section5.5janvier13:Layout 1 13/01/2009 10:07 Page 287
5.5 Measures to evaluate the effectiveness of tobacco product labelling policies
Background prohibit packaging elements that and legible” (WHO, 2003). Beyond are deemed to be misleading to these minimum requirements, Arti- The cigarette package serves as the smokers. As a consequence, cle 11 also states that warnings cornerstone of tobacco marketing labelling policies have begun to alter “should” cover 50% or more of a and advertising campaigns (Slade, the traditional appearance of the package’s principle surfaces, and 1997; Pollay, 2001). Package design cigarette package. “may” be in the form of pictures. helps to reinforce brand imagery The importance of tobacco label- To date, at least eight countries communicated through other media ling policies is highlighted in Article have implemented picture-based and plays a central role in retail 11 of the WHO FCTC (WHO, 2003). health warnings that meet the marketing. The importance of Article 11 sets international stan- FCTC’s “recommended” standard cigarette packaging only increases dards for packaging and labelling of (see Figure 5.27). A number of other as other forms of marketing are tobacco products in three broad jurisdictions, including the European restricted, as indicated in the fol- categories: 1) mandatory health Union, have recently implemented lowing quote from a Philip Morris warnings; 2) restrictions on brand prominent text warnings which meet executive: "Our final communication descriptors, such as the use of “light” the minimum FCTC standard. More vehicle with our smoker is the pack and “mild”; and 3) information on obscure text warnings remain in itself. In the absence of any other cigarette contents and emissions 1 many other markets, including the marketing messages, our pac- (Figure 5.26 ). USA, China, and Russia. kaging...is the sole communicator of our brand essence. Put another Health warning labelling Constituents & emissions way—when you don’t have anything labelling else—our packaging is our mar- Cigarette packages in the vast keting.” (Alechnowicz & Chapman, majority of countries carry a health There is general agreement that 2004). warning (Aftab et al., 1999). How- tobacco packaging should include Governments in many juris- ever, the position, size, and general at least minimal information about dictions have begun to apply greater strength of these warnings vary some of the hazardous and addi- restrictions on tobacco labelling. As considerably across jurisdictions. ctive constituents in tobacco and much as half of the package is now FCTC Article 11 requires that tobacco smoke. FCTC Article 11 used by regulators to communicate package health warnings must requires that packages contain the health effects of smoking. cover at least 30% of the package “information on relevant consti- Governments have also begun to surface and be “large, clear, visible, tuents and emissions of tobacco
1Tobacco labelling policies apply to a broad range of tobacco products, including a range of combustible products, such as cigars, and the packaging of loose or “fine cut” tobacco, as well as non-combustible tobacco products. However, much of this section will focus on labelling policies for factory-made, “pre-packaged” cigarettes given that they are the primary target of labelling policies, and the area in which most research has been conducted. Labelling policies for other types of products will be described briefly in a separate section to follow.
287 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 288
IARC Handbooks of Cancer Prevention
1. Each Party shall, within a period of three years after entry into force of this Convention for that Party, adopt and implement, in accordance with its national law, effective measures to ensure that:
(a) Tobacco product packaging and labelling do not promote a tobacco product by any means that are false, misleading, deceptive or likely to create an erroneous impression about its characteristics, health effects, hazards or emissions, including any term, descriptor, trademark, figurative or any other sign that directly or indirectly creates the false impression that a particular tobacco product is less harmful than other tobacco products. These may include terms such as “low tar”, “light”, “ultra-light”, or “mild”; and
(b) Each unit packet and package of tobacco products and any outside packaging and labelling of such products also carry health warnings describing the harmful effects of tobacco use, and may include other appropriate messages. These warnings and messages:
(i) shall be approved by the competent national authority, (ii) shall be rotating, (iii) shall be large, clear, visible and legible, (iv) should be 50% or more of the principal display areas but shall be no less than 30% of the principal display areas, (v) may be in the form of or include pictures or pictograms.
2. Each unit packet and package of tobacco products and any outside packaging and labelling of such products shall, in addition to the warnings specified in paragraph 1(b) of this Article, contain information on relevant constituents and emissions of tobacco products as defined by national authorities.
3. Each Party shall require that the warnings and other textual information specified in paragraphs 1(b) and paragraph 2 of this Article will appear on each unit packet and package of tobacco products and any outside packaging and labelling of such products in its principal language or languages.
4. For the purposes of this Article, the term “outside packaging and labelling” in relation to tobacco products applies to any packaging and labelling used in the retail sale of the product.
WHO (2003)
Figure 5.26 WHO FCTC Article 11: Packaging and labelling of tobacco products
products as defined by national three emissions in the mainstream generated under the ISO testing authorities.” At present, however, smoke: tar, nicotine, and carbon method are unrelated to individual national authorities have taken monoxide (CO). Emission levels levels of exposure or risk (Burns et much different approaches to are generated by machine- al. , 2001), there are growing calls labelling constituents and emis- smoking cigarettes according to a from within the tobacco control sions, and there remains con- standard set of puffing conditions; community for the ISO numbers to siderable disagreement regarding typically the International Stan- be removed from packages. Some what should be considered “rele- dards Organization (ISO) method, jurisdictions have supplemented vant” information. which serves as the current the ISO numbers with additional The current regulatory practice international standard. However, emission information. For exam- in many jurisdictions is to require in light of research indicating that ple, Canada increased the list of manufacturers to print levels for the tar and nicotine levels emissions that must be reported
288 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 289
Measures to evaluate the effectiveness of tobacco product labelling policies s g n i n r a w
d e s a b - e r u t c i P
7 2 . 5
e r u g i F
289 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 290
IARC Handbooks of Cancer Prevention
(by adding benzene, formal- false, misleading, deceptive or For example, consumers who buy dehyde, and hydrogen cyanide to likely to create an erroneous loose or fine-cut tobacco, without tar, nicotine, and CO), and impression about its any manufactured packaging, requires a second set of numbers characteristics, health effects, may not be exposed to product from a more intensive machine hazards or emissions, including health warnings. Even consumers smoking method for each emis- any term, descriptor, trademark, who buy fine-cut tobacco, sold in sion (Figure 5.28). Other figurative or any other sign that government-mandated packaging, jurisdictions have replaced quan- directly or indirectly creates the will have different patterns of titative emission values with false impression that a particular exposure than those who smoke descriptive, non-numerical infor- tobacco product is less harmful manufactured cigarettes, and who mation on hazardous emissions than other tobacco products. are likely to be exposed to the and toxicants. A consensus has These may include terms such as warnings each time they reach for yet to emerge on “best practices” “low tar,” “light,” “ultra-light,” or the package. As a result, studies for this area of tobacco labelling “mild”. Although there is evidence conducted in markets with a policy. to suggest that other packaging considerable proportion of fine-cut elements, such as the use of tobacco sales, such as the United Brand descriptor labelling colour, may also create mis- Kingdom, New Zealand, and leading perceptions of risk Thailand, may need to stratify for Tobacco manufacturers incor- (Wakefield et al. , 2002), “light” and fine-cut versus manufactured or porate a variety of common terms “mild” descriptors are the only mixed use. Smuggled or con- into the names of their cigarette packaging elements to be traband cigarettes may also alter brands. Words such as “light” and restricted to date. patterns of exposure in cases “mild” are ostensibly used to when the contraband product is denote flavour and taste; Methodological issues in evalu- not manufactured to the same however, “light” and “mild” brands ating tobacco labelling policies labelling specifications. are often promoted as “healthier” products and are typically applied Evaluating tobacco labelling Issues in attribution: dealing with to brands that generate lower policies presents several unique multiple sources of health infor - machine levels of tar (Pollay, challenges; this section reviews mation: 2001; Pollay & Dewhirst, 2002). some of the principal methodo- Not surprisingly, “light” and “mild” logical and analytical consi- Health behaviours with multiple brands are perceived by many derations. determinants present a challenge consumers to deliver less tar and to policy evaluation. The problem lower risk than “regular” or “full “Alternative” tobacco products: of attribution is particularly acute flavour” varieties despite evidence for health warning labels. First, to the contrary (Ashley et al., Labelling policies have generally labelling policies are often 2001; Shiffman et al. , 2001). been designed with factory-made, implemented simultaneously with A growing number of juris- pre-packaged cigarettes in mind. other tobacco control measures, dictions, including Brazil and the However, a substantial proportion including increases in taxation and European Union, have prohibited of tobacco users throughout the smoke-free policies. As a result, it the use of “light” and “mild” on world use tobacco products that is difficult to isolate the effect of an packages. Similar prohibitions are are either packaged in a different individual policy on overall pre- proposed in FCTC Article 11: way, or have no manufactured valence. Second, many of the "tobacco product packaging and packaging at all. This has specific themes and messages in labelling do not promote a tobacco important implications for patterns labelling policies are commu- product by any means that are of exposure to health warnings. nicated through other sources.
