Avoidance of Cigarette Pack Health Warnings among Regular Cigarette Smokers

Background— Previous research with adults and adolescents indicates that plain cigarette packs increase visual attention to health warnings among non-smokers and non-regular smokers, but not among regular smokers. This may be because regular smokers: 1) are familiar with the health warnings, 2) preferentially attend to branding, or 3) actively avoid health warnings. We sought to distinguish between these explanations using eye-tracking technology. Method— A convenience sample of 30 adult dependant smokers were recruited to participate in an eye-tracking study. Participants viewed branded, plain and blank packs of cigarettes with familiar and unfamiliar health warnings. The number of fixations to health warnings and branding on the different pack types were recorded. Results— Analysis of variance indicated that regular smokers were biased towards fixating the branding location rather than the health warning location on all three pack types (p < 0.002). This bias was smaller, but still evident, for blank packs, where smokers preferentially attended the blank region over the health warnings. Time-course analysis showed that for branded and plain packs, attention was preferentially directed to the branding location for the entire 10 seconds of the stimulus presentation, while for blank packs this occurred for the last 8 seconds of the stimulus presentation. Familiarity with health warnings had no effect on eye gaze location. Conclusion— Smokers actively avoid cigarette pack health warnings, and this remains the case even in the absence of salient branding information. Smokers may have learned to divert their attention away from cigarette pack health warnings. These findings have policy implications for the design of health warning on cigarette packs.


Introduction
A number of countries are now considering, or have already implemented, plain packaging of cigarettes, as recommended by Article 13 of the World Health Organisation's Framework Convention on Tobacco Control (WHO FCTC, 2005). Previous work from our laboratory has shown that plain packaging can increase attention to health warnings. In a series of eyetracking experiments, adult (Munafò et al., 2011) and adolescent (Maynard et al., 2013) nonsmokers and non-regular smokers made more saccades (eye movements) to health warnings, and fewer towards branding, on plain as compared with branded packs of cigarettes. This increase in attention to health warnings on plain packs is what would be predicted by models of natural image viewing (Parkhurst et al., 2002). Through sensory-driven bottom-up processes, attention is automatically drawn towards the most salient part of an image, where salience refers to the interplay between physical features and internal factors, such as the observer's intentions and goals, which can potentially override automatic shifts of attention (Vincent et al., 2009). However, this increase in attention to health warnings on plain packs was not observed among either adult or adolescent regular smokers. It is possible that the lack of attention to health warnings among regular smokers may be a result of attention being biased away from the visually salient bottom-up (physical) features of the health warnings with top-down, volitional control. There are a number of possible explanations for these voluntary shifts of attention, such as: 1) familiarity and therefore habituation to the health warnings, 2) a preference for branding, even in the absence of colours and logos, and/or 3) active avoidance of the health warnings.
Health warnings used in our previous eye-tracking studies were familiar to regular smokers, as they were taken from those currently used in the United Kingdom (UK). This familiarity perhaps led to reduced attention to health warnings because their impact had diminished over repeated exposure. Evidence from both Canada (Environics Research Group, 2007) and the UK (Hammond et al., 2007) suggests that new health warnings are more effective than old health warnings, which experience wear-out over time.
Alternatively, smokers' lack of visual attention to health warnings may be because their attention is instead drawn to the branding. In our previous studies, the plain packs still displayed the brand names of the cigarettes. It is possible that smokers continued to attend this area as they were interested in the branding that was available, thus reducing the amount of time spent attending the health warnings.
Finally, the lack of visual exploration of health warnings among regular smokers might be a result of active avoidance. Self-reported health warnings avoidance among smokers has been documented previously: interviews of Canadian regular smokers, found that 36% made some attempt at avoiding the warnings (Hammond et al., 2004). Health warning avoidance has also been shown to be greater for pictorial health warnings as compared with text-only warnings (Borland et al., 2013).
The present study aims to establish which of these three explanations accounts for why regular smokers do not attend cigarette pack health warnings. To establish whether familiarity was the most accurate explanation, smokers were presented with both familiar and unfamiliar health warnings. If familiarity was the cause, we predicted relatively greater attention to the unfamiliar as compared with the familiar health warnings. To establish whether regular smokers preferentially attend to branding or actively avoid health warnings, we included a 'blank' pack in addition to the branded and plain packs. The blank pack had all branding removed, and only the health warning present. If smokers preferentially attend to branding, they should allocate more attention to the health warnings than the blank region on the blank packs, as no branding is present. However, if smokers actively avoid health warnings, they should allocate more attention to the blank region on blank packs rather than the health warnings.

Design and Overview
This study used a repeated measures design with eye gaze location (i.e. saccade landing position) (health warning, branding), pack type (branded, plain, blank) and health warning familiarity (familiar, unfamiliar) as within-subjects factors. Eye-tracking equipment was used to measure the number of saccades made to health warnings and branding on the different pack types. Testing took place at the University of Bristol, and ethics approval was granted by the Faculty of Science Research Ethics Committee.

