil that covers cigarettes. HWL text for all conditions included four messages specified in 2012 for future U.S. implementation (Figure 1). The text-only “control” condition used the current U.S. HWL size and placement (ie, 50% of one pack side). Pictorial HWLs were 5 × 4.5 cm stickers (approximately 50% of pack front/back), included imagery based on prior research,⁴⁴ and were placed on the lower half of both the front and back of packs. Insert and HWL messages were rotated systematically in/on packs so that most participants would be exposed to each message multiple times, with the frequency depending on the number of packs they received. Packs were sealed in zip-lock bags to maintain cigarette freshness because cellophane was removed from packs.

Figure 1.

Labeling stimuli.

Sample

Eligibility criteria included age (ie, ≥18 years old before 2021, ≥21 after that due to federal increase in age of legal purchase); smoking at least 10 cigarettes per day in the prior month; smoking at least 100 cigarettes in their lifetime; and exhaled CO (Covita) of at least 8ppm to confirm current smoking, although this criterion was dropped after the COVID onset as it would require mask removal and, therefore, violate safety recommendations. People who used other nicotine products in the prior month were ineligible because of challenges around assessing compensatory behaviors (ie, increasing other nicotine product use while reducing smoking). Quotas for education (50% ≤ high school; 50% > high school) and sex (50% male; 50% female) were planned, though quotas were relaxed due to challenges with recruitment after COVID-19 onset. Participants were recruited in New York, South Carolina, and North Carolina, with recruitment involving both in-person (eg, intercepts at smoke shops) and online (eg, Facebook ads, Craig’s list) strategies, for which details are published elsewhere.⁴⁵ Sample size was determined based on relatively small effects in prior research on pictorial HWLs, with gains in statistical power from repeated measures based on a projected intra-class correlation (ICC) of 0.4. However, ICCs were higher than expected, reducing power; nevertheless, we found medium-to-large statistically significant main effects for labeling on forgoing/stubbing.

Measurements

Baseline Survey Moderators

At baseline, participants were queried on smoking-related perceptions and behavior, including quit intentions (dichotomized to intending to quit in the next 6 months vs. not)⁴⁶ and three self-efficacy to quit items⁴⁷ with high reliability (α = 0.81) that were averaged and assessed both with this average and with dichotomization at the median. Health literacy was assessed using the newest vital sign⁴⁸ (range = 0–6), which was dichotomized into adequate literacy (> 3) versus not, and educational attainment was dichotomized (ie, high school or less vs. higher). Assessment of time discounting involved a 1-minute validated task with branching logic over five questions about preferences for near- versus longer-term delayed rewards (range = 1 hour-25 years) that resulted in 32 possible outcome levels that were log-transformed to remediate skew (range: −9.12–3.18)⁴⁹; however, skew remained and so analyses involved dichotomizing this variable at the median.

Evening Surveys

Each day, the study-provided smartphone prompted participants at 7 PM to complete the 1–3 minute survey. A reminder was sent by text at 9 PM if the survey was not yet completed. The survey asked whether, in the prior 24 hours, they had stubbed out a cigarette before finishing it (yes/no) and whether they had foregone any cigarettes they would normally have smoked (yes/no), as these behaviors predict subsequent cessation attempts.⁴⁰ Responses to these two questions were combined into a single dichotomous variable to indicate either behavior (yes/no) during each day.

Analysis

To test our hypotheses, we used STATA v16.1 to estimate mixed-effects logistic models that accounted for individual-level repeated measures. Our outcome of forgoing or stubbing out cigarettes was drawn from evening reports collected on days 2–14 of the study, as smoking behaviors on days 1 (baseline) and 15 (final survey) were only partially collected, with varying coverage across participants depending on the time when they met with study staff. Details on the statistical design are provided in Appendix 1. The first set of models assesses moderation of insert effects (ie, binary insert only and insert + pictorial HWL vs. pictorial HWL only & control = reference; model 1, H2a in Appendix 1) and, separately, pictorial HWL effects (ie, binary pictorial HWL only and insert + pictorial HWL vs. insert only and control = reference; model 1, H2b in Appendix 1). In separate models, each baseline moderation variable was evaluated as both a main effect (model 1, H1 in Appendix; results not reported) and interaction with treatment, using binary variables with the higher risk group as the reference (ie, low health literacy = reference; low education = reference; low self-efficacy = reference; no intention to quit = reference; high time discounting = reference). When the interaction term was statistically significant, we stratified the data by the binary moderating variable and re-estimated the model. The next set of models evaluated whether exposure to packs with both inserts and pictorial HWLs dampened moderation effects. These models estimated the difference in moderation effects for the labeling-alone approach (pictorial HWLs alone or insert alone, evaluated separately) and the combined approach (pictorial HWLs + inserts; see Appendix 1, model 1, H3a and H3b, respectively). Wald tests for the associated regression parameters were then used to evaluate moderation effects. Most participants completed the majority of evening reports (ie, 78% completed ≥ 11 of the 13), and we assessed the impact of missing data (13.9% or 663/4771 possible observations) by comparing results from the full case analyses presented here and those using multiple imputations with chained equations. We also re-estimated models after and evaluated the consistency of results when the moderating variables of health literacy (range = 0–6), self-efficacy (range = 1–5), and time discounting (range = −9.12–3.18) were assessed as continuous.