290 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 291
Measures to evaluate the effectiveness of tobacco product labelling policies
European Union (United Kingdom): Three ISO emissions
Canada: Six ISO emissions and ‘Health Candada Intense’ emission
Australia: Descriptive information
Figure 5.28 Constituent labelling policies in the European Union, Canada and Australia
291 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 292
IARC Handbooks of Cancer Prevention
Mass media campaigns and implications for regulators in terms Youth: health professionals often target of ensuring periodic changes to the the same health effects, warnings, as well as studies that One policy-relevant question particularly with regards to com- compare labelling policies across concerns the impact of warning mon diseases such as cancer and jurisdictions. For example, a recent labels in reducing youth uptake. cardiovascular disease. The study found that new text-based Evaluating the impact of health impact of package-based labelling warnings, introduced in the United warnings among youth during policies may also be confounded Kingdom in 2003, were con- periods of smoking initiation with health warnings in other siderably more likely to be noticed requires a different conceptual settings. Various jurisdictions than Australian text-based war- approach. Given that the cigarette require health warnings in retail nings, which were only slightly package serves as the medium for outlets and warnings on print smaller, but had been in place for labelling policies, consumption advertisements for tobacco pro- more than eight years at the time of levels may be positively asso- ducts. Third, perceptions of risk the survey (Bornstein, 1989). ciated with knowledge of the and health knowledge are Ideally, labelling policies should be warning labels. In other words, influenced by an inter-related set evaluated at similar post-imple- individuals who smoke 20 of factors at the individual, social, mentation dates; at the least, cigarettes a day will be exposed to and environmental level. Few differences in follow-up periods the warnings more frequently than studies are able to measure more should be clearly noted and taken individuals who smoke less than than a small number of these into account when interpreting daily. Furthermore, “occasional” factors within a single study and findings. youth smokers are less likely to none can fully isolate the There is preliminary evidence buy their own package, reducing contributions of each. These reali- to suggest that not all measures of the likelihood of exposure to ties underscore the importance of effectiveness decline at the same warning labels, compared to more the methodological features rate over time. “Proximal” mea- regular smokers who are more described in Section 2.1. In sures of salience, such as noticing likely to buy their own package addition, environmental scans of warnings, may erode more quickly (Leatherdale, 2005). As a result, other mass media campaigns and than “distal” measures, such as individuals who smoke more policy interventions can provide reporting that health warnings frequently are more likely to recall important context. motivate quitting and increase the content, location, and other thoughts about the health risks of aspects of labelling policies, a “Wear-out” and impact over time: smoking (Hammond et al ., counter-intuitive association at first 2007a). It is even plausible that for glance (Robinson & Killen, 1997). It is widely accepted that the some smokers the impact of A second issue concerns salience of advertising and health health warnings could increase longitudinal studies that use communications is typically grea- over time. For example, the measures of exposure or test upon first exposure (Bornstein, cessation and telephone quitline knowledge as predictors of future 1989; Henderson, 2000). The initial information included in many smoking behaviour among youth. impact of comprehensive labelling health warnings may only become During youth and young policies, such as the introduction of relevant to smokers as they adulthood, the rate of smoking large graphic warnings on contemplate quitting. In a popu- undergoes significant increases. packages, is often magnified by lation-based survey, however, the As youth smoking behaviour media coverage. As a result, ebb and flow among individuals increases, so too will their measures of effectiveness are likely will balance out, and one would exposure to the package and their to be strongly associated with the still anticipate decreases in mea- knowledge of the warnings. Thus, implementation date. This has sures of effectiveness over time. whereas a negative association
292 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 293
Measures to evaluate the effectiveness of tobacco product labelling policies
between exposure and future scores the importance of suitable health warnings are tailored to smoking behaviour may be research designs and appropriate particular sub-groups of smokers. expected for anti-smoking cam- interpretations of the data when Warnings on the risks of smoking paigns in other media, this is not evaluating warning labels among while pregnant, for example, have the case for warning labels. youth. little relevance for older males. Failure to account for the Thus, it is conceivable that some somewhat counter-intuitive asso- Evaluation of individual messages warnings may perform very well ciation between smoking and & content: among sub-groups who comprise exposure to health warnings can the target audience, but relatively result in misleading interpretations Beyond the question of whether poorly among the population as a of data. For example, one study health warnings are generally whole. As a consequence, survey characterized an association effective, there is a growing body of measures may need to be between increased smoking and research on the individual elements adapted and the findings may increased knowledge of health of a warning. These elements can need to be stratified among warnings as “paradoxical,” and broadly be categorized in terms of relevant sub-groups. One might also found evidence that US design and content components. expect the tailoring of warnings to health warnings were ineffective To date, much of the research has increase, as the use of picture- (Robinson & Killen, 1997). This focused on important design ele- based warnings increase, along may have been the case; how- ments including the size, position, with the typical number of rotating ever, without a comparison group, and use of pictures on the package warnings in a given jurisdiction. the authors had no way of (Strahan et al ., 2002). In contrast, In general, population-based knowing whether the increases in relatively few studies have exa- surveys may be most appropriate smoking behaviour were greater, mined the content of individual for identifying the overall ef- less, or no different than they messages. fectiveness of a set of health would have been if no warnings or Population-based surveys that warnings. However, the task of more comprehensive warnings compare labelling policies across evaluating the content of individual had been implemented. It may be, time or jurisdictions are somewhat warnings is best suited to for example, that fewer youth ill-suited to the task of evaluating experimental or qualitative designs, initiated smoking than would have individual warnings. Policies typi- in which the content and design otherwise occurred without the cally differ on more than one di- can be systematically varied. health warnings. In fact, this was mension, and policy changes ty- the pattern reported in a longi- pically involve increases in the size, Geographic & cultural differences: tudinal study comparing changes number, position, and type of in- in youth smoking in Canada and formation presented in each Very little research has examined the USA following the introduction warning. Evaluating individual com- potential geographic and cultural of graphic warning labels on ponents or messages becomes differences in the effectiveness of Canadian packages. Smoking more complicated as the number of health warnings. Although the rates and knowledge of the warnings and complexity of in- fundamental principles underlying warnings rose among Canadian formation increases; it is far easier the effectiveness of warnings are youth as they aged; however, the to evaluate the effectiveness of a unlikely to vary across cultures increase in smoking was sig- single warning through survey- and regions, the effectiveness of nificantly less than among US bas ed research than to evaluate the individual messages may indeed adolescents and the increase in content of 16 individual warnings. perform differently. First, smokers knowledge of the warnings con- When assessing the impact of in different parts of the world have siderably greater (Fong et al. , individual warnings, it is also different levels of existing health 2002). Overall, this study under- important to consider that many knowledge. This has implications
293 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 294
IARC Handbooks of Cancer Prevention
for the type of messages to be information being removed is used the “light” and “mild” descriptors. included in warnings. For exam- as a brand identifier. In the case of These considerations are impor- ple, Australian smokers may have bans on the use of “light” and tant in terms of how the data are a relatively higher level of health “mild,” the terminology that was interpreted and how the “effective- literacy than smokers in other previously used to identify a class ness” of light and mild policies is regions, which may account for of products no longer exists. conceptualized. the decision to include a warning Smokers may retain the same for “peripheral vascular disease” misleading perceptions of these Defining misleading descriptors: on packages. Picture-based products after the terms have warnings may be particularly been prohibited, but survey There is widespread confusion important in populations with lower measures can no longer refer to among both consumers and many literacy rates (CRÉATEC, 2003). “light” or “mild” cigarettes in the within the tobacco control In addition, the images used in same way as in the past. community regarding several key one jurisdiction may not be equally Therefore, survey measures must terms relevant to labelling policy. effective in another. For example, be designed so that the wording Many fail to make the distinction several of the picture-based and meaning of questions remains between “light” and “mild” and “low warnings that appear on constant before and after the tar.” Whereas “light” and “mild” are Venezuelan and Uruguayan removal of these terms. This terms used in the name of a packages, and elsewhere, use creative challenge is only now brand, strictly speaking “low tar” symbols that may be culturally being confronted by researchers refers to the emission levels under specific. Finally, similar sets of with the recent advent of “light” machine testing. Although there is warnings may be more effective in and “mild” prohibitions. One a very strong correlation between areas where smokers have approach, discussed later in this the two (manufactures often relatively little access to anti- section, is to make the res- attach “light” and “mild” des- smoking information from mass pondents’ “own brand” the criptors to brands that generate media or health professionals. referent for questions. lower tar levels under the ISO Few of these issues have been Another implication of the smoking machine), one can have addressed to date; however, they “removal” of brand information is a “light” cigarette that does not are likely to gain prominence as a that the beliefs associated with generate “low tar” levels and vice growing number of jurisdictions in “light” and “mild” cigarettes are versa. Strictly speaking, in Asia, Africa, and the Middle East likely to persist for some time after jurisdictions with bans, “light” and enhance their labelling policies to the descriptors disappear from “mild” cigarettes do not exist, meet Article 11, and must rely on packages. This situation is similar whereas “low tar” cigarettes do. an evidence base that derives to advertising, promotion, and To complicate matters further, the from relatively few Western and sponsorship bans; one should not terms “light” and “mild” can also be Latin American countries. expect beliefs to change imme- used to refer to sensory properties diately upon the implementation of of a cigarette. Thus, smokers may Evaluating the removal of infor - the policy, but more gradually over still retain the concept of a mation: time. Indeed, anecdotal evidence cigarette as “light” or “mild” even suggests that many retailers and in the absence of a brand Unlike other labelling policies, consumers continue to use the descriptor. Given the potential for restrictions on brand descriptors terms “light” and “mild” well after confusion, survey measures result in the removal, rather than their removal. Other packaging should be explicit about the the provision of, information. This elements and aspect of cigarette intended meaning of these terms presents a challenge to evalu- design may also reinforce the and should avoid using them ation, particularly when the same beliefs and perceptions as interchangeably. This becomes
