Negative affect, message reactance, and perceived risk: How do pictorial cigarette pack warnings change quit intentions?

Abstract

Objective

Pictorial warnings on cigarette packs increase motivation to quit smoking. We sought to examine the potential mediating role of negative affect, message reactance (i.e., an oppositional reaction to a message), and perceived risk in shaping quit intentions.

Methods

In 2014 and 2015, we randomly assigned 2,149 adult US smokers to receive either pictorial warnings or text-only warnings applied to their cigarette packs for four weeks. Analyses used structural equation modeling with bootstrapped standard errors to test our theorized mediational model.

Findings

Pictorial warnings increased negative affect, message reactance, and quit intentions (all p<.001), but not perceived risk (i.e., perceived likelihood and severity of harms of smoking). Negative affect mediated the impact of pictorial warnings on quit intentions (mediated effect=.16, p<.001). Message reactance weakened the impact of pictorial warnings on quit intentions, although the effect was small (mediated effect=−.04, p<.001). Although pictorial warnings did not directly influence perceived risk, the model showed additional small mediation effects on quit intentions through negative affect and its positive association with perceived risk (mediated effect=.02, p<.001), as well as reactance and its negative association with perceived risk (mediated effect=−.01, p<.001).

Conclusions

Pictorial cigarette pack warnings increased quit intentions by increasing negative affect. Message reactance partially attenuated this increase in intentions. The opposing associations of negative affect and reactance on perceived risk may explain why pictorial warnings did not lead to observable changes in perceived risk.

Introduction

Tobacco use is the leading cause of preventable morbidity and mortality worldwide, causing nearly six million deaths each year.[1] The World Health Organization Framework Convention on Tobacco Control calls for its signatory countries to implement large pictorial images on cigarette packs based on evidence that pictorial warnings are more effective than text-only warnings.[2] Indeed, systematic reviews and a recent randomized controlled trial demonstrate that pictorial warnings elicit stronger quit intentions[3, 4] and increase subsequent cessation behavior, compared to text-only warnings.[3, 5] Characterizing how pictorial warnings strengthen quit intentions can help policymakers design more effective warnings.

The 2009 Family Smoking Prevention and Tobacco Control Act requires pictorial warnings on cigarette packs in the US.[6] However, a 2012 lawsuit brought forth by the tobacco industry stalled implementation of this facet of the law.[7] In the court case, the warnings were criticized for being “unabashed attempts to evoke emotion.”[7] Indeed, research has shown that pictorial warnings elicit fear and other negative emotions.[4, 8–12] These negative emotions may serve an important information processing role by making the warning content accessible and salient, thereby more effectively influencing outcomes such as risk perceptions and quit intentions.[8, 9] Emotional responses are typically formed with greater confidence and more quickly than cognitive appraisals, and are therefore likely to precede cognitive responses such as perceived risk.[13] Affect may also help people process risk information.[14–18] Understanding the role of negative affective responses to pictorial warnings may be relevant to future legal battles in the US and may also inform the creation of strong warnings in countries that are implementing new pictorial warning policies or changing their existing pictorial warnings.

In addition to negative affect, pictorial warnings may elicit message reactance, defined as cognitive and emotional resistance to a health message in response to a perceived threat to one’s freedom.[19–21] Several studies have found that pictorial warnings elicit greater reactance than text-only warnings.[22–25] However, the extent to which reactance weakens the impact of pictorial warnings on outcomes such as quit intentions remains an important research gap. Moreover, research is needed to understand whether negative emotions and message reactance are associated with changes in perceived risk.

In the current study, we sought to determine the psychological mechanisms by which pictorial cigarette pack warnings elicit stronger quit intentions using experimental data. Specifically, we aimed to understand whether pictorial warnings heightened negative affect and reactance, and whether, in turn, negative affect and reactance were associated with quit intentions via perceived risk (i.e., perceived likelihood of harm from smoking and perceived severity of harm from smoking).