Results

A full description of the study sample has been reported elsewhere.³² Briefly, 443 people were eligible and enrolled in the study. We excluded those who did not attend study orientation (n = 56), had too much difficulty self-administering the baseline survey (n = 1), experienced health concerns (n = 2), had already participated in the study (n = 1), received incorrectly labeled cigarettes (n = 1), or did not complete any cigarette or evening surveys (n = 15). Of the final analytic sample (n = 367 participants, 4196 evening surveys; see Table 1), most were female (61%), White (81%), had greater than high school education (58%), had “adequate” health literacy (ie, score > 3 on newest vital sign = 71%), and, at baseline, neither intended to quit smoking within six months (67%) nor had tried to quit in the prior 12 months (70%). The modal age was 36–55 years old (52%), with younger (18–35 years old = 28%) than older (> 55 years old = 20%) participants, and modal smoking frequency was 16 to 20 cigarettes per day (41%), with roughly equal proportions of the participant whose smoking was more (> 20 CPD = 30%) and less (10–15 CPD = 29%) frequent. No statistically significant differences in baseline characteristics were found across experimental conditions.

Table 1.

Participant Characteristics by Experimental Condition

Participant characteristics		Control n = 101	Insert n = 87	Pictorial HWL n = 90	Insert + pictorial HWL n = 89	Total n = 367
Age	18–35	27(27%)	27(31%)	26(30%)	21(24%)	101(28%)
	36–55	54(53%)	41(48%)	44(50%)	49(55%)	188(52%)
	≥56	20(20%)	18(21%)	18(20%)	19(21%)	75(20%)
Sex	Male	41(41%)	35(41%)	37(42%)	28(31%)	141(39%)
	Female	60(59%)	50(59%)	51(58%)	61(69%)	222(61%)
Race	White	77(76%)	74(85%)	73(81%)	72(81%)	296(81%)
	Non-White	24(24%)	13(15%)	17(19%)	17(19%)	71(19%)
Education	≤High school	48(48%)	36(42%)	30(34%)	39(44%)	153(42%)
	>High school	53(53%)	49(58%)	57(66%)	50(56%)	209(58%)
Income	<$10 000	17(17%)	11(13%)	14(16%)	14(16%)	56(15%)
	$10 000–29 999	43(34%)	18(21%)	24(27%)	24(27%)	100(28%)
	$30 000–44 999	19(19%)	12(14%)	16(18%)	22(25%)	69(19%)
	$45 000–59 999	14(14%)	19(22%)	8(9%)	13(15%)	54(15%)
	$60 000–74 999	6(6%)	11(13%)	9(10%)	6(7%)	32(9%)
	>$75 000	9(8%)	11(13%)	13(15%)	9(9%)	42(11%)
	No answer	2(2%)	3(4%)	4(5%)	1(1%)	10(3%)
Health literacy	Limited	9(9%)	7(8%)	5(6%)	7(8%)	28(8%)
	Possibly limited	26(26%)	19(22%)	19(21%)	19(21%)	83(22%)
	Adequate	66(65%)	61(70%)	66(73%)	63(71%)	256(70%)
Cigarettes per day	10–15	30(30%)	22(26%)	27(31%)	26(29%)	105(29%)
	16–20	44(43%)	38(44%)	30(34%)	38(43%)	150(41%)
	>20	27(27%)	26(30%)	31(35%)	25(28%)	109(30%)
Intend to quit (next 6 months)	Yes	32(32%)	29(34%)	31(35%)	27(30%)	119(33%)
	No	69(68%)	57(66%)	57(65%)	62(70%)	245(67%)
Quit attempt (last 12 months)	Yes	30(30%)	26(30%)	29(33%)	22(25%)	107(29%)
	No	71(70%)	60(70%)	58(66%)	65(73%)	254(70%)
	Don’t know	0(0%)	0(0%)	1(1%)	2(2%)	3(1%)
Self-efficacy to quit–mean (SD)1		2.32(1.06)	2.30(0.92)	2.39(1.05)	2.21(0.81)	2.31(0.97)
Time-discounting–mean (SD)2		−4.19(2.18)	−4.81(2.12)	−4.63(2.26)	−4.68(2.32)	−4.56 (2.23)
Evening surveys submitted		1164	979	1031	1022	4196

The ns shown are at the level of the individual participant; HWL = health warning label.