294 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 295
Measures to evaluate the effectiveness of tobacco product labelling policies
apparent when measuring these model, as illustrated in Figure The extent to which information is concepts in jurisdictions where the 5.29. Other psychosocial varia- processed or elaborated upon has “light” and “mild” brand descriptors bles, such as social norms and been shown to be the most have been removed. beliefs about the tobacco industry, important determinant of memory could also be added to this model, and attitude change in response to Measures but have been excluded in the new information (Anderson, interest of brevity. The following 1990). A number of measures This section provides an overview sub-section begins with a review have been developed to assess of the key constructs and of quantitative measures, followed cognitive processing of health individual measures that have by qualitative measures, and a warnings as a general indicator of been used to assess labelling brief discussion of the role of their salience. These measures policies. The constructs range industry documents (Tables 5.27- range from more “shallow” from the extent to which labelling 5.39; see also Appendices 9 and measures of processing, such as policies are noticed and pro- 10). a general awareness of warnings, cessed, the extent to which they to “deeper” measures of pro- alter key beliefs (such as levels of Measures of labelling salience cessing, including reading the health knowledge), to their impact and processing: warnings and thinking about them upon downstream behavioural when they are not in sight outcomes. These measures can Health warnings must be cog- (Borland & Hill, 1997a; Canadian be organised within a conceptual nitively processed to be effective. Cancer Society, 2001; Hammond
Construct Noticing Health Warnings
Measure “In the last month, how often, if at all, have you noticed the warning labels on cigarette packs?” (Never, Rarely, Sometimes, Often, Very Often)
Sources Hammond et al ., 2006a; Hammond et al ., 2007a
Validity The time reference varies across different versions: some questions include no time reference (“How often do you notice…”), whereas others refer to the “last month” or “last three months.” The response categories also vary and are often collapsed into a smaller number of categories in analysis. The basic question can also be asked within the context of noticing other forms of anti-tobacco media (e.g. “In the last 6 months, have you noticed advertising or information that talks about the dangers of smoking, or encourages quitting in any of the following places? (Yes, No to a list of 9 media channels, including on cigarette packages )).
Comments Overall, a straightforward measure of the salience and processing of warnings that should be considered within the core set of variables to assess health warnings. As close to a “gold standard” in this domain as exists. Using the same wording to ask about salience of other media channels provides a useful comparative index for the salience of various health information channels. A recommended and essential measure for evaluating health warnings.
Table 5.27 Essential Measure of Labelling Salience and Processing
295 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 296
IARC Handbooks of Cancer Prevention
POLICY Health Warnings
Salience & processing POLICIY- SPECIFIC Awareness of warnings Knowledge of warnings MEDIATORS Noticing of warnings Depth of processing
Cessation Brand appeal Affective Health knowledge knowledge reactions GENERAL Attractiveness Perceived risk MEDIATORS Fear Beliefs about Self-efficacy Disgust Quit-line health effects information Thinking/worry Website about health risks information
Avoidance Quit Covering warnings intentions Requesting packs
Smoking behaviour
Changes in consumption Quit attempt OUTCOMES Abstinence
Figure 5.29 Conceptual framework for the evaluation of health warning policies
296 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 297
Measures to evaluate the effectiveness of tobacco product labelling policies
Construct (a) General Awareness
Measure “Have you seen health warnings on cigarette packages?” (Yes, No)
Sources Borland & Hill, 1997a; Health Canada, 2005
Validity Good face validity; associated with policy strength.
Variations Response categories are consistent across measures. Alternative wordings include: “Are you aware of health warnings on cigarette packages?” and “Are there health warnings on packages?” Some questions refer specifically to the release of new warnings. For instance, “Are you aware of any recent changes to health warnings on cigarette packs?” and “Have you noticed any changes to the health warnings on cigarette packages since [date]?”
Comments Provides an overall measure of general awareness. Limited value in examining changes and comparing across jurisdictions, given almost universal awareness among smokers. Most useful for examining policy implementation and rollout when the question makes reference to “new” warnings, or in jurisdictions with very weak health warnings and no previous research. Overall, an informative measure, but only recommended under these circumstances.
Construct (b) Reading/Looking Closely at Health Warnings
Measure “In the last month, how often, if at all, have you read or looked closely at the warning labels on cigarette packs?” (Never, Rarely, Sometimes, Often, Very Often)
Sources Hammond et al ., 2006a; Hammond et al ., 2007a
Validity Face validity; good convergent validity with other measures; good predictive validity for strength of policy and motivation to quit smoking.
Variations The time reference varies across different versions. Also, some versions refer to reading , other versions use broader language, such as looking closely, and some versions include both terms. Looking closely may be more appropriate for pictorial warnings.
Comments Strong correlation with noticing, but conceptualized as a “deeper” measure of processing. Overall, a recommended and important, but not essential, measure of salience and processing that may be particularly relevant for textual aspects of warnings.
Construct (c) Discussing the Health Warnings With Others
Measure “In the last month, how often, if at all, have you talked about the health warning with others?” (Never, Rarely, Sometimes, Often, Very Often)
Source Hammond et al ., 2003
Table 5.28 Additional Measures of Labelling Salience and Processing
297 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 298
IARC Handbooks of Cancer Prevention
Validity Good face validity, convergent validity, and predictive validity for motivation to quit and future smoking behaviour when included as part of a composite measure.
Variations Variations of this measures use slightly different terms, including discussed and mentioned rather than talked about , as well as different response options, such as Never, Rarely, Sometimes, Frequently.
Comments These measures provide a “deeper” measure of processing for labels and may be useful for comprehensive evaluations of labelling policies. Recommended, but not essential.
Construct (d) Thinking About Health Warnings
Measure “In the last month, how often have you thought about what the health warnings have to say?” (Never, Rarely, Sometimes, Often, All the time)
Sources Canadian Cancer Society, 2001; Hammond et al., 2003; Christie & Etter, 2004
Validity Good face validity, convergent validity, and predictive validity for motivation to quit and future smoking behaviour when included as part of a composite measure.
Variations “In the last month, have you ever thought about the warning labels or what they had to say when a cigarette pack wasn't in sight?” This variation of the measure requires a higher threshold of processing than the items above.
Comments These measures provide a “deeper” measure of processing for labels and may be useful for comprehensive evaluations of labelling policies. Recommended, but not essential.
Table 5.28 Additional measures of labelling salience and processing
et al. , 2003; Christie & Etter, 2004; the uncertainty regarding when “wear-out” (i.e. decrease in the Hammond et al ., 2004a; Health health warnings begin appearing salience of the warning labels) of Canada, 2005; Koval et al. , 2005; on packages. health warnings (Health Canada, Hammond et al., 2006a; Ham- In contrast to general mea- 2005; Hammond et al ., 2007a). mond et al. , 2007a). sures of awareness, the extent to Additional data of this type may Measures of general “aware- which smokers notice, read, and help to answer the question as to ness” are typically endorsed by a think about the warnings appears whether the rate of decline among vast majority of respondents, to be highly dependent on the measures of salience is asso- including non-smokers, regardless size, type, and location of the ciated with design features, such of the type of warning level. These warning (Borland & Hill, 1997a; as the size of warnings and the questions are often used to Health Canada, 2005; Hammond use of pictures. In most cases, examine the implementation, or et al. , 2007a). These measures of these measures have been “roll-out,” of new package war- processing are also subject to the analyzed as individual items, nings following a change in policy. implementation date. Several although in one case a depth of This information is critical for any studies have used measures of processing scale was developed population-based survey conduc- processing collected from the and tested (Hammond et al., ted shortly after the imple- same population over time and 2003). In that instance, nine items mentation of a new policy, given can be used to measure the were used to create a scale to
298 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 299
Measures to evaluate the effectiveness of tobacco product labelling policies
Construct Health Warnings - Eye Tracking
Measure Participants wore eye-tracking equipment and viewed US cigarette advertisements with health warnings.
