Article Text


Effects of brand variants on smokers’ choice behaviours and risk perceptions
  1. Janet Hoek1,
  2. Philip Gendall1,
  3. Christine Eckert2,
  4. Joya Kemper1,
  5. Jordan Louviere3
  1. 1Department of Marketing, University of Otago, Dunedin, New Zealand
  2. 2University of Technology Sydney, Business School, Marketing Discipline Group, Ultimo 2007 NSW, Australia
  3. 3Institute for Choice, University of South Australia, Sydney, Australia
  1. Correspondence to Dr Janet Hoek, Department of Marketing, University of Otago, P.O. Box 56, Dunedin 9054, New Zealand; Janet.hoek{at}


Background Australian tobacco companies have introduced evocative variant names that could re-create the aspirational connotations plain packaging aims to remove. To inform future regulation, we explored how brand descriptors affected smokers’ responses to plain packs featuring different variant name combinations.

Methods An online survey of 254 daily smokers or social smokers aged between 18 and 34 used a within-subjects best-worst experiment to estimate the relative effects of variant names. A 2×4×4×4 design contained four attributes: quality (premium or none), taste (smooth, fine, rich or none) connotation (classic, midnight, infinite or none) and colour (red, blue, white or none). In a between-subjects component, respondents evaluated one of two alternative packs according to its perceived harm and ease of quitting.

Results The most important variant attribute was connotation, followed by taste, colour and quality; within these attributes, the most attractive descriptors were ‘classic’ and ‘smooth’. We identified four distinct segments that differed significantly in their sociodemographic attributes and variant preferences, although not in their perceptions of the harm or quitting ease associated with two different variants.

Conclusions Some descriptors significantly enhance the appeal of tobacco products among different groups of smokers and may undermine plain packaging's dissuasive intent. Policymakers should explicitly regulate variant names to avoid the ‘poetry on a package’ evident in Australia. Options include disallowing new descriptors, limiting the number of descriptors permitted or banning descriptors altogether.

Statistics from


Tobacco companies use packaging to create connotations that allure new smokers and reinforce existing smokers’ behaviour.1–4 Brand livery, which includes colours, logos and other imagery, encourages consumers to draw inferences about their likely consumption experience and ensures smoking not only satiates a physiological need, but maintains and projects desired social identities.5–7

Plain packaging removes these appealing stimuli, replaces them with dissuasive colours and much larger warnings, and transforms packs from a sophisticated marketing medium into a highly unattractive accessory.8 In Australia, the only country thus far to have introduced plain packaging, the brand name appears in a prescribed font and size on the front of the pack face and on top of the pack lid. Tobacco companies have resisted plain packaging and opposed it in every jurisdiction where it has been mooted by taking (or threatening) legal action9 and misrepresenting research evidence.10 ,11 Evidence from Australia suggests companies are also subverting plain packaging's intention by promoting colour connotations, using more evocative and detailed variant names, introducing new ‘value’ brands, offering ‘bonus’ sticks and developing novel flavouring devices such as menthol capsules.12

Although Australian legislation anticipated many of the tobacco industry's likely responses to plain packaging, it does not restrict variant names, also known as brand descriptors.13 Since brand descriptors have a long history of deception, many countries now prohibit terms such as ‘light’ and ‘mild’, which create misleading reduced harm connotations.14–18 Tobacco companies responded to these restrictions by introducing synonyms; terms such as ‘smooth’, ‘fine’ and ‘rich’, to recreate and perpetuate the deceptive connotations ‘light’ and ‘mild’ generated.15 ,16 ,18–20 Their replacement strategy appears to have been successful as smokers report shifting from brands featuring banned descriptors such as ‘light’ to brands using synonymous replacement descriptors.21

In anticipation of further regulation, tobacco companies began colour coding packs as an additional means of communicating variant information.22–24 Red packages denote regular variants, blue indicate ‘mild’ options, while white and silver define ‘light’ and ‘ultra-light’ variants, respectively.23 ,24 These colours reinforce smokers’ mistaken belief that some variants are less harmful than others. Such beliefs are now so well established that, even when shown packs featuring only a colour and no other information, smokers infer strength, and thus perceived harm, from the colour.20 ,22 ,25 Tobacco companies have prepared for further regulation by incorporating colours into descriptors that feature on tailor-made as well as RYO brands, for example, ‘Marlboro Red’. Continued pairing of visual stimuli and verbal descriptors ensures plainly packaged tobacco products featuring these variant names evoke historic brand livery and the appealing attributes this imagery formerly conveyed.