1. response scale = 1–5; 2. Log-transformed range = −9.12–3.18.

Pictorial HWL Effects

Models assessing pictorial HWL effects (Table 2) indicated statistically significant moderation effects for educational attainment: When exposed to pictorial HWLs (vs. not) those with high versus low education were relatively less likely to forgo/stub out cigarettes (OR = 0.19, p = .004). In stratified models, exposure to pictorial HWLs was unassociated with forgoing/stubbing out cigarettes among those with high education (OR = 0.91, p = .89) whereas pictorial HWL exposure was significantly associated with forgoing/stubbing out cigarettes among those with low education (OR = 4.68, p < .001). Figure 2A shows the prevalence of these outcomes for each education group by labeling condition.

Table 2.

Moderation of Cigarette Labeling Effects on Forgoing or Stubbing Out Cigarettes by Sociodemographic and Psychosocial Characteristics¹

Moderating variables				Labeling group contrasts
	Pictorial HWL vs. not2			Insert only vs. insert + Pictorial HWL			Insert vs. not3			Pictorial only vs. insert + Pictorial HWL
	OR	(95% CI)	p-val	OR	(95% CI)	p-val	OR	(95% CI)	p-val	OR	(95% CI)	p-val
Education	0.20	(0.06,0.60)	.004	5.45	(1.18,26.53)	.036	2.18	(0.72,6.61)	.167	0.49	(0.10,2.42)	.380
Literacy	1.13	(0.33,3.83)	.849	0.59	(0.10,3.47)	.563	0.46	(0.14,1.57)	.216	1.45	(0.24,8.71)	.682
Quit intention	0.66	(0.21,2.08)	.479	0.78	(0.15,4.11)	.770	1.89	(0.60,5.92)	.276	0.27	(0.05,1.42)	.123
Self-efficacy	1.19	(0.39,3.58)	.758	0.79	(0.16,3.89)	.771	1.19	(0.39,3.58)	.759	0.79	(0.16,3.80)	.768
Time discounting	0.92	(0.33,2.94)	.969	0.58	(0.11,2.68)	.459	3.66	(1.15,10.37)	.027	0.15	(0.03,0.75)	.021

¹ Results in the table are mixed effect model coefficients for the interaction term between the labeling group contrast in the column heading and the binary moderating variable, with each interaction coefficient assessed in a different model.

² Models involve a group contrast between pictorial HWL only or pictorial HWL + insert groups versus insert only or control groups.

³ Models involve a group contrast between insert-only or pictorial health warning labels + insert groups versus pictorial HWL-only or control groups.

Figure 2.

Prevalence of forgoing or stubbing out cigarettes across labeling conditions: Results for statistically significant moderating variables. (A) Pictorial HWL effects by levels of education, (B) Insert only versus Insert + Pictorial HWL by levels of education, (C) Insert effects by level of time discounting, (D) Pictorial HWL only versus Pictorial HWL + insert by levels of time discounting. See Figures S1 and S2 for the prevalence of forgoing or stubbing out cigarettes across labeling conditions, by level of education and time discounting.

Models assessing whether moderation effects differed for participants exposed to inserts only compared to inserts + pictorial HWLs (ie, evaluating whether adding pictorial HWLs to inserts dampened moderation effects) again indicated statistically significant moderation only for education (Table 2). Results from stratified models were not statistically significant, but among those with low education, inserts + pictorial HWLs were more likely to promote forgoing/stubbing out relative to inserts only (OR = 3.09, 95%CI = 0.73, 13.10, p = .126), whereas the opposite pattern was found for those with high education (OR = 0.64, 95%CI = 0.23, 1.80, p = .401). Figure 2B shows the prevalence of forgoing/stubbing across these labeling conditions for high and low-education groups.