Sources Fischer et al. ,1989b; Krugman et al. , 1994
Validity Good predictive validity for recall and recognition of health warnings
Variations Viewing time serves as another measure of attention, where warnings are flashed on a screen and the amount of time is recorded (Peters et al. , 2007).
Comments Eye tracking measures can help to identify the most salient design aspects of warning labels and serve as an objective measure of attention; however, these measures are limited to “laboratory” based research designs.
Table 5.29 Physiological Measures of Salience and Processing
measure cognitive processing wordings (Canadian Cancer include a number of rotating labelled as “depth of processing.” Society, 2001; Hammond et al ., health warnings, emission label- Responses to each of the nine 2004; Health Canada, 2005; Ham- ling is consistent across packages items were rated using a 5-point mond et al. , 2007a). for a given brand. As a result, Likert-type format going from "not there may be little reason for at all/never" to "all the time/a lot" Contents & emissions: smokers to read or attend to this and values added to create an information on a regular basis. As index. Examples of items included Several studies have assessed a consequence, we have not were “How carefully have you ever the extent to which smokers recommended a specific measure read the messages on the outside process emission information in this section. of a cigarette package?” and “How printed on the side of packages. often have you thought about what These measures mirror the pro- Physiological measures of messages on the inside of cessing items used to gauge salience and processing: packages have to say?” health warnings, although a more Although the wording of items limited set has been used. Both Physiological measures have is relatively similar across surveys, studies of which we are aware, been used in conjunction with different time periods are used in indicate that smokers are less survey measures to quantify both the question and the likely to read or look at emission attention to and processing of response option in many cases. information than health warnings health warnings. These mea- For example, whereas some on the face of packages sures have an advantage in that “noticing” questions refer to the (Thompson et al. , 2006). More they are more “objective” given past month, others refer to the generally, it is unclear whether that they do not rely on self- past three months, or use no time salience and processing type reporting. In several cases, they reference at all (Tables 5.27 and measures are as informative for have been used to compare the 5.28). Nevertheless, findings from emission labelling policies as for salience of warnings with pac- the same population are relatively health warning policies. Unlike kage design or within the context similar across different question health warnings, which typically of a tobacco advertisement. For
299 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 300
IARC Handbooks of Cancer Prevention
Construct (a) Health Warnings - Location
Measure Without looking at a cigarette package, where on the pack are the warnings or messages located?” (Open ended)
Source Hammond et al ., 2003
Validity Good face validity.
Variations The same question has been asked without the prefix (“Without looking at a cigarette package…”), as well as with a diagram in self-completed surveys.
Comments Useful measures for identifying basic knowledge about health warnings; however, it becomes complicated in jurisdictions with warnings on the inside and outside of packages. Unclear how emission and contents information should be treated, especially when provided by industry.
Construct (b) Health Warnings – Content
Measure “Without looking at a cigarette package, what specific health warning messages can you remember seeing on cigarette packages in Canada?” (Open ended)
Source Health Canada, 2005
Validity Good face validity.
Variations The same question has been asked without the prefix, which is typically included in telephone surveys to ensure the participant is not looking at the package during the call.
Comments Useful measures for identifying basic knowledge about health warnings and, potentially, for identifying individual messages that are particularly salient. However, this measure will be difficult to answer in jurisdictions with comprehensive health warnings, including multiple warnings on different areas of the package.
Table 5.30 Measures of Knowledge of Health Warnings
example, eye movements during communication measure (Krug- have been asked using un- exposure to an ad have been man et al. , 1994) (Table 5.29). prompted recall (e.g. “Where are used as physiological indicators the warnings located?”), as well as of attention to tobacco warnings. Knowledge of health warnings using recognition tasks (e.g. These measures that are directly “Please tell me which of the linked to cognitive processing Items assessing smokers’ know- following warnings appear on have been useful to investigate ledge of health warnings are cigarette packages…”) (Table the relationship between visual among commonly used survey 5.30) (Hill, 1988; Richards et al. , attention and a more traditional measures. Knowledge questions 1989; Rootman et al ., 1995;
300 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 301
Measures to evaluate the effectiveness of tobacco product labelling policies
Construct Emission Side Panel – Content
Measure “Without looking at a cigarette package, can you name any chemicals or substances that are currently listed on cigarette packages in [country]?” (Open ended)
Source Health Canada, 2003
Validity Face validity.
Variations A common alternative is to ask about the quantitative level of specific emissions, such as tar “Without looking at a pack, can you tell me the tar level of your cigarettes?”
Comments This measure examines basic recall of emission information printed on packages, and is often compared against “objective” data collected from other sources in order to evaluate accuracy of self-report recall. This measure should be interpreted alongside measures on the comprehension and use of this information (described later).
Table 5.31 Measures of Knowledge of Constituents and Emissions
Construct Health Warnings – Affective Reactions
Measure “Have you experienced any fear as a result of the health warnings?” (Not at all / A little / A lot)
Source Hammond et al. , 2004a
Validity Good face validity; good predictive validity for future smoking behaviour.
Variations Alternatives have used more comprehensive scales and asked about different affective reactions, including disgust and anger (Peters et al. , 2007).
Comments Affective reactions have been evaluated to a greater extent in qualitative evaluations of warning labels; however, survey-based measures may be a key mediator of downstream measures of impact.
Table 5.32 Measures of Affective Reactions to Health Warnings
Borland & Hill, 1997a; Borland & identify which individual warnings among nonsmokers. Except for Hill, 1997b; Robinson & Killen, may be most effective. In the few questions that refer to a 1997; Hammond et al ., 2003; jurisdictions with a large number respondent’s “own” cigarette pac- Brown et al. , 2005; Health Cana- of warnings, this task can be kage, most measures of da, 2005; O’Hegarty et al., 2006; particularly helpful. awareness and knowledge appear Thompson et al. , 2006). Measures Many of these measures have to work equally well among of unprompted recall for warning been assessed among the nonsmokers. Indeed, nonsmokers label content can be used to general population, including have been found to have
301 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 302
IARC Handbooks of Cancer Prevention
Construct Health Warnings – Avoidance
Measure “In the last month, have you made any effort to avoid looking at or thinking about the warning labels?” (Yes, No)
Sources Hammond et al ., 2004a; International Tobacco Control Policy Evaluation Survey (The ITC Project)
Validity Good face validity; good predictive validity for future smoking behaviour.
Variations Several follow-up questions may be asked of those who respond “yes” to the initial question, above. For example, “Have you made any effort to avoid the warnings by: (1) Covering the warnings up? (2) Keeping the pack out of sight? (3) Using a cigarette case or some other pack? (4) By not buying packs with particular labels?” (Yes, No to each question)
Comments These measures can indicate the prevalence of avoidance behaviours and whether they reduce the effectiveness of warnings. The follow-up questions are only necessary for in- depth exploration of avoidance.
Table 5.33 Measures of Avoidance
surprisingly high levels of aware- to be greater among heavier First, when asking about the ness and recall for prominent smokers. This association is likely location of health warnings, one health warnings and picture-based to be more pronounced within issue is whether respondents warnings in particular (Health samples that include a broad consider emission information, Canada, 2005). However, both range of smokers, and are likely to which may be printed on the sides recall and recognition of particular be greatest in studies that of the package, as a health messages has been shown to be compare regular smokers with warning. Canadian data suggests highly dependent on the com- occasional or nonsmokers. The that some smokers are aware of plexity of the health warning and its association between consumption this information, but fail to cite it as implementation date. For example, and knowledge is also likely to be a location. Second, in telephone virtually all Canadian smokers are stronger in jurisdictions with a or web-based surveys, some aware of the health warnings on greater number and complexity of participants may have a pack packages, although we are warnings. For example, packages visible as they respond to the unaware of any research indicating in Canada carry information on the survey. As a result, some that smokers have correctly been side panel, one of 16 health measures explicitly ask smokers able to identify all 16 health warnings on the outside of not to look at the package to avoid warnings that appear on packages . packages, and one of 16 this situation to the extent Analyses must take into additional warnings on the inside possible. Third, measures of account the consumption level of packages. In such cases, a knowledge can often be difficult to when assessing knowledge of greater number of exposures will compare across labelling policies. health warnings. Given the be required to recall various For example, smokers from the inevitable link between heaviness aspects of the warnings. USA, where a total of four different of smoking and viewing the There are several limitations text warnings appear on warning labels, knowledge is likely with measures of knowledge. packages, have a much greater
302 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 303
Measures to evaluate the effectiveness of tobacco product labelling policies
Construct (a) Health Warnings – Believability/Credibility
Measure “Overall, do you believe the health warning message(s)?” (Not at all, A little, A lot)
Source Health Canada Youth Smoking Survey (http://www.hc-sc.gc.ca/hl-vs/pubs/tobac-tabac/yss-etj-2002/index-eng.php)
Validity Good face validity.