Brand families have evolved to include multiple variants, each with a unique descriptor. Since plain packaging, these descriptors have evolved to include more abstract adjectives that attempt to recreate the evocative connotations previously aroused by carefully designed brand livery.7 ,12 Aspirational descriptors such as ‘Peter Jackson Hybrid Blue Fresh’ or ‘Easy Flow Ultimate’ also occupy more pack space, thus displacing dissuasive colours and potentially ameliorating the dissuasive effects of plain packaging.12

Evidence that, as with colours, these more abstract descriptors also influence consumers’ behaviour suggests these newly introduced terms aim to shape smokers’ perceptions and actions.26 Recent experimental work found that descriptors affected smokers’ perceptions of taste, tar and nicotine delivery and quality.17 However, this earlier study was undertaken before plain packaging's design had been confirmed and thus could not use the stimuli specified in the Australian legislation. Further, as descriptors have become increasingly evocative since plain packaging came into effect, the study could not test the variety of variant names now evident in Australia, or the combinations in which these are used. We extended this initial research by using the Australian plain pack design, and testing different types and combinations of variant names.



To evaluate smokers’ preferences for different brand variants, we used a best-worst scaling (BWS) choice task (the results reported are based on the ‘best’ choice responses). Respondents evaluated experimentally designed choice scenarios in which they were offered three choices and asked to select their most preferred and least preferred options. BWS is an alternative to discrete choice experiments and the particular BWS experiment we employed is known as case 3 BWS.27

Our BWS experiment used a 2×4×4×4 orthogonal design with four attributes: quality (premium or no descriptor), taste (rich, smooth, fine or no descriptor), connotation (classic, midnight, infinite or no descriptor) and colour (red, blue, white or no descriptor). This design produced 128 different potential variant combinations that varied in length and composition; for example, ‘Premium’, ‘Premium Fine’, ‘Premium Fine Midnight’, ‘Premium Fine Midnight Red’. We identified potential variant names after reviewing Australian retail websites and selecting those names that corresponded to the attributes we wished to test. A graphic artist created images corresponding to each of the possible variant combinations; online supplementary figure S1 contains examples of the stimuli used.


We used a balanced incomplete block design to generate 16 blocks; each block comprised 16 choice sets, and each set featured three options (see online supplementary figure S1). Respondents saw one randomly chosen block of choice sets and were asked to imagine they had gone into a store to buy cigarettes or tobacco, and had only the options shown available to purchase. They were also told that each option sold for the same price, before being asked: “Which of these packs would you be most likely/least likely to buy?” Respondents could select a ‘none of these’ option, which allowed for the possibility that they did not prefer any of the packs.

In a between-subjects component of the study, respondents saw one of two images that featured either the single descriptor ‘red’ or the compound descriptor ‘premium rich midnight red’. Respondents used a zero–10 scale to evaluate the variant's perceived harmfulness and the ease with which smokers could quit that variant. We also collected details of respondents’ smoking behaviour, quit experiences and intentions, and demographics.


We sourced a sample of self-defined daily smokers or occasional smokers aged between 18 and 34, from the New Zealand Research Now online panel provider, and collected data between 30 July and 7 August 2013. We selected this demographic because smoking prevalence in New Zealand peaks among young adults, making them a priority group for policy interventions. Panel members received an email inviting them to participate in a study that examined tobacco packaging and directing them to the survey website. Once they had accessed the survey website, respondents answered screening questions about their smoking status (never smoker; former smoker; daily smoker, social smoker who smokes only with others, or social smoker who smokes mainly with others). Those who were not self-defined daily or social smokers were excluded from the study.


We used Scale-Adjusted Latent Class Models (SALCMs) to identify a statistically defensible number of preference and scale classes from the choice experiment data. A recent development in discrete choice experiments, SALCMs recognise that respondents may differ in their preferences for the options presented as well as in their scales (or choice consistency), or both. By removing the scale factor confound in multinomial choice models, SALCMs avoid biases that occur as a result of choice variability; the resulting unbiased estimates are also statistically more efficient.28 SALCMs produce probabilistic segments; that is, segment boundaries are fuzzy and the arbitrary case allocation that typically occurs at the margins of traditional segments is avoided.

