Article Text
PDF
Abstract
Background Norway passed legislation banning smoking in restaurants, bars and other public spaces in 2004. This study tracks changes in hospitality workers' attitudes towards Norway's ban over three time points, using growth modelling analysis to examine predictors of attitude change.
Methods Participants were a national sample of 1525 bar and restaurant workers. Surveys were conducted, by phone or internet, one month before the ban's implementation and at 4 and 12 months thereafter. Exploratory principal components analysis of nine survey items revealed one primary attitude component. A latent growth model was fitted to the data to examine trajectories of attitude change and individual differences in rate of change.
Results Respondents supported the ban before implementation and increased support at 4 months (p=0.021) and again at 12 months (p=0.001). Concern for one's job followed a quadratic trend, increasing at 4 months and decreasing at 12 months (p<0.001). All demographic categories were associated with attitude increase; rate of increase was greater for females than males. Two within-person variables—change in smoking status and change in job concern—strongly predicted (p<0.001) respondents' deviations from their predicted group trajectories, explaining over 70% of residual between-person slope variance.
Conclusions Norway's hospitality workers increased their support of the ban over its first year. The strong influence of the within-person variables leads to two primary policy recommendations. First, support should be provided to assist cessation efforts and prevent relapse. Second, informational campaigns should inform hospitality workers about evidence that smoking bans are not economic threats to the industry.
- Cessation
- environmental tobacco smoke
- public opinion polls
- public policy
Statistics from Altmetric.com
The flourishing momentum in this decade for nationwide clean air legislation has changed the political landscape for tobacco control policy. Comprehensive national smoking bans have improved the quality of indoor air and the health of workers, and have reduced both overall smoking prevalence and the level of cigarette consumption by continuing smokers.1–5 Ireland, Norway and New Zealand all went smoke-free in 2004, and the list of countries has expanded steadily since then.6 7 Nations that have ratified the Framework Convention on Tobacco Control, numbering 168 as of January 2010,8 are committed to adopting comprehensive clean air legislation, and thus there is keen interest in the experiences of countries that are early adopters of these policies.
Because of the political and economic contexts surrounding efforts to implement national clean air legislation, supportive public opinion is a critical ingredient in preparing this legislation and making it successful. Therefore it is noteworthy that smoking prohibitions have proved popular with the public, including many smokers.9 10 Evidence is accumulating that public support for smoking bans increases after the bans have been implemented, based on independent population surveys conducted before and after the legislation takes effect.10–12 Support among restaurant and bar workers has also been found to increase, in studies based on both independent surveys13 and panel surveys that track the same respondents before and after implementation.14–16
Nevertheless, there remains a great deal still to learn about public attitudes towards smoking bans, especially among critical populations such as bar and restaurant workers. The studies of attitudinal change have thus far been primarily descriptive, and generally have involved only two time points. To date we still know little about which demographic subgroups may be most or least supportive at the outset, how the patterns change over time, or which factors may be influential in either strengthening or weakening individuals' support. The exploration of many of these important questions requires that samples be followed over multiple measurement points.
The hospitality industry, particularly restaurants and bars, is a pivotal sector for establishing smoke-free legislation. Opponents, frequently with backing from the tobacco industry,17 have argued that smoking bans will lead to economic losses, especially in bars. Those fears have been convincingly refuted,18 19 but restaurants and bars are still sometimes exempted from full compliance with clean air requirements. Restaurant bans have been called a ‘tipping point’20 for public acceptance of smoking bans, and the support of workers and patrons in food and drink establishments is widely recognised as a critical political element in achieving comprehensive legislation.10 17 From a public health perspective, the hospitality industry is also important because of the hazards to these workers' health: in the absence of clean-air controls, bar and restaurant employees are often exposed to particularly high levels of environmental tobacco smoke.21 22
In this article we address the topic of change over time in attitudes towards Norway's nationwide smoking ban among a national sample of bar and restaurant workers. Our data are drawn from a panel survey that was conducted over three time points—one month preceding the June 2004 introduction of the ban, and at 4 and 12 months thereafter. We use growth curve modelling procedures to examine how the attitudes of these respondents changed over the course of the year, with particular attention to identifying any policy implications and strategies that may emerge for promoting the success of comprehensive clean air legislation.