Methods

Participants

From September 2014 to August 2015, we recruited[26] a convenience sample of adult smokers in North Carolina and California, US to participate in a trial comparing the impact of pictorial and text versus text-only warnings.[3] Participants were age 18 or older, proficient in English, and current smokers, defined as having smoked at least 100 cigarettes during their lifetime and now smoking every day or some days. Exclusion criteria included pregnancy, current enrollment in a smoking cessation trial, smoking only roll-your-own cigarettes, smoking fewer than seven cigarettes per week, and living in the same household as another trial participant. Details regarding recruitment, design, and methods have previously been reported.[3]

Procedures

In our trial, smokers received warnings on their own cigarette packs for four weeks using a protocol developed by our team.[3, 27] Participants brought in an eight-day supply of cigarettes weekly. At baseline, we randomly assigned participants to receive one of eight warnings: four pictorial warnings with text, and four text-only controls in a two-arm trial testing pictorial vs. text-only warnings. The four pictorial warnings contained text required by the Tobacco Control Act and a picture to illustrate a health harm of smoking selected from the FDA’s originally-proposed set of images (Figure 1).[28] We chose these four warning images because they performed well in a previous internet study and avoided many of the criticisms in the lawsuits (e.g., using a cartoon or a rare health harm of smoking).[29, 30] We removed the FDA quitline number from the images, which was a source of contention in litigation against the warnings.[7, 31] Staff applied the pictorial warning labels to the top half of the front and back panels of participants’ cigarette packs, in accordance with the FDA requirements.[6] The four text-only control warnings used the US Surgeon General’s warning statements that have been required on the side of cigarette packs since 1985. Staff applied the text-only warning labels on the side of the packs covering the existing US Surgeon General’s warnings. We applied the new warning labels on top of the existing warnings to control for the effect of putting a label on smokers’ packs. Randomization created groups that did not differ on demographics assessed (all p>.05).[3] For this reason, we did not adjust analyses for demographics. Trial participants were diverse, including a substantial number of sexual minority, African American, low-education, and low-income smokers (Table 1).

Figure 1.

Intervention warnings (left) and control warnings (right)

Table 1.

Participant characteristics (n=2,149)

	Text-only warnings (n=1,078)		Pictorial warnings (n=1,071)
	n	(%)	n	(%)
Demographics
Age
18–24 years	171	(16.1)	152	(14.5)
25–39 years	377	(35.5)	398	(37.9)
40–54 years	338	(31.8)	304	(29.0)
55+ years	176	(16.6)	195	(18.6)
Mean (SD) years	39.7	(13.4)	39.8	(13.7)
Gender
Female	548	(51.2)	512	(48.2)
Male	507	(47.4)	532	(50.0)
Transgender	15	(1.4)	19	(1.8)
Gay, lesbian, or bisexual	173	(16.3)	195	(18.8)
Hispanic	92	(8.6)	89	(8.5)
Race
Black or African American	484	(45.8)	510	(48.9)
White	393	(37.2)	358	(34.3)
Other/multiracial	134	(12.7)	117	(11.2)
Asian	28	(2.7)	42	(4.0)
American Indian or Alaska Native	7	(0.6)	11	(1.0)
Native Hawaiian or other Pacific Islander	11	(1.0)	6	(0.6)
Education
High school graduate or less	333	(31.1)	344	(32.5)
Some college	519	(48.5)	502	(47.4)
College graduate	156	(14.6)	156	(14.7)
Graduate degree	63	(5.9)	58	(5.5)
Household income, annual
$0–$24,999	566	(53.3)	589	(55.8)
$25,000–$49,999	272	(25.6)	266	(25.2)
$50,000–$74,999	110	(10.3)	92	(8.7)
$75,000+	115	(10.8)	109	(10.3)
Low income (≤ 150% of Federal Poverty Level)
No	506	(47.0)	477	(44.8)
Yes	570	(53.0)	589	(55.2)
Trial site
California	594	(55.1)	592	(55.3)
North Carolina	484	(44.9)	479	(44.7)
Cigarettes smoked per day, mean (SD)	8.8	(6.6)	8.7	(7.3)
Smoking frequency
Non-daily	211	(19.6)	207	(19.3)
Daily	866	(80.4)	864	(80.7)
Mediators and outcome at baseline, mean (SD)
Perceived likelihood of harm from smoking	3.3	(0.9)	3.3	(0.9)
Perceived severity of harm from smoking	3.7	(0.6)	3.7	(0.6)
Quit intentions	2.2	(0.9)	2.3	(0.9)