We used the syntax module of Latent Gold V.5.0 software to undertake these analyses as this software allows preference parameters to differ for discrete, but unobserved (latent) classes of respondents, while also allowing the underlying variability of random errors to differ between several discrete latent classes. SALCMs also allow membership of latent preference and scale classes to be predicted as functions of demographic and other covariates. The analyses presented use preferred pack data (ie, analyse the options respondents indicated they would be most likely to buy).

To examine perceptions of the harm and quitting ease of the two stimuli used in the comparison test, we used independent sample t test to estimate differences between the mean scores and sample proportions. These analyses were undertaken using SPSS V.21.


Sample characteristics

A total of 254 respondents completed the study. Table 1 summarises the overall sample characteristics.

Table 1

Sample characteristics

BWS model

The best statistical model had four preference classes and two scale classes. Respondents in scale class 2 had a scale 3.6 times larger than those in scale class 1, indicating that they were much more consistent in their choices. Women had more consistent choice patterns than men, as did daily smokers relative to social smokers, and New Zealand Europeans relative to Ma¯ori, Pacific or Asian respondents. These respondents’ more strongly defined preferences imply they may be less amenable to interventions designed to promote behaviour, such as making a quit attempt.

Figure 1 illustrates the relative importance overall of the four attributes tested. The most important is the connotation descriptor (classic, infinite, midnight), followed by taste (smooth, fine, rich), colour (red, white, blue) and finally, ‘premium’ quality. The significance of the likelihood ratio test for the difference between the full model and the model without the connotation descriptor is p=0.0001; for taste it is also p=0.0001, while for colour and quality, significance p<0.05. All the attributes thus significantly improve the fit of the model estimated.

Figure 1

Overall attribute importance.

We first developed parameter estimates for the whole sample using an aggregate Multinomial Logit Model, and then for the four preference segments identified by the SALCM. Table 2 presents the results for each of the four attributes tested. Although not a feature of the experimental design, the position of each image on a show card had a significant effect on respondents’ choice behaviours. This finding is not surprising since the Latent Gold software treated the choice ‘none of these’ as an attribute level for position; consequently, a preference for not choosing any of the three options is captured in this positioning effect (and some respondents may also use image position as a choice criterion). The estimates reported in table 2 are thus controlled for position of image in choice set as well as for differences between scale classes.

Table 2

Multinomial logit and latent class parameter estimates

The aggregate sample model estimates indicate differences in preferences between some attribute levels; specifically, premium, smooth and classic were preferred to other options, while the no colour option was significantly more attractive than the red, blue or white options. Figure 2 graphically displays the overall preferences for the attributes and levels tested.

Figure 2

Specific attribute importance.

The preferred variant was ‘Premium Smooth Classic’, followed by ‘Premium Smooth Classic Red’. However, while the attributes ‘Classic’ and ‘Smooth’ were significantly preferred to the other attribute levels tested, the term ‘Premium’ was only marginally preferred as a quality descriptor. While the preference for no colour descriptor relative to red, blue or white may appear anomalous, it is explained by strong variations in each segment's colour preferences, which mirror their current variant choices (shown in table 3). Since aggregate results such as this can mask differences between different preference groups, we now focus on the segment responses.

Table 3

Characteristics of preference segments

The SALCM approach simultaneously reveals differences in preferences and choice consistency (scale), and thus allowed us to parsimoniously capture differences within discrete classes or segments. Members of the first segment preferred the smooth taste descriptor, found the midnight connotation significantly less appealing, and preferred ‘white’ or no colour descriptor over ‘blue’. The second segment preferred the premium quality descriptor, the rich taste descriptor, the midnight connotation and the red colour, but had significantly lower preferences for the fine, infinite and white descriptors. Members of segment 3 preferred either the rich or no taste descriptor, the classic or no connotation and the red descriptor, and had significantly lower preferences for the smooth and fine taste descriptors, midnight and infinite connotations, and blue or white colour descriptors. The final segment preferred the smooth taste descriptor, the classic connotation and blue, white or no colour descriptors. Members of this segment had significantly lower preferences for the rich taste descriptor, the midnight connotation and the red colour descriptor.