Method
Sample and procedure
The survey sample consisted of 1525 employees working in restaurants, cafeterias, bars and other eating and drinking establishments in Norway in May 2004. To select respondents, a random sample of businesses was first identified from a public register listing all businesses in the Norwegian hospitality industry. These establishments were contacted by telephone, and employees from each establishment were identified and invited to participate. To avoid selection bias, the identification of employees to be interviewed followed a procedure involving random selection of the first letter of employee surnames. The overall participation rate was estimated at 53%. Participants were given the choice between being interviewed by telephone (n=1337) or completing the questionnaire online (n=188).
The study was approved by the Regional Committee for Medical Research Ethics for Western Norway, located at the University of Bergen's Faculty of Medicine. Baseline data were collected in May 2004 (T1), shortly before the ban went into effect in June. Follow-ups were conducted in September/October 2004 (T2) and May 2005 (T3). For each participant, the mode of data collection (phone or online) remained consistent across follow-up waves.
Measures
Overall support for the smoking ban
This was assessed with the following item (in translation): ‘How do you feel about serving places being smoke-free as of 1 June? Are you generally positive, negative, or neutral?’
Opinions about the smoking ban and clean air policies
Nine opinion items (listed in table 1), each rated from strongly disagree (1) to strongly agree (5), addressed beliefs, affective responses and behaviours relating to the ban, and were considered potential indicators of respondents' underlying attitudes towards the ban.23 24 Based on findings from principal components analysis, four of them were later used to construct a latent attitudinal factor, which was assessed at each time point and became the dependent measure of the growth curve analysis.
Changes over time in restaurant and bar employees' responses relating to the smoking ban
Current smoking status
Categories were non-smoker, occasional smoker or daily smoker.
Demographics
Variables included gender, age (combined into four categories from 15–24 through 45–69), educational level completed (lower secondary, upper secondary or university) and type of establishment (11 options combined into dichotomous categories of bar/nightclub and restaurant/café/other).
Data analysis plan
We first summarised the scores on the overall support item and the nine opinion items, examining them for change over time using χ2 tests and analysis of variance. The opinion items were then analysed, using exploratory principal components analysis, to determine whether some subset could be utilised as a unidimensional attitude scale. These analyses were conducted using SPSS 17.0.
Following this, a latent growth model was fitted to the data to examine the overall trajectory of attitude change over time and any individual differences in rates of change. Multilevel modelling addresses the dependency issues of repeated measurements by decomposing outcome variance into between-person (BP) and within-person (WP) variance. Growth modelling25 uses time as a WP predictor to model individual trajectories using, in this case, the intercept and linear slope.
Latent growth modelling is conducted in a structural equations framework, and latent growth factors, whose means and variances correspond to the fixed and random trajectory effects, are invoked as the underlying causes of change.26 We tested an invariance model and several subsequent growth models, using Mplus v5.2 and the WLSMV estimator.27
Results
Sample information and attrition analysis
The sample consisted of slightly more females than males (52.9% vs 47.1%). Most respondents (52.3%) were 34 or younger. The modal level of educational attainment (58.0%) was upper secondary (consisting of 13 years of schooling in Norway). Most respondents (60.9%) worked in restaurants, cafeterias, coffee bars or other settings, while 39.1% worked in bars, clubs or discos. At baseline, 52.9% of respondents were daily smokers, 7.7% were occasional smokers and 39.4% were non-smokers. Additional descriptive details on the sample are available elsewhere.28
Among the 1525 respondents at T1, 879 (57.6%) participated at T2 and 579 (38.0%) participated at T3. An extensive sample attrition analysis, reported previously,28 found that participant attrition did not introduce significant levels of bias with regard to variables measuring demographics and smoking behaviours. In addition, for this analysis we conducted comparisons on survey completers versus non-completers at both T2 and T3, for all previous responses on the overall support item and the nine opinion items. Of 30 statistical comparisons, only two were significant at α=0.05, suggesting that attrition was not a significant problem for these analyses.