Insert Effects

Moderation of insert effects on forgoing/stubbing out were statistically significant only for time discounting (OR = 3.66, 95% CI = 1.15 to 10.37, p = .027; Table 2). Stratified models including those with low time discounting indicated that insert exposure was associated with a greater likelihood of forgoing/stubbing out (OR = 4.35, p < .001), whereas this association was not significant among those with high time discounting (OR = 1.26, p = .561). The higher relative prevalence of forgoing/stubbing out by insert exposure for each time discounting group is shown in Figure 2C.

In models evaluating the contrast between those whose packs included only pictorial HWLs and those whose packs included inserts + pictorial HWLs (ie, evaluating whether adding inserts to pictorial HWLs dampened moderation effects; Table 2), moderation effects were only statistically significant for time discounting (OR = 0.15, p = .021). Models stratified by time discounting indicated that among participants with low time discounting, those exposed to pictorial HWL-only labels were less likely to forgo/stub out cigarettes relative to those exposed to inserts + pictorial HWLs (OR = 0.37, 95% CI = 0.16 to 0.89, p = .026). Among those with high time discounting, the prevalence of forgoing/stubbing in both labeling conditions was similar (OR = 0.96, 95% CI = 0.38 to 2.44, p = .930). Figure 2D shows the prevalence of forgoing/stubbing for each time discounting group by labeling condition.

Sensitivity Analyses

Results from sensitivity analyses involving multiple imputations of missing data were consistent with those reported here in terms of the direction of association and statistical significance (Table S1). Models that included moderating variables as continuous were also consistent with one exception: the results for moderation of insert effects by time discounting were no longer statistically significant (Table S2).

Discussion

This study found some evidence of moderation of cigarette package labeling effects on the cessation-related outcome of forgoing or stubbing out cigarettes, for which our prior research with data from this study found evidence of the positive effects of both inserts and pictorial HWLs.³² In particular, we found evidence that fear arousing pictorial HWLs (relative to small, text-only warnings) appears especially effective amongst those with low educational attainment. Furthermore, inserts with efficacy messaging appear to be more effective for those with lower time discounting. The lack of evidence for moderation by self-efficacy to quit, health literacy, and quit intentions suggest that the labeling strategies we evaluated may be equally effective across key subgroups of adults who smoke—at least when considering short-term cessation-related behaviors.

Our finding that pictorial HWLs were more effective in promoting forgoing/stubbing out of cigarettes among those with lower education is consistent with some other research.⁹ Indeed, only amongst those with lower education did pictorial HWLs promote foregoing/stubbing out, whether when evaluated as a simple contrast (pictorial HWLs vs. not) or when comparing exposure to inserts only with inserts and pictorial HWLs. Further research with longer follow-up is needed to determine whether these behaviors translate into quit attempts and cessation. Recent quasi-experimental evaluation research comparing the effects of United States and Canadian labeling policies suggests that Canada’s policy—which includes both pictorial HWLs and inserts with efficacy messages—does not exacerbate and may even promote health equity.⁵⁰ The lack of evidence for additive or multiplicative effects of both pictorial HWLs and inserts over either strategy by itself is inconsistent with theory and empirical evidence. We speculate that including both message types for people who have been exposed to the same small text-only HWLs for almost 4 decades may have been overwhelming and generated greater defensive responses, rather than dampening undesired outcomes. Research with longer periods of follow-up are needed, as well as research on including inserts in the many countries where smokers are already exposed to pictorial HWLs.

We also found evidence that time discounting moderated the effects of inserts but not pictorial HWLs, whether evaluating the simple contrast between receiving packs with inserts versus not or when evaluating pictorial HWLs alone versus pictorial HWLs and inserts. The messages used on inserts to promote response efficacy focused on financial savings from quitting and cardiovascular benefits, mentioning both relatively short- and longer-term benefits. Our results suggest that such messages are particularly effective for people who value long-term outcomes but not for people who more strongly value the near term, perhaps because the reinforcing behavior of interest is addictive. Self-efficacy enhancing messages with cessation tips around delaying cigarettes and substituting other activities when experiencing cravings may also be ineffective for those with high time discounting. Indeed, such strategies are often embedded in contingency management interventions that are less effective for those with high-time discounting.³⁹ As such, fear-arousing imagery on pictorial HWLs that works through affective pathways³² may be more effective for this higher-risk group of people, even though pictorial HWLs focus on longer-term effects of smoking. Those with high time discounting tendencies likely require more intensive cessation assistance than static labeling messages can provide, although labeling can promote awareness of and linkage to cessation resources.²⁵ The development of scalable cessation strategies for this population should be the subject of future research.