Segments differed not only in their preferences, but also in their membership profiles. The first preference segment constituted 14% of the sample and its members were significantly more likely than expected to be men; they were also significantly more likely to smoke an ‘other’ cigarette colour. The second segment constituted 26% of the sample; it was significantly more likely than expected to have a high level of education. Segment 3 constituted 21% of the sample and its members were significantly more likely than expected to be women, not Asian, daily smokers with less education, who smoke red tobacco brands and roll their own cigarettes. The final segment was the largest and constituted 39% of the sample. Its members were significantly more likely than expected to be social smokers; they were also more likely to smoke brands with blue, white or silver descriptors, and they had higher 6-month quit intentions. There are also relatively large differences among the segments on ethnicity, but the small numbers of Maori, Pacific and Asians in the sample means many of these differences do not reach statistical significance. Table 3 summarises the covariates that are significantly associated with preference segment membership probabilities.

Harm and quitting ease perceptions

In the between-subjects component of the study, we examined how respondents perceived one of two stimuli that featured either a compound or simple descriptor. As table 4 shows, irrespective of the variant they saw, respondents did not differ in their perceptions of that variant's harmfulness or the ease with which smokers could quit smoking it. Nor did perceptions vary across variants, although members of segment four, who were significantly less likely to report smoking a ‘red’ variant or to select a variant featuring ‘red’ in the choice experiment, attributed the highest harm score to the ‘red’ pack. By contrast, members of segment 3, who were significantly more likely to report smoking a ‘red’ variant and to select a variant featuring ‘red’, gave the lowest perceived harm score to the ‘red’ variant. The relatively large number of segments resulted in small cell sizes, which reduced our ability to identify significant differences. Overall, respondents generally saw variants as harmful, though not very harmful, and moderately easy to quit.

Table 4

Perceptions of harm and quitting ease for two variants


Some brand variant labels influenced smokers’ cigarette pack choices, and a pack featuring the compound descriptor: ‘Premium Smooth Classic Red, Blue or White’ was significantly preferred to one with no variant label, or to one featuring a single colour descriptor, such as ‘Red’, ‘Blue’ or ‘White’. Our results also suggest that different variant names appeal to distinct smoker segments; for example, ‘Rich Classic Red’ appealed more to women, Ma¯ori or Pacific, daily smokers with less education, who smoke roll-your-own tobacco. Perhaps predictably, members of this segment reported they currently smoked cigarettes (or loose tobacco) featuring a ‘red’ descriptor; however, instead of choosing brands featuring only this single descriptor, they preferred a more evocatively labelled brand.

‘Rich, Classic, Red’ offer more complex connotations that extend well beyond those offered by ‘Red’ alone: ‘Classic’ and ‘Rich’ suggest a high-quality brand, while the double-entendre of ‘Rich’ emphasises both flavour and prestige. These metaphors seem intuitively likely to appeal to women with less education who occupy lower socioeconomic positions, and who preferred this compound descriptor to options featuring only their existing brand descriptor, ‘Red’. This interpretation reflects findings from earlier qualitative studies, which explicated the important symbolic qualities brands confer on smokers,5 and is consistent with our wider understanding of tobacco branding, which smokers use to access identities that offer a more appealing reality.4 ,6 ,22

Plain packaging removes brand imagery and logos, the overt visual stimuli on which smokers draw to construct social personae.3 ,8 However, our findings suggest that, even in the absence of these pack elements, variant names may evoke connotations formerly aroused using visual imagery. Use of terms such as ‘classic’ and ‘rich’, and newly emerging descriptions noted by Scollo et al,12 such as ‘New York blend’ and ‘silver fine scent’, suggest tobacco companies are developing more metaphoric variant names to offset the removal of brand imagery.

For practical reasons, we tested a limited number of variant attributes and labels; tobacco companies have now developed a much wider array of descriptors as they attempt to recreate the associations visual brand imagery once elicited. Our findings demonstrate that using certain words or combinations of words as brand descriptors makes cigarettes more attractive to smokers. The results also suggest that compound descriptors are more attractive than simple descriptors, presumably because they arouse more powerful and diverse connotations.