Note. Study characteristics, mediators and outcomes at baseline did not differ by trial arm.[3] Missing demographic data ranged from 0.7% to 2.2%. The baseline surveys could not assess negative affect or message reactance as participants had not yet seen the warnings.

Participants completed two computer surveys at the first visit (i.e., baseline and immediately after seeing their assigned warning, which was immediate post-test), and one survey at each visit thereafter for four weeks. Participants received a cash incentive at the end of each visit, totaling up to $185 in North Carolina and $15 higher in California due to the higher cost of living. At the end of the final follow-up appointment, participants received information about local smoking cessation programs. The University of North Carolina Institutional Review Board approved the procedures for this trial.

Measures

The baseline survey and the week 2 follow-up survey assessed perceived likelihood of harm from smoking and perceived severity of harm from smoking. We measured negative affect elicited by the warning at immediate post-test.[32] We originally planned to examine fear alone as a mediator given its importance in the Extended Parallel Process Model,[33] but sensitivity analyses revealed that fear and the other negative affect items exhibited a nearly identical pattern of associations, and confirmatory factor analysis supported treating negative affect as a single latent factor. The immediate post-test survey also assessed message reactance using the Brief Reactance to Health Warnings Scale.[25] This measure captures the three main components of reactance: anger toward the warning, perceived threat to freedom, and counterarguing against the warning.[19, 20, 24] Finally, we measured quit intentions at baseline, immediate post-test, and all follow-up surveys.[34] Item wording, response scales, and factor loadings for all measures appear in Table 2.

Table 2.

Latent variables used in the measurement and structural equation models (n=2,149)

Latent variable (Timepoint used in analysis)	Indicator item wording [response scale]	Factor loading
Negative affect (Immediate post-test)	How much did the warning on your cigarette packs make you feel… Scared? [not at all (coded as 1), a little (2), somewhat (3), very (4), extremely (5)]	.93
	Regretful? [not at all (1), a little (2), somewhat (3), very (4), extremely (5)]	.91
	On edge? [not at all (1), a little (2), somewhat (3), very (4), extremely (5)]	.88
	Disgusted? [not at all (1), a little (2), somewhat (3), very (4), extremely (5)]	.88
	Sad? [not at all (1), a little (2), somewhat (3), very (4), extremely (5)]	.86
Message reactance (Immediate post-test)	Please say how much you agree or disagree with each statement below about the warning we put on your packs. The health effect on this warning is overblown. [strongly disagree (1), somewhat disagree (2), neither agree nor disagree (3), somewhat agree (4), strongly agree (5)]	.82
	This warning is trying to manipulate me. [strongly disagree (1), somewhat disagree (2), neither agree nor disagree (3), somewhat agree (4), strongly agree (5)]	.77
	This warning annoys me. [strongly disagree (1), somewhat disagree (2), neither agree nor disagree (3), somewhat agree (4), strongly agree (5)]	.75
Perceived risk1 (Week 2)
	Perceived likelihood of harm from smoking2	1.00
	What is the chance that you will one day get heart disease if you continue to smoke cigarettes? [no chance (1), low chance (2), moderate chance (3), high chance (4), certain (5)]	.94
	What is the chance that you will one day get cancer if you continue to smoke cigarettes? [no chance (1), low chance (2), moderate chance (3), high chance (4), certain (5)]	.93
	What is the chance that you will one day get a permanent breathing problem if you continue to smoke cigarettes? [no chance (1), low chance (2), moderate chance (3), high chance (4), certain (5)]	.87
	Perceived severity of harm from smoking	.56
	How much would getting heart disease because of smoking affect your life? [not at all (1), a little (2), a moderate amount (3), a lot (4)]	.97
	How much would getting cancer because of smoking affect your life? [not at all (1), a little (2), a moderate amount (3), a lot (4)]	.97
	How much would getting a permanent breathing problem because of smoking affect your life? [not at all (1), a little (2), a moderate amount (3), a lot (4)]	.97
Quit intentions (Week 4)	How much do you plan to quit smoking in the next month? [not at all (1), a little (2), somewhat (3), very much (4)]	.98
	How interested are you in quitting smoking in the next month? [not at all interested (1), a little interested (2), somewhat interested (3), very interested (4)]	.95
	How likely are you to quit smoking in the next month? [not at all likely (1), a little likely (2), somewhat likely (3), very likely (4)]	.94