Variations Other alternatives refer to the accuracy, trustworthiness, credibility, believability and true/false nature of the warnings or the importance of information (Cecil et al ., 1996; Borland & Hill, 1997a; Canadian Cancer Society, 2001; Hammond et al, 2004a; Brown et al. , 2005; Health Canada, 2005; O’Hegarty et al ., 2006). Some surveys have also included more comprehensive, but also more time consuming, scales involving numerous items.
Comments A useful, brief measure to examine credibility of message content. The measure can be used to examine whether different design and content features change the believability of information among smokers. This question can be asked of individual health messages, such as in qualitative or experimental research, or to refer to a set of warnings, as is common in population-based surveys. Note that responses to this item will also reflect denial, self- exempting beliefs, etc.
Construct (b) Health Warnings – Public Opinion/Support
Measure “Do you approve of the health warnings on cigarette packages?” (Yes, No)
Source Borland & Hill, 1997a
Validity Good face validity.
Variations Other alternatives include measures of agreement with the warnings and references to appropriateness or desire for more information (Canadian Cancer Society, 2001; Hammond et al ., 2004a; Brown et al ., 2005., Health Canada, 2005)
Comments This measure is a combination of previously administered questions and has yet to be administered exactly as worded. Though measures of public support or approval may be less important as a measure of effectiveness, they are a critical measure for regulators and policy makers, and for demonstrating support for more comprehensive policies.
Table 5.34 Measures of Credibility and Public Support
likelihood of correctly identifying when Canada revised its labelling cases, neither the total number all the messages than smokers in policy in 2000 to include pictures, nor the proportion of messages the United Kingdom where 16 the number of individual mes- correctly identified, provide a different text messages appear on sages doubled from eight to 16 suitable basis for comparing packages. The same issue arises (not counting 16 additional policies given that the in pre-post studies of a new messages that appeared on the denominator is different. More- labelling policy. For example, inside of packages). In such over, it is both time consuming
303 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 304
IARC Handbooks of Cancer Prevention
Construct Health Warnings – Thinking About Health Risks
Measure “To what extent, if at all, do the warning labels make you think about the health risks of smoking?” (Not at all, A little, A lot)
Source Hammond et al. , 2007a
Validity Good face validity; good convergent validity; associated with strength of policy.
Variations Similar questions ask about the extent to which warnings affect the level of concern or worry about health risks.
Comments A key mediator of the effectiveness of health warnings. This should be considered among the essential measures.
Table 5.35 Measures of Health Knowledge and Perceived Risk
Construct Emissions Information – Comprehension
Measure “If you were to look for a safer or less harmful cigarette, would you use information about the amounts of chemicals listed on the cigarette packs to help you find a less harmful brand?” (Yes, Maybe, No)
Sources Gori, 1990; Health Canada, 2003
Validity Good face validity.
Variations Similar questions ask smokers to compare different tar levels of cigarettes in terms of delivery and health risks.
Comments A critical measure to evaluate emission policies that include quantitative emission levels. The question can also be used to refer to specific emissions, such as tar or nicotine. This measure is essential in any survey that also asks about recall or awareness of emission numbers on packages. The current wording can be used to refer both to descriptive (i.e. text-based) and quantitative emission information.
Table 5.36 Measurement of Comprehension of Emissions Information
and awkward to prompt survey Krugman and Robinson presented visual information to be presented respondents for 16 different participants with diagrams of to participants, must be ad- warnings. various warnings in a recognition ministered either face-to-face or Finally, some knowledge mea- task (Krugman et al. , 1994; using web-based modalities. sures may not work across all Robinson & Killen, 1997). Any survey modalities. For example, such measures, which require
304 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 305
Measures to evaluate the effectiveness of tobacco product labelling policies
Construct (a) Light / Mild Descriptors – Comparative Risk
Measure “Light cigarettes are less harmful than regular cigarettes.” (Strongly agree, Agree, Neither agree nor disagree, Disagree, Strongly Disagree)
Source The ITC Project
Validity Good face validity; good convergent validity (Borland et al. , 2004).
Variations This question can be adapted to refer to other descriptors, such as mild or smooth . In some cases, the terms light and mild are used in the same question.
Alternatives ask smokers about differences in the “tar” or “nicotine” of light versus regular cigarettes (Smokers of light cigarettes take in less tar than smokers of regular cigarettes). These measures have been widely used, but require a basic familiarity with tar and nicotine, which may not exist in all smokers in some jurisdictions (Kozlowski et al ., 1998b; Shiffman et al. , 2001; Hamilton et al ., 2004).
Other alternatives have asked smokers to report the number of light cigarettes that would need to be smoked to equal the harm from 10 regular cigarettes; however, this approach requires a level of numerical literacy beyond the capacity of smokers in many jurisdictions (Kozlowski et al ., 2000; Shiffman et al., 2001).
Comments: This is an essential construct, although there is no single “gold standard” question for its measurement. The recommended measure has been selected because is it the most direct and may be most appropriate for smokers in low- and middle-income countries. Nevertheless, the question may need to be preceded by a general awareness questions (e.g. “Have you ever heard of light cigarettes?”) in some markets or rural areas. There are also issues with the interpretation of this measure in jurisdictions where light and mild descriptors have been prohibited.
Construct (b) Light/Mild Descriptors – Addiction
Measure “Light cigarettes are less addictive than regular cigarettes.” (Strongly agree, Agree, Neither agree nor disagree, Disagree, Strongly Disagree)
Source The ITC Project
Validity Good face validity; good convergent validity (Borland et al. , 2004).
Variations This question can be adapted to refer to other descriptors, such as mild or smooth. In some cases, the terms light and mild are used in the same question.
Comments A straightforward question with the same format and response options as above. A recommended question to address perceptions of light and mild cigarettes, but not as essential as the comparative risk question, above.
Table 5.37 Measures of Light, Mild, and Brand Descriptors
305 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 306
IARC Handbooks of Cancer Prevention
Construct Brand Appeal – Health Warnings
Measure “Do you think the new warnings make cigarettes packages look less attractive, more attractive, or has it made no difference to their attractiveness?” (Not at all, A little, A lot) “How often have you put your cigarette package away because you didn’t want others to see the warning on the package? Have you done this?” (Never, Sometimes, Often)
Source Canadian Cancer Society, 2001
Validity Face validity.
Variations Alternatives refer to quality of advertisements with and without warnings, whether youth would want to “use” the product, intentions to purchase the product in the future, and a measure of perceived economic values of brands (Hyland & Birrell, 1979; Brubaker & Mitby, 1990; Canadian Cancer Society, 2001; Willemsen et al. , 2002; Thrasher et al ., 2007). “Attractiveness” scales have also been used (Loken & Howard-Pitney, 1988).
Comments These measures provide a straightforward evaluation of whether health warnings have altered the general appeal of packaging. The second of the two measures has a higher threshold and represents a more distal measure of appeal, which may also tap into social norms. Both of the measures are recommended for surveys that wish to provide a comprehensive evaluation of warnings, but are not essential.
Table 5.38 Measures of Brand Appeal
Constituents & emissions: nicotine numbers may be printed Strong emotional responses to on the package, but few are able messages are also associated A number of studies have to recall the tar or nicotine levels with greater behaviour change examined whether smokers can printed on their usual brand of when supportive or “efficacy” recall the emission information cigarettes. To our knowledge, no related information is also pre- commonly printed on the side measures have been developed sented. To date, several studies panel of cigarette packages to measure smokers’ knowledge have used measures of affective (Table 5.31) (Chapman, 1986; of tobacco contents. reactions to assess the impact of Cohen, 1996b; Health Canada, warnings labels (Environics Re- 2003; O'Connor et al ., 2006c). search Group, 2000; Elliot & These items typically ask Affective reactions to health Shanahan Research, 2002; participants to name the emi- warnings: Environics Research Group, 2003; ssions printed on packages using Hammond et al., 2004a; Health unprompted recall tasks, or ask Research in the field of health Canada, 2006; Peters et al ., 2007). them to report the number communication indicates that These measures are common in associated with a particular messages with emotionally qualitative evaluations of individual emission (usually “tar”). The data arousing content are more likely to warning labels and have been indicates that many smokers have be noticed and processed by particularly influential in develop- a general awareness that tar and smokers (Witte & Allen, 2000). ment of picture-based warnings in
306 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 307
Measures to evaluate the effectiveness of tobacco product labelling policies
Construct (a) Changes in Foregoing – Health Warnings
Measure “In the last month, have the warning labels stopped you from having a cigarette when you were about to smoke one?” (Never, Once, A few times, Many times)
Sources Borland & Hill, 1997a; Hammond et al. , 2007a
Validity Good face validity; convergent validity; associated with strength of policy.