Participants
Thirty two participants were recruited from the staff and student population at the University of Bristol, and the general population. Participants were required to smoke five or more cigarettes a day and smoke within one hour of waking. All participants were required to be aged between 18 and 40, to have lived in the UK since 2008 (when pictorial warnings were introduced in the UK) and to primarily purchase their cigarettes from within the UK (more than 90% of the time, minimising exposure to non-UK health warnings).

Materials
Visual stimuli of branded and plain packs of cigarettes were identical to those used in our previous eye-tracking studies (Maynard et al., 2013;Munafò et al., 2011). Blank packs were created by removing all text from the plain packs, leaving only the health warning. Examples of the three pack types are shown in Figure 1. The 11 pictorial health warnings currently used in the UK come from a larger set of 42 European Union health warnings. The effectiveness of these health warnings was assessed in pre-study piloting. During this piloting, each of the health warnings were presented on screen individually and non-smokers and smokers rated each on four measures of effectiveness on a nine-point Likert scale: 1) 'Would this image prevent you from starting smoking? / Does this image make you think about quitting smoking?' 2) 'Does this image make you think about the health consequences of smoking?' 3) 'How unpleasant do you find this image?' and 4) 'How threatening is this image?' Scores for each of these questions were summed for each health warning, to provide an overall measure of effectiveness. These effectiveness scores were then used to select 20 of the 42 health warnings for use in the present study: 10 from those currently in use in the UK, to create a set of familiar health warnings, and 10 from those not used in the UK, to create a set of unfamiliar health warnings, ensuring that these two sets were matched for effectiveness [F (18) The 20 health warnings were paired with each of the 10 branded and plain pack stimuli, to create a total of 400 stimuli (200 branded, 200 plain). Each warning was also paired with the blank pack. Each participant was shown all 20 blank packs and a pseudo-random selection of 20 branded and 20 plain packs (where each health warning was presented once and each brand was presented twice, paired with both a familiar and an unfamiliar health warning).

Procedure
Following informed consent, participants completed questionnaires assessing compliance with the inclusion criteria and a carbon monoxide breath test to confirm smoking status. Participants were sat 57 cm from the LCD computer screen and fitted with an Eyelink II eye tracker (SR Research Ltd, ON, Canada) to measure eye movements from their dominant eye. A nine-point grid calibration and validation preceded the experiment, and eye movements were recorded at a sampling rate of 500 Hz and a spatial resolution of typically less than 0.4 degrees of visual angle. The main eye-tracking phase of the experiment was administered using MATLAB (version 2009a) and participants viewed five blocks of 12 images. Each block included two images from each of the six different stimuli types (branded, plain and blank packs, each with familiar and unfamiliar health warnings). Stimuli were presented for 10 seconds and were followed by a fixation cross, which acted to correct for drift due to head movements and ensured a fixed starting point for each trial. A recall phase followed each block to ensure that participants actively attended the images in the test phase. In the recall phase, participants were shown two cigarette pack stimuli and had five seconds to decide whether they had seen each image in the previous block. Participants then completed the Fagerström Test for Nicotine Dependence (Heatherton et al., 1991) and the quitting smoking Contemplation Ladder (Biener and Abrams, 1991). Participants were then fully debriefed and reimbursed £5.

Data Analysis
Eye-position data were analysed in the same way as in our previous studies (Maynard et al., 2013;Munafò et al., 2011). A 2 (eye gaze location: health warning, branding) × 3 (pack type: branded, plain, blank) × 2 (health warning familiarity: familiar, unfamiliar) analysis of variance (ANOVA) was used to analyse the data on the number of saccades. Interaction effects were explored by further stratified analyses corrected for multiple comparisons, using the Bonferroni method. In cases where Mauchly's Test of Sphericity indicated that the assumption of sphericity had been violated, Greenhouse Geisser corrected values were used. Effect sizes were calculated using Cohen's d for t-tests and eta-squared for ANOVA. For consistency, we use the term 'branding' to refer to the top section of the cigarette pack stimuli, even for the blank packs, where no branding is present.
A power analysis indicated that a sample size of n = 28 would be required to detect a difference of three eye movements (SD 8.5) towards the branding as compared to the health warning , with 80% power at an alpha level of 5%. This effect size (dz = 0.56) was estimated based on data from our previous studies (Maynard et al., 2013;Munafò et al., 2011), and assumes a correlation between conditions of r = 0.8.

Characteristics of Participants
Two participants were excluded from further analysis due to an inability to track their eye movements. Participants were therefore 30 regular smokers (63% male) with an average age of 21 years (SD = 3). On average, participants smoked 11 cigarettes a day (SD = 5), smoked their first cigarette within 45 minutes of waking (SD = 15) and started smoking by age 15 (SD = 2). None of the participants were attempting to quit smoking at the time of the study, based on scores of six or less on the quitting smoking Contemplation Ladder.
Conversely, an equal number of saccades were made to health warnings on branded and plain packs (t (28) < 0.001, p = 1.00, d = 0.08), but more saccades were made to health warnings on blank packs than either branded packs (t (28) = 3.85, p < 0.001, d = 0.53) or plain packs (t (28) = 4.00, p < 0.001, d = 0.44). There was no main effect of familiarity and no interactions including this factor.