Note. Table reports standardized factor loadings.

¹Modeled as a second-order latent factor

²For purposes of estimation, we fixed the value of the loading for perceived likelihood at 1

Data Analysis

Analyses used Stata/SE version 14.1 and Mplus version 7.4 with two-tailed tests and a critical alpha of .05. Analyses used measures of negative affect and message reactance taken at immediate post-test, and measures of perceived likelihood and perceived severity taken at week 2 follow-up (Table 2). The outcome was quit intentions at week 4 follow-up.

First, we examined associations among the constructs using bivariate correlations; this analysis used mean scores for each construct. Next, we examined the association between the individual items and their latent constructs using a measurement model that treated the mediators and outcome as latent variables. The model specification process pointed toward modeling perceived risk as a second-order latent variable that was a function of perceived likelihood and perceived severity. Then, we examined the impact of pictorial warnings on quit intentions in an unadjusted structural model. Finally, we examined the simultaneous influence of the hypothesized mediator and suppressor variables using a final, adjusted structural model. Suppression occurs when the direct and mediated effects have opposite signs, in this case demonstrating that the mediator detracts from the effectiveness of pictorial warnings.[35] In contrast, a direct and mediated effect with the same sign signals mediation, indicating that the mediator contributes to the effectiveness of pictorial warnings.[35] Candidate mediator/suppressor variables were negative affect, message reactance, and perceived risk. The model tested theoretically-driven predictions[4, 15, 33, 36] about the indirect effects of pictorial warnings on quit intentions first via negative affect and message reactance, and then through perceived risk.

Intent-to-treat analyses included all participants randomized, using the last observation available.[37] Participants who had quit smoking at week 4 did not answer the quit intentions items, so we used the last observation available for them as well. When data remained missing, the models employed full information maximum likelihood estimation, an approach commonly recommended for structural equation models that makes use of all available observations.[38–40] We report results as standardized path coefficients (βs). Mediation analyses used bootstrapped 95% confidence intervals (CI) with 1,000 repetitions, as this approach does not assume that indirect effects are normally distributed.[41] We evaluated several indicators of acceptable model fit, including the root mean square error of approximation (RMSEA<.08),[42] the Tucker-Lewis Index (TLI>.95),[43] and the Bentler Comparative Fit Index (CFI>.95).[44]

Results

The measurement model fit the data well (RMSEA=.040 [90% CI=.036, .043], CFI=.997, TLI=.997); Table 2). Factor loadings for the indicator variables were all statistically significant and ranged from .75 to .98. Bivariate correlations between the model constructs ranged from −.24 to .38 (Table 3). Pictorial warnings increased negative affect, message reactance, and quit intentions (all p<.001, Table 3). Pictorial warnings did not increase perceived risk (p=.65, Table 3).

Table 3.