Variations Similar measures have referred to not smoking when tempted.
Comments This question has a lower “threshold” than other measures that assess the behavioural effects of health warnings.
Construct (b) Reductions in Smoking – Health Warnings
Measure “Are you smoking any less or more as a result of the new warnings , or are you still smoking the same amount?” (Less, Same amount, No difference)
Source Hammond et al. , 2007a
Validity Good face validity; convergent validity; associated with strength of policy.
Variations Similar measures have referred to not smoking when tempted.
Comments The wording “as a result of the warnings” needs to be emphasized when asking this question. This item is not intended to provide a precise measure of changes in consumption as a result of the warnings; changes in consumption happen in response to a wide range of related factors. However, this question does provide a good general measure of the extent to which smokers have been affected by the warnings.
Construct (c) Likelihood/Motivations to Quit
Measure “To what extent, if at all, do the warning labels on cigarette packs make you more likely to quit smoking?” (Not at all, A little, A lot)
Source Hammond et al ., 2007a
Validity Good face validity; convergent validity.
Variations Alternatives refer to motivations to quit and thinking about quitting , with some differences between response categories.
Comments The recommended wording refers directly to the likelihood of quitting smoking, which is somewhat broader than motivation alone. In practice, however, there appears to be few differences with regards to how these measures perform in practice given the consistency of findings from similar samples. The question has the potential to provide a very good summary measure of the self-reported impact of health warnings and should be considered within the core set of items to evaluate labelling policy.
Table 5.39 Measures of Behavioural Outcomes
307 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 308
IARC Handbooks of Cancer Prevention
Construct (d) Quit Attempts & Abstinence
Measure “To what extent, if at all, were the following reasons for your current quit attempt…warning labels on cigarette packages?” (Not at all, A little, A lot)
Source The ITC Project
Validity Good face and convergent validity.
Variations Alternatives have also asked about the effect of the warnings on staying quit in the future. This question can be asked as part of a list of reasons for quitting, which provides some useful context on the relative influence of other potential influences on quitting.
Comments Retrospective measures, such as this, should be interpreted with caution given that they are subject to recall biases, particularly as the time since the quit date increases. In addition, smokers often cite a number of complementary reasons for quitting and endorsement of this item does not mean that the quit attempt is solely attributable to health warnings.
Table 5.39 Measures of Behavioural Outcomes
several jurisdictions. Measures of manufacturers have been ac- communicating the threatening negative emotions, including fear cused of marketing covers consequences of smoking there and disgust, have also been used specifically intended to cover pic- would be no reason to avoid them. in population-based surveys and ture- based warnings, prompting Furthermore, one study found that shown to predict future cessation- calls for regulatory bans on the smokers who attempted to avoid related behaviour (Table 5.32). sale of such covers (Table 5.33) the warnings were no less likely to Overall, measures of emotion have (Wilson et al., 2006). see the warnings, think about considerable promise as a proximal Although avoidance behaviours them, or engage in cessation measure of effectiveness which may be undesirable to some behaviour at a 3-month follow-up can be used in both qualitative and extent, these examples of fear (Hammond et al. , 2004a). quantitative research. control behaviour do not neces- sarily reflect an adverse outcome Credibility & public support: Avoidance: or inherent weakness of package warnings. Research has demon- In order to be effective, the health Warnings that result in unpleasant strated that avoidant behaviours information presented in warnings emotions may lead some smokers and attempts at thought sup- must be credible. The credibility of to avoid the warnings. Indeed, pression often have the opposite warnings relates not only to the several studies indicate that a effect of increasing the presence health information contained in a considerable portion of smokers of the unwanted thoughts warning, but also to its design and make some attempt to avoid the (Wegner, 1994). In the context of source or attribution. Some have warnings, including covering or the warning labels, avoidant even speculated that there may be hiding the warnings, using another behaviour might be more rea- a trade-off between the vividness case, or requesting different packs sonably interpreted as a measure of the information in health to avoid particular warnings. In of effectiveness. Indeed, if the warnings and its credibility among some jurisdictions, tobacco warnings were ineffective in smokers. In others words, if
308 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 309
Measures to evaluate the effectiveness of tobacco product labelling policies
pictures and text become too and “Do you approve of the health lung cancer and heart disease, as striking or graphic, smokers may warnings on cigarette packages?”) well as health effects on begin to question the accuracy of (Health Canada, 2001; Health nonsmokers, and lesser-known the information and become more Canada, 2003). Public opinion health effects. Including lesser- resistant to the messages. data may be particularly effective known health effects can be Although some validated for policy makers in gauging particularly effective in attributing scales have been used to political support for new or existing changes in knowledge to specific evaluate the believability of health labelling polices (Table 5.34). labelling policies. Ideally, longi- warnings (e.g. Beltramini, 1988; tudinal studies, assessing Loken & Howard-Pitney, 1988; Health knowledge & perceived changes in health knowledge, Cecil et al. , 1996), many studies risk would also select the health have used single questions with effects based upon the effects that face validity (Borland & Hill, The primary objective of cigarette are targeted in the warnings. In 1997a; Canadian Cancer Society, warning labels is to communicate other words, studies should 2001; Hammond et al ., 2004a; the health effects from smoking. include health effects that: a) are Brown et al., 2005; Health Thus, measures of health know- already included on packages at Canada, 2005; O’Hegarty et al., ledge and perceived risk baseline (before policy change) 2006; Peters et al. , 2007). represent critical components in and will remain on packages at Together, the findings suggest any evaluation of health warnings follow-up; b) health effects that are that health warnings represent a (Table 5.35). To date, studies not on packages at baseline, but credible source of information, have taken two main approaches will appear at follow-up; and c) particularly when attributed to a to measuring the impact of health effects that are not on well-respected department of warnings on health knowledge. packages at either baseline or health, or a well-respected non- One approach is to ask par- follow-up. This type of design governmental authority, such as a ticipants to self-report whether provides a measure of specificity cancer society (Guttman & Peleg, health warnings have changed the with respect to changes in 2003; Health Canada, 2003; BRC extent or frequency with which labelling policies. Marketing & Social Research, they think or worry about the A similar approach has been 2004). The levels of credibility do health effects of smoking. Alter- taken with respect to emission not appear to be associated with natively, some studies have information. At least one study has the type or design of warning assessed health knowledge examined whether knowledge of labels; just like for text-based directly and examined changes the emissions in tobacco smoke is warnings, smokers report high over time or across jurisdictions in higher in jurisdictions where they levels of believability for graphic levels of knowledge. Given the are printed on the package picture-based warnings as well. number of health effects caused (Hammond et al ., 2006a). As with Several studies have also by smoking, we are unaware of health effects, lists should include sought to assess general mea- any study that has attempted to emissions that are, and are not, sures of public support for health measure a complete list. How- printed on packages, in order to warnings (Borland & Hill, 1997b; ever, studies typically measure examine the specificity of the effect. Brown et al. , 2005; Hammond et beliefs about a range of specific Overall, research conducted to al., 2004a; O’Hegarty et al ., 2006). health effects to determine date suggests that increases in To our knowledge, two items have knowledge levels. Some studies the size, number, and content of been developed to examine sup- have included “bogus” health warnings are associated with port among smokers for emission effects in the list in order to identify greater thoughts about the health labelling (“Overall, do you believe response bias. Most lists include risks of smoking (Health Canada, the health warning message(s)?” “major” health effects, such as 2005; Hammond et al ., 2007a).
