The data were derived from the 2018 Asia Best Workplace Mainland China (ABWMC) programme, which aims to support companies in building a healthy workplace. The ABWMC programme was designed by Peking University and organized by the American International Assurance Co. All companies may voluntarily join the programme and are free to withdraw. The inclusion criteria were as follows: (1) registered legal companies in China; (2) agreement to participate in the programme; and (3) at least 100 workers who are full-time employees.
Data were collected by employee questionnaires. The questionnaires were designed by Peking University and accessed via an online link through a survey company, Ipsos Inc. The questionnaire consisted of 50 items, including demographic and sociological information, payment and welfare information, individual health literacy and lifestyle information, smoking-related behaviour and quitting intentions, and disease and sick leave information.
The human resource departments of each company delivered the questionnaires to all employees. When employees first opened the link, content related to informed consent was shown, and the employees were able to choose whether to complete the questionnaire or quit. We considered them to have agreed to participate if they submitted the questionnaire through the link. The self-check function of the online survey system automatically identified missing data, logical errors and illegal characters. All participants were informed that the research team would analyse the data anonymously.
Our analyses used all participants for whom the variables of interest were available, with no imputation for missing data.
The measurement of alcohol use and smoking and quitting intention
Alcohol use was identified by a question that asked respondents, ‘How often do you drink alcohol?’ The response options were A: everyday, B: always C: sometimes, D: I never drink alcohol. These four groups were collapsed into two groups of non-drinkers (D) and drinkers (A or B or C).
Smoking was measured by the question, ‘Do you smoke now? A: yes, every day, B: yes, only sometimes, C: I have quit smoking, D: never.’ Participants who chose A or B were classified as smokers.
In the survey, participants were asked, ‘Are you going to quit smoking? A: yes, within a month, B: yes, within 6 months, C: yes, but not within 6 months, D: no plan for quitting’. Participants who chose A or B were classified as having an intention to quit.
Measurement of SHS exposure and SF workplace policy
In the survey, participants were asked, ‘How many days do you usually suffer from SHS exposure more than 15 minutes a week in the workplace? A: almost every day, B: 4-6 days, C: 1-3 days, D: never.’ Only participants who chose D were classified as having no SHS exposure.
Although there are legal ‘recommendations’ regarding SF workplaces, mainland China does not have national legislation for either comprehensive SF public places or SF workplaces. Some companies have voluntarily banned smoking due to safety requirements and health concerns. Therefore, in this study, we used company-level SF workplace bans as a measurement of indoor smoking policies. We measured worksite SF policy by asking about smoking rules in the workplace. The response options were as follows: A: no smoking ban, B: only ban smoking in parts of indoor area, C: complete smoking ban inside building, D: I have no idea. Only the participants who chose C were classified as working in a company with a SF policy.
We controlled for several variables of individual characteristics, such as gender, age, body mass index (BMI), marital status, ethnicity, education, yearly income, chronic disease and job position.
1. Mediation analysis to establish the mediating effect of alcohol drinking
To examine whether the association between smoking prevalence and SHS exposure was mediated by alcohol use, linear regression models were fitted based on the procedures outlined by Baron and Kenny16. The first equation regressed the mediator on the independent variable. The second equation regressed the dependent variable on the independent variable. The third equation regressed the dependent variable on both the independent variable and mediator.
The present study utilized the following criteria to establish mediation17:
- The independent variable (smoking) should be significantly related to the mediator (alcohol drinking) and the dependent variable (SHS exposure).
- The mediator (alcohol drinking) must be significantly related to the dependent variable (SHS exposure).
- The association between the independent variable (smoking) and the dependent variable (SHS exposure) must be attenuated when the mediator (alcohol use) is included in the regression model.
We then performed Sobel tests to estimate how much of the effect was mediated through the channel of alcohol use. The Sobel test is widely used to investigate the size and significance of indirect relationships. It is basically a specialized t-test used for examining whether the effect of the independent variable has a statistically significant reduction after the mediator variables are included18. In addition, as a supplemental method, we tested the mediation effects using a bootstrap test. As the result was almost the same, we report only the Sobel test results. We conduct the mediation analysis for male and female separately, because of significant differences in cigarette smoking by gender among adults in China (The 2018 China Adult Tobacco Survey shows 50.5% of males and 2.1% of females are smokers)19.
2. Structural equation modelling (SEM) to evaluate the role of alcohol drinking when SF workplace policies are imposed
We applied a structural equation modelling (SEM) approach to test two hypothesized models. For the first model, we used a full sample to test the role of alcohol consumption in the pathways between SF workplace policy and SHS exposure. For the second model, we added smoking amount and quitting intention to the model and tested the role of regular alcohol drinking in the pathways between workplace SF policy and quitting intention among smokers. The SEM approach can be used to test overall models rather than individual coefficients, incorporating multiple dependents as well as mediating variables20, 21.
We used the following model fit statistics that have proven to be meaningful in SEM20, 21:
Bentler’s comparative fit index (CFI): recommended>0.95;
Tucker-Lewis index (TLI): recommended>0.95;
Root mean square error of approximation (RMSEA): <0.06.
We used AMOS 24.0 for SEM and STATA 14.0 for mediation analysis.