Evidence that different smoker groups exhibited varied responses to different descriptors indicates that variant names enable tobacco manufacturers to continue targeting their brands at specific groups. Tobacco companies have developed descriptors that resonate with smokers who seek quite different attributes. Whether they wish to associate themselves with prestige (‘premium’ and ‘rich’), seek the illusion of reduced harm (‘fine’ and ‘white’), or hope to sanction their behaviour by associating it with cleansing attributes (‘white’), descriptors allow smokers to identify with brands that continue to promise these attributes.

Our study has some limitations. First, we did not use a simple random sample; however, following similar studies, the sample was randomly selected from the panel and was not chosen because of any attribute related to the research objectives.17 Second, we examined a specific age demographic; although there is no a priori reason to expect responses to differ across other age groups, replication studies are required to test this assumption. Third, although BWS studies estimate choice behaviour, this measure is not necessarily equivalent to actual behaviour. Thus, while our findings extend perceptual experiments, more naturalistic studies are required to estimate actual behaviour. Finally, the large number of segments identified reduced the power required to test whether segment members had different perceptions of each variant's harm and addictiveness. Future experimental work could draw on a larger sample to allow more detailed analyses by smokers’ current behaviour, particularly with respect to the brand and variant currently smoked, and enable comparison of a wider array of variant combinations. Such work could use our findings and emerging practice in Australia to identify compound descriptors that merit further analysis.

Plain packaging aims to reduce the appeal of smoking as well as prevent tobacco companies from using colour or other brand imagery to create misleading associations, as they did when ‘light’ and ‘mild’ were deemed misleading and deceptive.19 ,23 Our results show respondents had similar levels of misperception irrespective of whether they saw a single (‘red’) or multiple (‘premium rich midnight red’) descriptor. Further research, using multiple exposures and more varied descriptor combinations, could extend these findings. Notwithstanding the need for replication studies, evidence that descriptors enhanced pack appeal among some groups suggests policymakers drafting plain packaging legislation should consider regulating variant names.

Given how Australian tobacco companies have proliferated variant names, regulations that prohibit specific terms seem unlikely to provide adequate protection as synonyms will quickly emerge. Instead, policymakers in France, the UK, Ireland and New Zealand, all of which have announced their intention to implement plain packaging, could refuse to allow descriptors not in use for a defined period prior to the introduction of plain packaging legislation. Alternatively, they could limit how many descriptors packs may feature, as Uruguay has done. Yet while these measures would prevent the proliferation of descriptors seen in Australia, they would not eliminate the attractive connotations created by existing descriptors.23 ,24 To address this problem, policymakers could ban descriptors altogether, on the grounds that these create nuanced appeals that target specific smoker groups and undermine the intent of plain packaging. We suggest further consideration of this latter option, which would eliminate a marketing loophole that allows use of metaphoric attributes to appeal to specific smoker subgroups.

What this paper adds

  • Tobacco variant names or descriptors exert a significant influence on smokers’ choice behaviours and enable tobacco companies to target their brand offerings at different segments of smokers.

  • Policymakers introducing plain packaging legislation should regulate variant names so tobacco companies cannot use evocative descriptors to recreate associations formerly evoked by on-pack brand imagery and undermine the intent of plain packaging.

  • Possible policy options include disallowing descriptors not in use prior to the introduction of plain packaging legislation, limiting the number of descriptor terms permitted, or banning all descriptors, given these create nuanced appeals that target specific smoker groups and undermine the intent of plain packaging.


View Abstract
  • Supplementary Data

    This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.

    Files in this Data Supplement:


  • Correction notice This article has been corrected since it was published Online First. A new affiliation has been provided for Christine Eckert.

  • Acknowledgements The authors would like to thank Professor John Rose, Institute for Choice, University of South Australia, who assisted with the experimental design development, and Julie Jeon, the graphic artist who prepared and refined the test stimuli.

  • Contributors JH conceptualised the project and, with PG, designed the questionnaire and oversaw the data collection. JK was a student who worked under JH's and PG's supervision; she contributed to the research design and data collection, and to the initial analyses and interpretation. CE and JL analysed the choice data. JH led the manuscript development and is the guarantor; all the authors have seen and approved the final manuscript version.

  • Competing interests None.

  • Ethics approval University of Otago Delegated Authority acting for the University Human Ethics Committee.

  • Provenance and peer review Not commissioned; externally peer reviewed.

Request permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.