Study design and data
The present study is based on survey data concerning smoking behaviour and responses to CPHWs. Data was collected from February to March 2017 by The Institute for Empirical Social Studies (IFES) on behalf of the authors. A random sample of 500 persons was drawn from the Austrian population of both current smokers and ex-smokers (i.e. smoking cessation during the last 12 months). Given that smoking usually develops in the period from adolescence to young adulthood [15] and that an early onset of smoking is strongly related to future smoking patterns [16, 17], smoking prevention is of great public health interest among the younger age group in particular. We therefore aimed for a disproportionately higher rate of 15–23 year olds in order to increase statistical power to detect potential differences between the younger and the older age groups. Self-reported data were collected using computer-assisted telephone and web interviews (CATI/CAWI).
Measures
The conceptualization of the survey variables was largely based on previous survey studies examining the effects of CPHWs on smokers’ behaviour and intentions [7, 18–21]. The variables used in this study comprised behavioural, cognitive, and emotional responses to the CPHWs, measures of salience of the CPHWs, smokers’ assessments of the health risks of smoking, smokers’ intentions to quit, smoking history and dependence, and socio-economic and socio-demographic factors.
Smoking status
Participants were categorized according to their smoking status, i.e. to the fact whether they smoked every day (regular smokers) or occasionally (occasional smokers), or whether they were ex-smokers that had quit smoking during the last 12 months (ex-smokers).
Smoking cessation because of pictorial CPHWs
To assess the direct effects of health warnings on the decision to quit smoking, ex-smokers were asked to declare whether or not the new cigarette packs with the pictorial warnings were the main reason or at least one of the reasons why they quit smoking. Response options were: 0 =no; 1 =yes, one reason among others; 2 =yes, the main reason.
Intention to quit smoking
To quantify quitting intentions, regular and occasional smokers (but not ex-smokers) were asked how strongly they wanted to quit smoking. Response categories were: 0 =not at all; 1 =somewhat; 2 =quite a lot; 3 =a lot. In order to facilitate interpretability in subsequent regression analysis, we dichotomized this variable by combining response categories 1, 2, and 3. The resulting categories thereby represented no intention to quit smoking on the one hand, and at least a slight intention to quit smoking on the other hand.
Warnings impact score (WIS)
In order to obtain a total score for the impact of textual and pictorial health warnings, we summed up the dichotomized scores of four items (for a similar approach see [20]):
-
1
Change in feelings because of the new CPHWs: Regular and occasional smokers were asked whether or not the new CPHWs had changed their feelings about smoking. Response options were: 0 =no, feelings have not changed; 1 =yes, I feel bad smoking; 2 =yes, I feel very bad smoking. We dichotomized this item score by combining response categories 1 and 2 into one category.
-
2
Forgoing cigarettes because of pictorial CPHWs: Regular and occasional smokers were asked to specify whether or not they had reduced cigarette consumption due to the pictures on the packs. Response categories were: 0 =no; 1 =yes, I smoke a little less; 1 =yes, I smoke much less. Again, we combined response categories 1 and 2 to dichotomize this item score.
-
3
Frequency of reading warnings: Regular and occasional smokers were requested to indicate how often they had read the textual warnings on the cigarette packs during the last weeks. Response categories were: 0 =never; 1 =once; 2 =repeatedly; 3 =often; 4 =almost always. For dichotomization, we combined response categories 0–1 and 2–4 into one category, respectively.
-
4
General attitude toward pictorial warnings: Regular and occasional smokers were asked whether or not they would agree that it is generally a good idea to show pictorial warnings on the cigarette packs as a deterrent (0 =no; 1 =yes).
Higher values on the WIS represented more positive responses in terms of provoking the effects intended by the health warnings. This sum score was then used in regression analysis. Confirmatory factor analysis confirmed the one-dimensional structure of this scale, and internal consistency was sufficient (Cronbach’s α=0.75).
Assessment of the validity of warnings
Regular and occasional smokers were requested to state whether they believed that the warnings on the cigarette packs are valid, on a 7-point rating scale ranging from 1 =very untrue of what I believe, to 7 =very true of what I believe.
Smoking risk assessment
To quantify smokers’ knowledge about the health risks of smoking, regular and occasional smokers were requested to assess (1) smokers’ risk (compared to non-smokers of the same age and sex) to develop a life-threatening disease, (2) smokers’ risk to develop a life-threatening disease while they keep smoking, and (3) smokers’ risk to die earlier than if they quit smoking. Response categories ranged from 1 =my risk is much lower to 7 =my risk is much higher for the first item and from 1 =very low to 7 =very high for the latter two items, on a 7-point Likert-scale, respectively. We summed up the three item scores and used the resulting sum score in a subsequent analysis. Internal consistency was found to be good (Cronbach’s α=0.86).
Use of neutral covers
Regular and occasional smokers were asked to indicate whether or not they used a neutral cover to hide the pictures on the cigarette packs (0 =no; 1 =yes).
Cigarette dependence
The measure for cigarette dependence was based on the score used in the Fagerström Test for Cigarette Dependence (FTCD) [22, 23]. Internal consistency was found to be good in the present study (Cronbach’s α=0.82). Again, these test items were only presented to regular and occasional smokers.
Additional variables
Both the responses to the CPHWs and the intentions to quit smoking may be affected by individual-level differences in factors such as smoking history and socio-demographic/socio-economic backgrounds [10, 20, 24, 25]. We thus considered the number of years smoking (≤3 years; 4–10 years; >10 years), household income (≤2000 Euros; 2001–3000 Euros; >3000 Euros), educational level (highest level reached: still in school or compulsory education; apprenticeship certificate or diploma from vocational school; high school diploma; university degree); age group (15–23 years vs. >23 years), and sex (male vs. female).
Statistical analysis
First, descriptive analyses were used to explore the sample characteristics for both regular and occasional smokers, each stratified by age group. We conducted χ2-tests of independence and t-tests in order to investigate potential differences with regard to smoking status and age group. Second, we used multiple linear regression analysis in order to examine the determinants of the health warnings impact (WIS) and multiple logistic regression analysis in order to predict the intention to quit smoking using the WIS. Other predictor variables included the assessment of the validity of warnings, smoking risk assessment, the question whether or not neutral covers were used, the cigarette dependence score (FTCD), smoking status, and the additional variables.
In total, 5.08 % of all data used in this study were missing. The proportion of missing values across the variables varied between 0 % and 29 %. In order to handle missing data, we repeated regression analysis for each of the m=50 multiple imputed datasets and pooled the results according to Rubin’s rules [26]. In the imputation model, we used the entire set of predictor and outcome variables to impute the missing data. All statistical analyses were carried out using R version 3.4.2 [27], and we used the R-package mice version 2.30 for multiple imputation by chained equations [28].