Changes in overall support for the ban
Table 2 lists the percentages of respondents, for those who were present at all three waves, who reported positive orientation towards the smoking ban on the overall support item. The level of support went from 49.2% of respondents in May 2004 to 60.3% in May 2005, with statistically significant increases occurring at both time points. More restaurant workers than bar workers expressed support at each wave, but bar worker support increased sharply between T2 and T3. Both daily and occasional smokers were far less supportive than non-smokers, but support of occasional smokers rose dramatically between T1 and T2 (statistically significant despite the small number), while daily smokers' support rose sharply between T2 and T3. A large proportion of non-smokers expressed support at baseline (68.6%), and this figure was modestly higher one year later.
Changes across time in restaurant and bar employees' support for the smoking ban*
Changes in opinions on the ban and clean air policies
On the nine opinion items relating to the ban and clean air policies, there were, with one exception, highly significant (p<0.001) linear changes between T1 and T3 (table 1). The exception was the item about concern for one's job (‘I am afraid for my job because of the ban.’). For that item, the mean score rose between T1 and T2 (representing heightened concern), and then returned to its baseline level at T3. Thus this item had a non-significant linear trend but a highly significant quadratic trend, which was present for both the bar and restaurant worker subsamples. Several of the other items had significant quadratic trends in addition to linear trends; for items 1–3, this indicates that responses supportive of the ban rose more steeply in the second measurement interval than the first.
The opinion items were then examined using exploratory principal components analysis (PCA). Separate PCAs were conducted for responses at each of the three waves, and results across waves were strongly parallel. In each case, one primary component comprised a proportion of total variance varying between 38.4% (at T3) and 41.3% (T1). Loadings were strongest for the first four items listed in table 1. The pattern matrices at the three time points were highly consistent as well, with loadings ranging from 0.776 to 0.832 at T1, 0.760 to 0.850 at T2 and 0.756 to 0.854 at T3. These four items all reflect agreement with the legislation's purpose and effect, and are consistent in the direction of their scaling (high score signifies positive orientation towards the ban). Grouping the items as a scale, the mean scores correlated highly at each wave (between 0.732 and 0.746) with the overall support item (displayed in table 2), indicating strong convergent validity. Finally, internal consistency reliability (Cronbach's α) for this four-item scale ranged between 0.814 (T3) and 0.830 (T1). Based on this strong psychometric evidence, these four items were used as indicators for the latent attitudinal factor in the growth curve analysis. In addition, the item focusing on job concern was used as a time-varying predictor of attitude in the analyses that followed.
Growth model analysis
Using the four polytomous (5-level) attitude items, we constructed a latent attitude factor at each occasion and modelled factor-level growth. These models conceptualise a latent continuous and normally distributed response variable underlying each observed response, and a set of cut points is estimated to produce the observed counts.29 Factor score growth models require longitudinal factorial invariance as a necessary assumption.30 We tested a model of strong longitudinal factorial invariance31 32 and obtained excellent model fit (comparative fit index (CFI)=0.986, Tucker-Lewis index (TLI)=0.995, root-mean-square error of approximation (RMSEA)=0.039).