Means and bivariate correlations among variables in the multiple mediation model (n=2,093)

	Mean (SD)	1	2	3	4	5
Pictorial warnings (1)	–	–	–	–	–	–
Negative affect (2)	2.25 (1.21)	.38**	–	–	–	–
Message reactance (3)	2.08 (.89)	.22**	.03	–	–	–
Perceived risk (4)	3.55 (.63)	.01	.20**	−.24**	–	–
Quit intentions (5)	2.57 (1.07)	.08**	.35**	−.15**	.23**	–

Note. Analyses excluded 56 participants with missing data on at least one of the variables. Pictorial warnings coded as 1 (vs. text-only, coded as 0). Mediators measured at immediate post-test (negative affect, message reactance) or the week 2 follow-up survey (perceived risk). Quit intentions measured at week 4.

^*p<.05,

^**p<.001

The final structural model exhibited excellent fit (RMSEA=.038 [90% CI=.034, .041], CFI=.997, TLI=.997; Figure 2; Supplementary Table 1). Pictorial warnings generated higher quit intentions than text-only warnings (β=.08, p<.001). As expected, negative affect mediated the impact of pictorial warnings on quit intentions, such that pictorial warning exposure increased negative affect (β=.40, p<.001; Figure 2) which, in turn, was associated with greater quit intentions (β=.41, p<.001; mediated effect=.16, p<.001; Table 4). Negative affect was also associated with greater perceived risk (β=.30, p<.001), and perceived risk was associated with greater quit intentions (mediated effect through negative affect and perceived risk=.02, p<.001).

Figure 2.

Structural equation model assessing the impact of pictorial warnings on quit intentions (n=2,149)

Note. To simplify presentation, the figure omits factor loadings, non-significant pathways, and residuals. RMSEA=.038 [90% CI=.034, .041], CFI=.997, TLI=.997.

*p<.05

Table 4.

Mediation of association between pictorial warning exposure and quit intentions (n=2,149)

Mediation path	Mediated effect	95% CI
Pictorial warning ➔ negative affect ➔ quit intentions	.16	(.14, .19)
Pictorial warning ➔ negative affect ➔ perceived risk ➔ quit intentions	.02	(.01, .02)
Pictorial warning ➔ reactance ➔ quit intentions	−.04	(−.06, −.02)
Pictorial warning ➔ reactance ➔ perceived risk ➔ quit intentions	−.01	(−.02, −.01)

Note. Table reports standardized path coefficients for mediated effects. Mediators measured at immediate post-test (negative affect, message reactance) or the week 2 follow-up survey (perceived risk). Quit intentions measured at week 4.

Pictorial warnings also increased message reactance (β=.25, p<.001), and message reactance was associated with lower quit intentions (β=−.17, p<.001; mediated effect=−.04, p<.001, Table 4). Message reactance was associated with lower perceived risk (β=−.36, p<.001), which was in turn associated with greater quit intentions (mediated effect through reactance and perceived risk=−.01, p<.001).

Discussion

Pictorial cigarette pack warnings elicited stronger quit intentions than text-only warnings. Negative affect (which included fear, guilt, disgust, anxiety, and sadness) was a key driver of the effect of pictorial warnings on quit intentions. Compared to text-only warnings, pictorial warnings elicited more negative affect, which was associated with greater quit intentions. We also found in adjusted analyses that negative affect was associated with greater perceived risk of harm from smoking. Reactance, however, was associated with lower perceived risk. Perceived risk was associated with stronger quit intentions, building on prior work showing associations between risk perceptions and intentions and behavior.[45–48] The opposing associations of negative affect and reactance with perceived risk may offer insight into the well-established finding that pictorial cigarette pack warnings do not cause observable changes in risk perceptions.[4, 8, 9, 49]

Our study suggests that the emotion evoked by the warnings may be a precursor to beneficial changes in quit intentions, building on prior research about emotional reactions to pictorial warnings.[4, 8–12] This is an important point that the US courts did not acknowledge when criticizing pictorial warnings simply for evoking emotion.[7] A recent legal analysis of the pictorial warning lawsuits in the US concluded that the “warnings do not bypass reason simply by reaching for the heart,” and that the emotions evoked by the warnings should not deem them unconstitutional.[50] Indeed, we found pictorial warnings were effective because of the emotions they elicited.