309 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 310
IARC Handbooks of Cancer Prevention
More prominent warnings have meaning of the emission levels, ultra-light” and “regular” brands also been associated with although a substantial proportion using qualitative or descriptive increased knowledge for specific associate health benefits with categories to describe exposure health effects (Borland & lower numbers. This type of data levels and health risks. These Hill,1997a; Hammond, 2006a). is critical to place measures of qualitative response categories Most of these findings derive from knowledge into context; prominent have also been used to compare population-based surveys, al- labelling that succeeds in in- perceived sensory properties and though one study reported creasing knowledge of emission addiction levels of “light/mild” significantly higher beliefs about levels is of little value if smokers cigarettes compared to “regular” health effects following presen- do not understand the meaning of brands. At least one study com- tation of graphic versus text these numbers. Indeed, the data bined items to create a “sensory” warnings within an experimental appear to indicate that com- index and a “health effects” index setting (O’Hegarty et al. , 2006). municating quantitative emission (Shiffman et al. , 2001). Overall, levels promotes erroneous per- both qualitative and quantitative Constituents & emissions: ceptions about exposure levels measures appear to yield similar and health risks that can be findings, and indicate that a A number of studies have sought expected from different products. substantial proportion of smokers to examine the extent to which In general, this set of findings perceive health benefits from smokers understand and interpret underscores the importance of cigarettes with “light” and “mild” quantitative cigarette emission assessing more than basic recall descriptors (Table 5.37). information (Table 5.36). These of information (Figure 5.30). At least one study, the Inter- studies ask smokers to report national Tobacco Control Policy either the “meaning” of the Light & mild descriptors: Evaluation Survey (the ITC Pro- numbers, or the extent to which ject) (Borland et al ., 2004), has the numbers translate into dif- A variety of surveys have adopted an alternative approach ferences in exposure from examined perceptions of “light” to comparative estimates. Rather different brands (Gori, 1990; and “mild” brand descriptors than asking smokers to compare Cohen, 1996a; Health Canada, (Kozlowski et al. , 1998b; Kozlow- “regular” and “light” cigarettes, 2003; Thompson et al ., 2006). ski et al ., 2000; Ashley et al. , 2001; participants were asked to Other questions ask smokers to Shiffman et al ., 2001; Etter et al. , compare their “usual” brand with predict the health consequences 2003c; Borland et al. , 2004; regular cigarettes (e.g. “Do you of different tar levels, without Hamilton et al. , 2004). Both think that the brand you usually explicit reference to labelling poli- quantitative and descriptive mea- smoke, [current brand], might be a cies (Gori, 1990; Cohen, 1996a). sures have been used to assess little less harmful, no different, or Indeed, a number of studies on this the health consequences of a little more harmful, compared to topic were conducted in the USA, smoking “light/mild” cigarettes. other cigarette brands?”). where there are no mandatory Several studies have asked Separate items were used to requirements to print emission smokers how many light or ultra- collect the name, descriptors, and levels on packages, they appear on light cigarettes would need to be relevant attributes of participants’ packages less than 15% of the time, smoked to inhale the equivalent “usual” brand. This approach has and are at the discretion of the level of tar as regular cigarettes. the benefit of personalizing the manufacturer (Davis et al., 1990). Some of these measures used “10 question, and is particularly useful Regardless of the jurisdiction cigarettes” as a reference point, to implement following the or the labelling policy, the findings whereas others were open-ended. removal of “light” and “mild” terms, indicate that smokers have very Smokers have also been asked to at which point questions with little or no understanding of the make comparisons between “light/- direct reference to “light” and
310 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 311
Measures to evaluate the effectiveness of tobacco product labelling policies
POLICY
Emission information
Salience & processing Indirect effects POLICY-SPECIFIC Industry Awareness of MEDIATORS response information Knowledge of information Moderators • Public education about rationale and meaning • Variant smoked Beliefs about GENERAL contents and MEDIATORS emissions
General percieived risk OUTCOMES Perceptions of relative risk between products Quit intentions
Brand switching Smoking behaviour
Changes in consumption Quit attempt Abstinence
Figure 5.30 Conceptual Framework for the Evaluation of Emissions and Contents Labelling Policies
311 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 312
IARC Handbooks of Cancer Prevention
“mild” cigarettes become awkward dictions that have banned “light” (Hyland & Birrell, 1979; Loken & and confusing. The question also and “mild,” may wish to consider Howard-Pitney, 1988; Brubaker & has a broader frame of reference. additional measures that examine Mitby, 1990; Hammond et al. , This is an advantage in the sense the substitution of terms in their 2004b; Thrasher et al ., 2007). that it captures the effect of other place. Market-based research, Future research might also potential misleading descriptors or such as cataloguing the infor- explore whether larger graphic product elements. The dis- mation printed on packages, can health warnings undermine the advantage is that information on provide “objective” data on the visual appeal of cigarette displays the respondents’ own brand must substitution of terms which may be at retail outlets. also be available (see Section helpful in interpreting self-reported 5.4), and there is less specificity brand data (see Section 5.4). Behavioural outcomes with respect to the brand elements Largely, the selection of that underlie differences in measures in this area may depend There are several approaches to perceptions of risk. A similar upon the current state of policy predicting “downstream” cessa- conceptual approach has recently more so than other areas (Figure tion-related outcomes from health been taken with respect to 5.31). models. As with health effects, evaluating print advertisements. some studies have used mea- Rather than asking smokers to Brand appeal sures of processing and compare the risks implied by the knowledge of the warnings, and expressions “light” versus “regu- Health warnings target psycho- modelled their effects on moti- lar” cigarettes, respondents were social variables other than vation to quit and patterns of asked to rate the perceived risk to perceived risk and health know- smoking behaviour (see Section their health derived from adver- ledge. More recent labelling 3.1 for measures of tobacco use tisements for different products, policies include themes of addic- and Section 3.2 for psychosocial and the ratings for advertisements tion, industry manipulation, outcomes). This has produced of “light” versus “regular” ciga- aesthetic costs, financial costs, significant findings in longitudinal rettes were compared (Hamilton and cessation beliefs, among studies to date (Hammond et al., et al ., 2004). In most cases, others. A range of psychosocial 2003). However, this approach is follow-up questions may be measures have been developed somewhat limited when it comes necessary to identify which to assess each of these con- to evaluating changes in health specific elements underlie per- structs, although these measures warnings. Unless both survey ceptions of reduced harm. have rarely been used to evaluate waves are conducted when the Descriptors other than “light/- warning labels. same set of health warnings is on mild” are likely to receive One area that has been ex- the pack, the baseline measures increased attention in the coming plored is the impact of health of processing or knowledge relate years, particularly within juris- warnings on measures of brand to the “old” warnings, whereas any dictions where “light/mild” terms appeal (Table 5.38). In theory, cessation-related activity at follow- have already been prohibited. To replacing brand imagery with up presumably reflects the impact our knowledge, only one study health warnings has the potential of the “new” warnings. has developed measures to to change perceptions of the An alternate strategy that can evaluate health perceptions based cigarettes and packaging. To also be used in cross-sectional on other brand descriptors, date, the limited findings in this studies is to ask smokers to including the words “smooth” and area appear to support this directly report the extent to which “ultra” (Thompson et al. , 2006). hypothesis, although it has yet to warnings have influenced their Furthermore, studies with a focus be explored in much depth with motivation to quit and smoking upon brand descriptors in juris- respect to warnings on packages behaviour (Borland & Hill, 1997a;
312 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 313
Measures to evaluate the effectiveness of tobacco product labelling policies
POLICY
Removal of descriptors such as Light and Mild
Incidental effects Salience & POLICY-SPECIFIC processing MEDIATORS Industry use of alternative Awareness of descriptors descriptors Moderators
Beliefs about • Information provided descriptors/brands by health groups Addiction and government on Health risks rationale for removal
Taste • Brands smoked Sensory properties GENERAL MEDIATORS
General perceived risk
Perceptions of relative risk between products Quit intentions OUTCOMES
Smoking behaviour
Changes in consumption Quit attempt Brand switching Abstinence
Figure 5.31 Conceptual Framework for the Evaluation of Brand Descriptor Policies
313 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 314
IARC Handbooks of Cancer Prevention