Table 3 shows the model fit, effects and pseudo-R2 for the unconditional growth model of attitude (Model A) and four conditional models. These models fully address issues of dependency due to repeated measurements through two mechanisms: using occasions as variables and specifying an unstructured random effect covariance matrix. We introduced time-invariant, between-person (BP) and time-varying, within-person (WP) predictors in successive models (B and C) to examine the relative strength of these predictors. After fitting models using all available data (n=1515) under the full information, maximum-likelihood assumption of data missing-at-random, we fitted the fully conditional model (including both BP and WP predictors) to data from participants who were present for at least one of the follow-ups in addition to baseline (n=1054, Model D). As a further check on sensitivity of estimates with missing data, we fitted a model (not shown) requiring responses at all three occasions (n=577); results were highly comparable, providing assurance that the pattern of results is not overly sensitive to missing data assumptions. Finally, we fitted a reduced model that removed non-significant interactions and main effects (Model E). While some effects move above or below the p=0.05 significance level across the successive models, the pattern of results is generally consistent and fit indices remain very good to excellent across all models.
Comparison of growth curve models
The growth model is presented in figure 1. Demographic and other T1 variables are time-invariant predictors affecting initial level (intercept) and linear change (slope), the attitude variable's latent growth factors. Two time-varying predictors (change in smoking status from T1 and change in job concern from T1) affect individuals' deviations from this linear change at both T2 and T3. Finally, the three occasion-specific attitude factors account for the four attitude indicators—the survey responses—at each measurement occasion.
Growth curve model of hypothesised influences on respondent attitudes towards the smoking ban. I-Att, S-Att: intercept and linear slope, respectively, of latent attitude variable. Coefficients are taken from conditional model, including respondents present for at least two time points (Model D, table 3). The S-Att coefficients of 0.33 and 1.0 refer to 4 months and 12 months, respectively. The time-varying predictors ‘Change in smoking status’ and ‘Change in job concern’ are correlated across time points.
As table 3 displays, different categories of respondents, based on the BP baseline predictors, vary in their initial attitude levels (represented by the intercept). Being female predicts a more favourable initial attitude (+0.392 compared to males), while each successive level of smoking or job concern at T1 predicts a decrease in initial attitude of 0.572 and 0.408, respectively.
However, all categories are predicted to become more favourable over the year. For the referent demographic condition (male, 15–24 years old, lower secondary education, non-smoker at T1, no concern for job security), the model predicts an initial level of 0 and an increase over the year of 0.785 units of support for the ban. With respect to effects on slope, only gender (+0.556, p=0.028) and the interaction of gender and age (−0.351, p=0.022) are significant BP predictors of attitude change. Compared to the referent male with a predicted change of 0.785, a comparable female is predicted to increase by 1.345 (0.785+0.556)—almost double—over the first year. But the effect of gender decreases as respondent age increases.
The strongest and most consistent effects are the time-varying changes (change across measurement points) in smoking and job concern. Everyone, including those smoking at T1 and those with high job concern at T1, is predicted to grow more supportive of the ban—unless there are increases in smoking and/or job concern at T2 or T3. Such increases, if large enough, will result in a slowing or actual decline in attitude growth.
This pattern is illustrated graphically in figure 2. The top left panel shows a 10% sample of predicted individual trajectories. The overall upward growth in favourability is apparent. However, some people are seen to decrease across the year, or between T2 and T3 after having increased from T1 to T2. In contrast, the top right panel shows the predicted trajectories for the same 10% sample with the time-varying effects not included in the plots. All slopes are positive. It follows that the descending lines seen in the first panel are entirely the consequence of the time-varying predictors: shifts in smoking status and/or job concern. Four selected cases in figure 2 illustrate this further. For reference purposes, the cut points for question 1 are superimposed on individual trajectory plots shown in figure 2.
Group and individual growth trajectories. For individual cases, T2 and T3 values of the time-varying predictors (change in smoking status and change in job concern) are expressed as change from the individual's status at T1. Levels of smoking status: 0, non-smoker; 1, occasional smoker; 2, daily smoker. Levels of job concern (see table 1, item 5): 1, lowest (strongly disagree); 2, low; 3, moderate; 4, high; 5, highest (strongly agree).
