Methods
A multi-method approach was used to develop and test warning statements designed to advise drinkers of the cancer risk associated with alcohol consumption. Ethics clearance for the study was obtained from the University of Western Australia Human Research Ethics Committee and the research reporting process adhered to the RATS guidelines for reporting qualitative studies. Initially, focus groups were conducted with current drinkers to (i) explore perceptions of the harm associated with various levels of alcohol intake, (ii) assess awareness of the link between cancer and consumption, and (iii) workshop message content. On the basis of the qualitative findings, 12 warning statements were developed and then tested in a subsequent quantitative phase to determine the extent to which they may be effective in warning drinkers of the cancer risks associated with alcohol consumption. These two data collection stages are described further below.
Qualitative Phase
A social research agency used their respondent databases and random digit dialing to recruit drinkers to participate in the qualitative phase of the study. Forty-eight individuals who consumed at least 2–3 standard drinks of alcohol per month participated in six focus groups that were stratified by gender and age (18–30, 31–45, 46–64 years). The first author moderated all the groups using an exploratory and emergent approach. The groups commenced with a general discussion of the role of alcohol in Australian culture and in the participants' lives. Motivations for and consequences of drinking were discussed, including any harms/diseases that participants considered to be associated with alcohol consumption. Perceptions of the potential for warning statements to be printed on alcoholic beverage containers were canvassed, and participants were invited to suggest wording for warning statements that could be effective in encouraging drinkers to reduce their intake. During this process, mock statements were constructed in line with the participants' comments and used to stimulate further discussion. The statements generated during the focus groups included examples of wording that varied according to message framing (negative vs positive frame), strength of suggested causality ('alcohol causes/can cause cancer' vs 'alcohol increases the risk of cancer'), types of cancer (general reference to cancer vs mentions of specific forms of cancer), and the use of the term 'Warning' (or not).
The groups ran from 60 to 92 minutes and were digitally audio-recorded. The recordings were transcribed and imported into NVivo9 software (QSR International, Pty Ltd) for coding and analysis. The coding schema included demographic variables and content nodes relating to the range of topics discussed in the groups. Examples of the latter included health-related beliefs about alcohol, types of drinking behaviors enacted, attitudes to cancer and its causes, types of message attributes mentioned, and perceptions of appropriate organizations/institutions to deliver cancer-related messages.
Throughout the focus group discussions, participants were primarily focused on the short-term effects of over-consumption and they frequently referred to their increased likelihood of engaging in other risky behaviors such as smoking and drink driving while intoxicated. There was also a clear understanding of the dangers associated with drinking while pregnant. The younger participants were additionally concerned about becoming over-emotional and being involved in accidents, acts of violence, and unprotected sex. Longer-term outcomes were less salient, but diseases such as liver problems, heart disease, cancer, and obesity were raised in most groups as potential outcomes from excessive alcohol consumption.
When the focus group discussions were directed to focus specifically on cancer risk, many participants appeared to believe that "everything gives you cancer", and that alcohol consumed in moderation does not constitute a level of cancer risk that is worthy of concern. They therefore felt that it would be difficult to convince people that they needed to change their drinking habits on the basis of cancer risk. There was considerable discussion around the ideal wording for cancer warning statements that could be included on alcoholic beverage containers. Many felt that it would be more appropriate to refer to an increased risk of cancer (e.g., alcohol increases your risk of cancer), while others preferred a stronger causation message (e.g., alcohol causes cancer). Some participants were in favor of using high levels of fear to motivate behavioral change, and some suggested the provision of facts about the alcohol-related cancer to make the risk more tangible. Most felt that it would be more effective to nominate specific forms of cancer, rather than mentioning cancer in general, and to have multiple messages that rotate to ensure exposure to a relevant form of cancer (e.g., prostate cancer for men and breast cancer for women). There were some, however, who were in favor of generic cancer messages on the basis that they would apply to all drinkers. There were also mixed views on including the word 'Warning' at the front of the message. Some believed this would assist in attracting attention, while others argued that brief messages would work better and hence all extraneous words should be omitted.