Exploratory analyses
To describe the focus of participants' attention, a time-course analysis was conducted for each of the three pack types (see Figure 3). For each participant, each 1,000ms trial was divided into 10ms time bins for branding and health warnings. A value of 1 was assigned to a bin if a saccade fell on the respective area within that interval. An average per participant was then taken for each pack type. Time bins where participants were fixating the area outside of the cigarette pack, where they were making the actual saccadic eye movement, or blinked, were not included.
Broadly the same pattern of results was seen for branded and plain packs: throughout stimulus presentation, participants were more likely to be fixating the branding than the health warning. Participants always started trials by fixating the branding, as the branding replaced the fixation cross. Immediately after stimulus onset, the percentage of fixations on branding reduced so that between approximately 1,000ms and 3,000ms for branded packs, and between 2,000ms and 3,000ms for plain packs, participants fixated the branding on only 50% of trials. After 3,000ms, for both pack types, participants were fixating the branding area for approximately 70% of trials until the end of the stimulus presentation.
In contrast, for the blank packs there was a sharper decline in fixations on the 'branding' (i.e., the blank) region of the pack at the beginning of stimulus onset, with participants fixating the health warnings more than the blank area after approximately 500ms. After 2,000ms, however, the pattern reversed, with participants more likely to fixate the branding than the health warning, and after 4,000ms the pattern was comparable to that for branded and plain packs.

Discussion
Our results show that familiarity is not the cause of regular smokers' lack of visual attention to health warnings, as whether the health warning was familiar or unfamiliar had no effect on visual attention to the warning. Rather, our results indicate that both a preference for branding and an active avoidance of warnings, explain regular smokers' lack of attention to health warnings. The time-course analysis shows that for blank packs, with no branding present at all, there is a large shift in attention towards the health warnings within the first few hundred milliseconds of stimulus onset. However, for both branded and plain packs, attention remains primarily in the area of branding and, as compared with blank packs, there is a slower and smaller increase in attention to the health warning in the first 2,000ms. We assume that fixations early in the time-course of viewing are strongly influenced by visual salience (Parkhurst et al., 2002). This indicates that branded and plain packs contain sufficient salient visual information about cigarette branding to compete with the health warning information. This suggests that smokers' lack of attention to health warnings on plain packs must in part be due to a preference for the branding name.
Approximately 2,000ms after stimulus onset, however, and for the remaining 8,000ms of cigarette pack presentation, for all three pack types, smokers made voluntary and sustained shifts in attention towards the branding, and away from the health warnings. We interpret these results as indicating that regular smokers actively avoid health warnings on cigarette packs, via top-down voluntary control of attention.
The effect of health warning avoidance is disputed in the literature, with some arguing that 'cigarette warning labels are useless' (Ruiter and Kok, 2005) because of the defensive reactions they promote among smokers, although others counter these claims, arguing that avoidance is an indicator of warning engagement Fathelrahman et al., 2009;Hammond et al., 2004).Future research should determine whether visual avoidance of health warnings, as observed here, is associated with self-reported levels of avoidance and with outcomes previously linked with avoidance, such as long-term recall of the warnings, knowledge of the health risks of smoking and future smoking cessation. If a negative association is observed, research should focus on understanding the reasons for this avoidance, and on designing cigarette packs and health warnings which prevent this.
A potential limitation of this study is the blank pack design. As intended, the 'blank' pack looked like a cigarette pack with the branding removed. However, it is possible that the attention to this area of the pack, which we have ascribed to warning avoidance, maybe the result of an interest in a particularly novel cigarette pack (i.e., one without any branding). While this is possible and may explain some of the attention directed to this area of the pack, it is unlikely that this explains why smokers attended this region of the pack for approximately 8,000ms, for each of the 20 blank packs shown to them. Previous studies have relied on self-reported measures of explicit behavioural avoidance of health warnings and have found low to moderate levels of avoidance among smokers (Borland et al., 2013;Environics Research Group, 2007;Hammond et al., 2007;Hammond et al., 2004). By measuring health warning avoidance at a more implicit, unconscious level, using eye-tracking technology, we have found clear evidence of visual avoidance of health warnings among regular smokers, even after accounting for smokers' interest in branding. Examples of branded, plain and blank pack stimuli, respectively.  Number of saccades to branding (grey line) and health warnings (black line) on the three pack types. Error bars represent adjusted standard errors corrected for within-subjects comparisons.  Time-course analysis across the entire stimulus presentation of 10,000 ms to show the percentage of trials (20 trials per pack type) in which participants were fixating the branding (grey lines) as opposed to health warnings (black lines) for branded (A), plain (B) and blank (C) packs of cigarettes.