The first individual case displayed is a 25–34-year-old female with lower secondary education, initially an occasional smoker with no job concern. Based solely on her demographic category and initial levels of smoking and job concern, her predicted trajectory follows the broken line. However, taking the time-varying predictors into account, her attitude is predicted to remain relatively flat at T2, at which she reports being a daily smoker (+1 unit change), and to decrease at T3, at which she remains a daily smoker and has increased job concern. In panel 2, the 35–44-year-old female respondent drops sharply from her predicted trajectory at T2 (at which she moves from non-smoking to daily smoking and increases her job concern), but rebounds considerably at T3 (non-smoking once again, though job concern remains high). In panel 3, the 35–44-year-old male increases sharply relative to predicted trajectory at T2 (change from daily smoking to non-smoking) but drops at T3 (back to daily smoking plus heightened job concern). In panel 4, the 35–44-year-old female respondent drops sharply at T3, because of increases in both smoking and job concern.
The powerful effect of the time-varying predictors is reflected in the change in random effects and pseudo-R2 values across models. Model B, with only BP predictors, explains 27.1% and 11.8%, respectively, of the BP intercept and slope variances estimated in the unconditional model. However, after including the WP predictors (Models C-E), the percentage of explained slope variance jumps to over 70%. This reflects unusually strong predictive power for these models.
Discussion
These findings from Norway's comprehensive smoking ban are consistent with studies from other countries, which report strong pre-legislative support that increases further after the law takes effect. In addition, the use of growth curve analyses to track a national sample of bar and restaurant workers over three time points allows for patterns of attitudinal change to be modelled among these respondents with very high accuracy.
Before implementation, significantly greater support existed among females relative to males, non-smokers relative to smokers, and those who had relatively little concern about the security of their jobs. Once the legislation took effect, two major patterns of attitude change emerged as particularly noteworthy. First, for each of the many demographic combinations of gender, age, education, smoking status and type of work establishment, respondents' attitudes became steadily more positive over the course of the ban's first year. Before the introduction of the time-varying WP factors, not a single group trajectory was negative, as illustrated in figure 2.
Second, two time-dependent WP factors were found to be extremely influential in predicting individuals' deviations from these demographically based linear trajectories: change in the respondent's smoking status (across categories of daily, occasional and non-smoker) and change in concern for the security of one's job. Respondents who became more negative towards the ban at either of the follow-up points had most likely increased their smoking, become more anxious about their job since baseline, or both. Conversely, respondents whose support grew particularly rapidly had probably experienced a decrease in smoking or job concern, or both. These predictors of individual attitude trajectory were the directions of change in the smoking and job concern variables, rather than the static levels of those variables at T2 or T3. Remarkably, more than 70% of the variance in attitude change (represented by slope variance) was explained by the BP and WP predictors in our analysis.
One might have expected these hospitality workers to display lower support for the ban than Norway's population at large, given that they smoked at levels considerably higher than Norway's national averages in comparable demographic strata,28 and that public debate before implementation included speculation about potential job losses in the hospitality sector. To some degree this difference in support did exist, but it was modest. A general population survey conducted in May 2004, one month before implementation, found that 54% of the Norwegian public supported the smoking ban.33 This was roughly similar to the level of support found in the present study among restaurant workers (52.3%, table 2), though bar workers' support was somewhat lower (44.4%). The reasonably high initial support among hospitality workers may have been due to several factors. One was a mass media campaign conducted by the Norwegian government between April and June 2004 in preparation for the ban, with one prominent theme being the right of hospitality workers to work in a smoke-free environment.33 Previous legislation in 1988 had banned smoking in other workplaces, but excluded restaurants and bars. Another was strong endorsement for the legislation from the Norwegian Confederation of Trade Unions, owing to the focus on improving working conditions.