We found no statistically significant differences in the effects of labeling manipulations on forgoing/stubbing out by quit intention, self-efficacy to quit, or health literacy. This is contrary to our expectations that the labeling messages would be more relevant and, therefore, more effective for those who intend to quit or have high self-efficacy to quit. From a population health perspective, equivalent effectiveness is desirable, as people with low levels of quit intentions, self-efficacy, or health literacy are generally less successful in efforts to quit smoking. In addition, studies finding differential labeling effects for health literacy,^10,33,34 quit intention,^35,36 and self-efficacy³⁵ are cross-sectional and/or involve single exposures; however, longitudinal observational studies have found no evidence of differential pictorial HWL effects by self-efficacy to quit.²⁰ Repeated exposure to labeling messages, as in our 2-week RCT, may give people more time to process messages, thereby minimizing some differences that are apparent when people are exposed to messages just once. Future research should explicitly evaluate the predictive validity of single-exposure studies for assessing moderation effects.

Although the null effects we found for moderation may be valid, other explanations may also account for these findings. The high ICC for our outcome (ICC = 0.70) reduced the potential for gains in statistical power due to the repeated measures from our EMA approach (³²). Furthermore, our sample was relatively addicted to smoking, as our inclusion criteria involved smoking at least 10 cigarettes per day: this level of addiction may have resulted in weaker labeling effects than would be found among people who smoke less often. Hence, our conclusions regarding limited differential effects should be interpreted with some caution.

Our study has a number of other limitations. While participants did not complete all study protocols (eg, 78% completed at least 11 of the 13 evening reports), results from our sensitivity analyses using multiple imputations to address missing data were consistent with those presented here, suggesting that missing data did not seriously impact our findings. Furthermore, our data quality is consistent with other EMA research.⁵¹ The results we found were also consistent when evaluating moderation variables as continuous. The exception was the lack of statistically significant interactions with time discounting; however, this discrepancy may be due to the skewed nature of the continuous variable, even after transformation to reduce skew. Finally, results from our study may not generalize to the broader population of adults who smoke in the United States. Some sample characteristics that are different from the general population may have dampened labeling effects (eg, smoking frequency and interest in quitting), although other sample characteristics may augment labeling effects (eg, more females, who tend to be more responsive to health messages) or have unknown effects on our estimates (eg, age). We excluded people who used other nicotine products; however, this exclusion likely does not seriously compromise generalizability since exclusive smoking is the dominant pattern of nicotine product use among U.S. adults.⁵²

Despite these limitations, our study provides evidence that pictorial HWLs may be particularly effective for adults who smoke with lower education, among whom smoking is increasingly concentrated. Furthermore, inserts containing efficacy messages appear to work well for people with low time discounting who value longer-term outcomes and who appear relatively more responsive to other cessation interventions, as well.

Supplementary Material

Supplementary material is available at Nicotine and Tobacco Research online.

Funding

Research reported in this publication was supported by the National Cancer Institutes of the National Institutes of Health under award number R01CA215466. 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

JFT and DH have served as expert witnesses on behalf of governments in litigation involving the tobacco industry. The remaining authors declare that they do not have any conflicts of interest.

Author Contributions

James Thrasher (Conceptualization [equal], Methodology [equal], Resources [equal], Writing—original draft [lead], Writing—review & editing [lead]), Liyan Xiang (Data curation [equal], Formal analysis [lead], Writing—review & editing [supporting]), Jason Yang (Conceptualization [supporting], Writing—review & editing [equal]), Yanwen Sun (Writing—original draft [supporting], Writing—review & editing [supporting]), Emily Hackworth (Project administration [equal], Writing—original draft [lead], Writing—review & editing [equal]), Stuart Ferguson (Conceptualization [equal], Funding acquisition [equal], Methodology [equal], Writing—original draft [equal], Writing—review & editing [equal]), Minji Kim (Conceptualization [equal], Writing—review & editing [equal]), David Hammond (Conceptualization [supporting], Funding acquisition [supporting], Writing—review & editing [equal]), James Hardin (Conceptualization [supporting], Data curation [supporting], Formal analysis [supporting], Funding acquisition [supporting], Methodology [supporting], Supervision [lead], Writing—original draft [supporting], Writing—review & editing [supporting]), and Jeff Niederdeppe (Conceptualization [supporting], Funding acquisition [supporting], Investigation [supporting], Methodology [supporting], Writing—original draft [equal], Writing—review & editing [equal])

Data Availability

The data underlying this article will be shared on reasonable request to the corresponding author.