Canadian Cancer Society, 2001; cluded in package health not possible with telephone based Health Canada, 2005; Koval et al ., warnings. Each of these studies surveys; and 2) focus groups are 2005; Willemsen, 2005; O’Hegarty reports significant increases in call well suited to open-ended ques- et al. , 2006). This approach does volumes (Willemsen et al. , 2002; tions and allow for more in-depth not have the same validity in terms Department of Health, 2006). discussion than structured sur- of measuring actual changes in Finally, several items have veys. In many cases, focus groups smoking behaviour, although it been created for use among are also used as a way to evaluate can be used to examine changes former-smokers. Typically, these the effectiveness of health across labelling policies. items ask about various reasons warnings on sub-groups, including A third alternative is to examine for quitting, including whether the younger smokers and those from changes in prevalence rates, or health warnings either motivated lower socio-economic groups. The population-based cessation acti- them to quit or have helped them primary disadvantage of focus vity, before and after the imple- to remain abstinent (Canadian groups is that the findings can be mentation of new warnings. To our Cancer Society, 2001; Hammond hard to summarize in a systematic knowledge, this approach has et al. , 2004b; O’Hegarty et al. , fashion, which complicates com- been used in only one study to 2006; Thompson et al ., 2006). parisons across groups and date: Gospodinov & Irvine (2004) These measures are, however, settings. As a result, conventional reported no discernable changes in subject to recall bias and should validity tests for quantitative data prevalence rates, and a reduction be interpreted with particular can not be conducted with focus of two cigarettes per week among caution (Table 5.39). group findings. Nevertheless, smokers in the months following qualitative findings help to com- the implementation of pictorial Formative research plement quantitative research in health warnings in Canada. How- this area, and represent an ever, as described earlier in this Formative research is often important step in the development section, there are serious problems conducted to help identify the of new labelling policies. in attributing changes in national content and design of new health Qualitative research has level trends to changes in health warning policies. Regulators must examined many of the same warnings, or any other individual decide what health effects to themes as population-based sur- policy measure. Indeed, as communicate, how many, and veys, and other quantitative Gospodinov & Irvine note, there how to present this information to methods. These include general were significant changes in price smokers on the package. Al- knowledge of the warnings, such over the same period of time, as though population-based surveys as the content and location, the well as considerable sub-national may help to guide these decisions, emotional impact of warnings, as tobacco control activity over the qualitative research is typically well as their general salience and same time period. undertaken as part of the policy noticability (Environics Research Yet another approach to mea- development process. Group, 2000; Elliot & Shanahan suring the impact of warnings on The most common approach Research, 2002; CRÉATEC, 2003; cessation behaviour has been to has been to conduct a series of BRC Marketing & Social Research, look at changes in the use of focus groups (i.e. semi-structured 2004; Health Canada, 2006). In cessation services as they relate interviews conducted within a many cases, these studies have to information on warnings labels. group setting). Focus groups have presented different health warnings Research conducted in the UK two important advantages over to participants in order to make and the Netherlands has exa- population-based surveys: 1) direct comparisons between mined changes in the usage of participants can be presented with labelling policies. These designs national telephone helplines after visual stimuli, including examples have proven particularly effective at the contact information was in- of health warnings in a way that is comparing the emotional reactions
314 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 315
Measures to evaluate the effectiveness of tobacco product labelling policies
elicited by picture versus text Alechnowicz & Chapman, 2004). Implications for study design & warnings, for example (Environics However, to date, no com- analysis: Research Group, 2000; Elliot & prehensive review of packaging Shanahan Research, 2002; BRC issues related to labelling policies No single study research design is Marketing & Social Research, has been undertaken. adequate to evaluate the impact of 2004). Focus groups have also labelling policies. Given the chal- provided critical information Summary lenges inherent in evaluating regarding the meaning and national level policies, individual comprehension of the information Few of the measures used to studies are inevitably subject to a communicated in labelling policies. evaluate warning label policies range of limitations. However, For example, focus group have undergone formal psycho- when taken collectively, the range measures developed, on behalf of metric analyses. Much of the of designs constitute a persuasive Health Canada, have helped to literature in this area has been body of evidence demonstrating demonstrate that, even though conducted on behalf of regulators, the effectiveness of compre- most Canadian smokers are aware which may account for the lack of hensive health warnings. Quali- of emission information on the side “formal” tests of validation more tative methods, including focus of packages, very few understand common to academic research. In groups, are essential for informing the actual meaning of the infor- addition, different studies have the early stages of design and mation (Environics Research used different measures to assess generating new insights into Group, 2003). Indeed, judging by the same construct. In many labelling policies. Experimental the findings of the focus group, cases, measures differ in the research is best suited to drawing most Canadian smokers are wording of questions and in the direct comparisons across misusing the emission infor- time references used in mea- warnings and to isolating the mation. Thus, carefully con- sures, such as noticing and effectiveness of individual design structed focus group measures can awareness. This complicates and content features. For this provide “deeper,” more com- comparisons across surveys and reason, experimental research prehensive measures of meaning across labelling policies. However, provides the highest level of that are difficult to ascertain most measures have high face internal validity. Alternatively, through structured population- validity and several have shown population-based surveys have based surveys. good predictive validity for down- the highest external validity and stream outcomes, including may provide the most com- Industry documents knowledge of health effects and prehensive measures of effec- self-reported motivation to quit, tiveness given adequate designs. Internal tobacco industry docu- and cessation behaviours. In External validity is particularly ments represent a potentially rich addition, the consistency of the important in the case of warning source of information about the findings across studies and survey labels, which operate over effectiveness of tobacco control modalities suggests that the repeated exposures that are tied policies. There are several infor- differences in the measures have to smoking behaviour. The pattern mative reviews of industry only a modest effect on outcomes of exposure is the defining feature activities and documents on pro- of interest. Nevertheless, virtually of product warnings and one that duct labelling, including many all of the constructs would benefit is impossible to replicate in a related to brand descriptors such from further developmental work, “laboratory” environment. As a as “light” and “mild” (Slade, 1997; including the standardization of result, the central question of Pollay, 2001; Pollay & Dewhirst, the wordings across surveys. whether labelling policies in- 2002; Wakefield et al ., 2002; fluence beliefs, attitudes, and Chapman & Carter, 2003; behavioural change can only be
315 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 316
IARC Handbooks of Cancer Prevention
assessed with population-based using descriptive, rather than and efficacy-related information. surveys. The inferences that can quantitative means. Greater These measures may include be made from these surveys are experimental and qualitative work survey based measures, as well considerably enhanced within must also be undertaken to as indicators from other data longitudinal and quasi-experi- explore how smokers interpret sources, such as usage rates from mental designs, as discussed in and use this information, and to telephone quitlines or web-based Section 2.1. compare different approaches services. more systematically. These issues Priorities for future work: are directly relevant to the ongoing Recommendations debate regarding how to As countries begin to implement communicate the risks of com- Comprehensive evaluations of restrictions on misleading pac- bustible versus non-combustible health warning labels should kaging elements, research must tobacco products. Historically, include recommended items from begin to examine elements other emission information has been each of the key constructs (see than “light” and “mild” brand des- used by smokers to evaluate the above). Population-based sur- criptors. These include other relative risks of different products. veys, seeking a more limited potentially misleading elements, As emission and content labelling evaluation of health warnings, such as the use of colour-coding policies are developed for the full should include proximal measures and package designs that falsely range of tobacco products, of noticing, along with inter- convey differences in strength. To regulators will need to consider mediate measures of perceived date, very limited work has been the delicate issue of what risk or health knowledge. Although conducted outside the tobacco fundamental message they wish measures of general awareness industry on these issues. There is to communicate to smokers. and knowledge of health warnings an immediate need to develop Quantitative emission and content can be informative, these mea- measures that can examine these information will inevitably be sures should be used with caution issues within population-based interpreted as indicators of risks, for the purpose of comparing samples, especially within juris- unlike descriptive information that labelling policies. dictions where “light” and “mild” is uniform across products. Evaluations of brand des- descriptors have already been In addition to developing new criptors, and other packaging prohibited. survey measures, existing mea- elements, should represent a A second priority for future sures must be administered more priority for tobacco control policy. In research is to examine contents widely, as a greater number of addition to examining “light” and and emission information more countries prepare to implement “mild” descriptors, research should closely. Up to now, much of the the provisions within Article 11 of consider other potentially mis- existing research has focussed the FCTC. In particular, few of the leading terms, as well as brand upon awareness and under- measures reviewed in this section elements such as colour and standing of ISO tar and nicotine have been assessed among package design. Unlike health numbers. There is an urgent need smokers in low- and middle- warnings, these policies require the for measures to evaluate new income countries. removal of information from the approaches to communicating Finally, measures should be package and present challenges in contents and emission infor- developed to examine the impact the wording of survey measures. mation. Population-based studies of the cessation information that is There is an immediate need to should be conducted within included in many labelling policies. develop measures that can jurisdictions that have developed Cigarette packages are among address these issues as more novel policies, such as com- the most prominent vehicles for countries implement recomm- municating emission information disseminating cessation services endations under Article 11 to
316 section5.5janvier13:Layout 1 13/01/2009 10:07 Page 317
Measures to evaluate the effectiveness of tobacco product labelling policies
prohibit misleading package ele- value. Rather, evaluations should Overall, the selection of mea- ments. focus upon the meaning and use sures to evaluate tobacco labelling Policies to communicate emis- of emission and content infor- policies will depend upon the sions and content information via mation. Given the lack of research method and scope of the evalu- packages, also present unique in this area, and the lack of ation, as well as the specific policy evaluation challenges. Unlike consensus regarding the best context. health warnings, measures of policy approach, there is a par- salience and processing for this ticular need for formative research type of information are of limited in this area.
317