What are the policy implications of these findings? Health officials should be attentive to the two factors that were found to disrupt the predicted growth over time in hospitality workers' favourable attitude towards the ban. First, they should attempt to counteract increases in individuals' smoking, which was one of the most powerful predictors in our model. We think it is likely, considering the age range of our respondents (79% were 25 or older), that a high percentage of the cases of smoking increases between T1 and T3 reflect relapse following attempts to reduce or quit, rather than first-time smoking uptake. Smokers who relapse after quit attempts may feel increasing discomfiture and disapproval of smoking restrictions, since bans tend to denormalise smoking and increase perceptions of its unacceptability.34 Relapse might also exacerbate employees' concerns about their job security in a non-smoking work environment, thus amplifying the effect of our second time-varying predictor. Therefore, providing resources to support quit attempts—eg, quitlines, clinics, peer support programmes and pharmacological supports—can probably accelerate endorsement of the ban, as more individuals join the ranks of non-smokers.
The strong effect of our second time-varying predictor, changes in job concern, suggests another policy recommendation: health officials need to counteract the misperception, perpetuated by the tobacco industry,17 18 that smoking bans threaten businesses and jobs. This view is clearly refuted by extensive evidence showing restaurants and bars to be unaffected, or positively impacted, by bans.18 19 Restaurant and bar employees will understandably be highly sensitive to this issue, and there should be energetic media strategies to provide the hospitality industry, as well as the public, with accurate information. In Norway, an economic evaluation of the ban's effects found that overall revenues in the hospitality industry were essentially unchanged (down by 0.8%) in the first year after the ban. The number of employees in the industry declined slightly in the last quarter of the ban's first year, but the decrease was well within standard patterns of fluctuation and rebounded to pre-ban levels by the next quarter (all comparisons seasonally adjusted).33 Our present results showed a highly significant quadratic trend in respondents' job concerns, increasing sharply immediately following the ban's implementation but declining to pre-ban levels by the end of the first year. The growth curve findings suggest that effective communications that allay hospitality workers' job anxieties can promote rapid attitudinal change favourable to the ban.
In conclusion, the accumulating evidence of the past several years has established that smoking bans are popular, but looking more deeply into patterns of attitude change will help health officials to anticipate public opinion trends and respond effectively, for example, by countering political opposition and promoting acceptance among reluctant groups. The present study has taken an individual differences approach, seeking to identify predictors of attitude change among members of one key audience. Research in this area is just beginning, and given the growing momentum of national initiatives for clean air legislation, these inquiries have the potential for timely and significant practical applications.
What this paper adds
Previous studies have shown that popular support for national smoking bans increases after the bans come into force, even among smokers. However, these studies typically use only two time points and are limited in their ability to explore patterns of attitude change towards clean air legislation.
This study uses Norway's national smoking ban to examine this attitude change more deeply among bar and restaurant workers. Using a year-long panel design that began one month before the ban's implementation and includes three points in time, the study examines both group and individual trajectories of change.
The results show that patterns of attitude change towards Norway's ban were well described by a linear trajectory across the three time points. Demographic subgroups had different starting levels and different rates of change, but all groups grew more positive across the year. However, individuals were found to deviate from their group-predicted values based strongly on only two variables: change in their smoking status and change in their job concern since the initial measurement. The findings suggest that initiatives to address these two concerns, once a smoking ban has been implemented, will substantially strengthen the growth of hospitality workers' support for the legislation.
Acknowledgments
The authors thank Rita Lill Lindbak of the Norwegian Directorate of Health, Karl Erik Lund of the Norwegian Institute for Alcohol and Drug Research, and their teams, for valuable discussions and contributions during the planning of the surveys of bar and restaurant employees. We also thank Alan Acock of Oregon State University for his valuable suggestions about data analysis, and Shauna Tominey of OSU for assisting with the analysis. Finally, we thank the journal's reviewers for their thoughtful and helpful comments.
References
Footnotes
Funding Norwegian Directorate of Health.
Competing interests None.
Ethics approval This study was conducted with the approval of the Regional Committee for Medical Research Ethics for Western Norway, University of Bergen Faculty of Medicine.
Provenance and peer review Not commissioned; externally peer reviewed.