Some have argued that discrete negative emotions fall into distinct categories and therefore play unique roles in shaping intentions and behaviors.[51–53] The Extended Parallel Process Model argues that fear is the most important affective motivator of behavior change.[33] However, we found support for a constructionist view of emotion[52] because – in the context of pictorial cigarette pack warnings – fear, guilt, disgust, anxiety, and sadness were highly correlated and functioned quite similarly. However, in contrast with negative affect, anger (a facet of reactance) was associated with lower quit intentions. This finding builds on previous research indicating that anger and fear produce opposite effects on intentions and behavior.[54–56] In the context of pictorial warnings, the distinctions between fear, guilt, disgust, anxiety, and sadness may not be meaningful, but it is crucial to distinguish between anger and other types of negative affect. Message reactance, which includes anger, works against the warnings’ purpose to motivate quitting, while increasing other types of negative affect advances this purpose.

As predicted, message reactance weakened the effect of pictorial warnings on quit intentions, although the magnitude of the mediated effect was small. Message reactance was also associated with lower perceived risk. Previous studies have found that pictorial warnings cause greater reactance than text-only warnings,[4, 22–24] but few studies have examined whether message reactance or other forms of defensive processing are associated with deleterious consequences, such as lower quit intentions.[24, 57, 58] Our experimental test of the suppression effects of message reactance following repeated exposure to pictorial warnings adds to this body of research. However, given research demonstrating the effectiveness of pictorial warnings, it would be unwise to conclude that pictorial warnings are counterproductive simply because they produce reactance, as others have argued.[23] In fact, the best warnings may be those that elicit strong reactions of all kinds, including reactance. Reactance had a small effect in the model and its influence did not undo the positive effects of warnings. Measuring message reactance can help to better identify the types of individuals who are resistant to the warnings and therefore may benefit from alternative interventions. Moreover, message reactance could be a particularly useful as a way of vetting candidate warnings in the early stages of message development and testing. Given a choice between two warnings that elicit similar levels of negative affect, message designers might choose the warning that elicits less reactance.

Study strengths include the use of an experimental design, a large and diverse sample of smokers who received the warnings on the cigarette packs they used every day, and the 4 week data collection period that allowed us to establish the temporality of mediated effects. However, the trial examined the potential effect of adding pictorial warnings to cigarette packs as well as implementing other label formatting changes required by the 2009 Tobacco Control Act compared with the present text-only warnings in the United States. While the trial aimed to compare the current warning policy to the new one in the Act, the use of this research design leaves open the possibility that the observed effects on mediators and intentions may be due to the combination of adding pictures and other changes (e.g., location, size, and content). The generalizability of these findings to different contexts (e.g., outside the US) or over a longer period of time has yet to be established. Finally, some of the mediated pathways were based on observational data, limiting our ability to assume causal associations between the model factors.

Conclusions

Understanding how pictorial warnings exert their influence can help researchers and policymakers design more effective warnings. Pictorial warnings elicited stronger quit intentions in the present study. However, message reactance partially suppressed pictorial warnings’ effect on quit intentions. We found that increased negative affect was a key mechanism by which pictorial warnings changed quit intentions. Negative affect was also associated with stronger perceived risk, which was in turn, associated with stronger quit intentions. Thus, negative affect is a vital and productive result of pictorial warning exposure.

What This Paper Adds

• In a four-week trial with smokers, we found that pictorial warnings increased quit intentions.

• Negative affect was a key driver of the association between pictorial warnings and quit intentions. Message reactance weakened this association, although pictorial warnings were more effective than text-only warnings on the whole.

• Negative affect and reactance had opposite and offsetting associations with perceived risk, which may explain why pictorial cigarette pack warnings do not change perceived risk.