On the basis of the qualitative findings and the existing evidence relating to the relationship between alcohol and cancer, a series of 12 statements was developed that included examples of the attributes considered most relevant by the focus group participants. Table 1 lists the statements and their primary attributes. As the statements emerged from the qualitative data, there was coverage of each the primary attributes, but no attempt was made to equalize the number of messages exhibiting each attribute. Instead, the statements considered most motivating and that were best supported by the existing literature were carried through to the next stage of data collection.
Quantitative Phase
An online survey of Australian drinkers (n = 2,168) was used to measure the believability, convincingness, and perceived personal relevance of the 12 warning statements. A large web panel provider was used to access the sample and disseminate the online questionnaire. To be eligible to participate in the study, respondents had to be at least 18 years of age (the legal age to purchase alcohol in Australia) and consume alcohol at least two to three times per month. Quotas were used to generate a sample that was roughly equivalent to the Australian adult population in terms of key demographic attributes. Table 2 provides the sample profile.
Of the 12 statements included in this phase of the research, one was a general health message ('Warning: Alcohol harms your health') that was used as a comparison for the 11 cancer warning statements that were the primary focus of the study. Along with a series of demographic (e.g., age, gender, education), behavioral (e.g., amount and frequency of alcohol consumption), and attitudinal questions (e.g., quantity and frequency of alcohol consumption, perceptions of the healthiness of alcohol), respondents were randomly exposed to three of the 12 statements. The statements were presented as plain text, and were not incorporated into label designs to avoid contamination effects from other label elements. After exposure to each statement, respondents were asked to report the extent to which they found the message to be believable, convincing, and personally relevant. Respondents thus responded to each statement individually, and statement order was randomly assigned to minimize order effects.
Analysis. Descriptive analyses were initially conducted to calculate the believability, convincingness, and personal relevance scores for each of the 12 statements. Tests for normality (skewness and kurtosis) were also conducted prior to any inferential analysis. These tests revealed that the variables under investigation were all normally distributed (skewness < 2.00, kurtosis < 4.00).
The factors influencing believability, convincingness, and personal relevance scores across the 11 cancer statements were then examined using hierarchical multiple linear regression. Factors entered as independent variables were gender (male vs female), age, type of alcohol consumed most often (beer, wine, spirits), and education level (tertiary, non-tertiary). In addition, the Australian National Health and Medical Research Council (NHMRC) alcohol guidelines were used to allocate respondents to one of four harm profiles. According to the guidelines, "no more than two standard drinks on any day reduces the lifetime risk of harm from alcohol-related disease or injury" and "no more than four standard drinks on a single occasion reduces the risk of alcohol-related injury arising from that occasion". The four profiles were therefore constructed as follows: low short-term/low long-term risk – no more than four drinks in a single sitting over the previous 12 months and an average consumption level of two or fewer standard drinks per day; high short-term/low long-term risk – more than four drinks in a single sitting and an average daily intake of two drinks or fewer; low short-term/high long-term risk – no more than four drinks in a single sitting and an average daily intake of more than two drinks; and high short-term/high long-term risk – more than four drinks in a single sitting and an average daily intake of more than two drinks.
Regression analyses were accomplished in two steps. First, separate univariate regression analyses were conducted for each possible predictor to avoid any complications due to multicollinearity. In the second step, significant univariate predictors were included in a simultaneous multivariate regression model to determine the unique contribution of each significant predictor. The assumption of no multicollinearity was satisfied in all multivariate regressions conducted, with all independent variables associated with a tolerance level above the minimum criterion of 0.20 and a variance inflation factor below the maximum criterion of 10. A p value of < .05 was used as the significance cut-off.
To assess whether specific aspects of the statements under investigation were related to outcomes, the 11 cancer statements were clustered according to key message attributes. Paired-samples t-tests were then used to compare believability, convincingness, and personal relevance across these various attributes.