This population-based cross-sectional study was conducted in Mainland China from July 10 to September 15, 2021. A multi-stage, hybrid sampling strategy was employed to recruit participants from 120 cities, aiming to capture geographic and socioeconomic diversity.

Stage 1: City Selection. A stratified hybrid sampling strategy was adopted. First, we purposively included all 32 provincial capitals and municipalities directly under the central government (e.g., Beijing, Shanghai) to ensure coverage of major administrative and population centers. Subsequently, for each province/autonomous region, we randomly selected 2 to 6 non-capital prefecture-level cities from a complete administrative division list using a random number table, resulting in a total of 120 cities.

Stage 2: Investigator Recruitment and Training. In each selected city, local investigators or investigation teams (with ≤10 members per team) were recruited through open channels, constituting a convenience sample. All investigators underwent standardized online training to ensure a consistent understanding of the survey protocol, questionnaire content, and ethical guidelines.

Stage 3: Participant Recruitment. Due to the absence of a comprehensive sampling frame for residents, a non-probability sampling method was employed at the individual level. Investigators distributed the online questionnaire (hosted on the Wenjuanxing platform) primarily through their personal and social networks (e.g., community and professional WeChat groups). Each independent investigator was tasked with collecting 30–90 valid questionnaires, while each team was responsible for 100–200. This approach combined elements of convenience and snowball sampling, with no randomized selection of individuals.

The selected cities collectively represent the major geographic (Eastern, Central, Western regions) and socioeconomic strata of mainland China, capturing a diverse range of occupational sectors and community types (see in the Supplementary material).

The data were derived from the 2021 China Family Health Index Survey (CFHI-2021). The survey was administered online via the Wenjuanxing platform. Investigators distributed questionnaire links one-on-one to residents in their networks. Participants provided electronic informed consent before completing the questionnaire, and investigators recorded the questionnaire numbers.

Inclusion Criteria: ① Chinese nationality; ② Permanent resident in China (annual time away from residence ≤ 1 month); ③ Voluntary participation with informed consent; ④ Ability to complete the questionnaire independently or with investigator assistance; ⑤ Capacity to comprehend the questionnaire items.

Exclusion Criteria: ① Delirium or severe mental disorders; ② Concurrent participation in similar research projects; ③ Unwillingness to cooperate.

A participant flowchart is provided in Supplementary Figure 3.

The primary outcome was the level of support for establishing smoke-free environments, measured by a self-reported support score. The questionnaire also collected comprehensive data on: Sociodemographic characteristics: sex, age, education, occupation, marital status, etc. (see Supplementary Table 1 for categories). Standardized Scales: 10-item Big Five Inventory (BFI-10) for personality (9), Perceived Social Support Scale (PSSS), Family Health Scale-Short Form (FHS-SF), Eating Behavior Scale-Short Form (EBS-SF), Fagerström Test for Nicotine Dependence (FTND, For never-smokers and former smokers who had quit, FTND scores were set to zero), EQ-5D-5L for quality of life. For the primary analysis, a total BFI-10 score was used as a parsimonious global measure of personality’s association with support, while domain-level analyses were reserved for exploratory models (see Supplementary material). Other factors: family structure, and financial status. The internal consistency (reliability) of the multi-item scales, as measured by Cronbach’s alpha in our sample, was as follows: BFI-10 (α = 0.72), PSSS (α = 0.91), FHS-SF (α = 0.88), EBS-SF (α = 0.80), indicating acceptable to good reliability (all in the Supplementary material).

To ensure data quality, the following measures were implemented: (1) Standardized online training for all investigators; (2) Use of a unified and validated online survey platform; (3) Logical checks and data validation rules within the questionnaire; (4) Monitoring of questionnaire completion time to identify and exclude careless responses.

It is important to acknowledge that the non-probability sampling method, while practical for large-scale national outreach, limits the generalizability of our findings and may introduce selection bias, as participants were recruited through social networks and may not be fully representative of the entire Chinese population.

Data were analyzed using SPSS (version 27.0). Descriptive statistics presented counts and percentages for categorical variables and means with standard deviations for continuous variables. Univariate analyses (independent samples t-tests for binary variables; one-way ANOVA for multi-level categorical variables) were conducted to explore associations between individual factors and the support score. Missing data were handled using list wise deletion. A multivariable linear regression analysis was performed to identify factors independently associated with the support score. We used the Enter method, forcing all pre-specified variables from the socio-ecological framework into the model simultaneously, without employing any automated variable selection procedures. The final model included all pre-specified variables from the socio-ecological framework. All variables listed in Section 3.3 were considered as potential independent variables. Categorical variables were dummy-coded (see Supplementary Table 1). The assumptions of linear regression (linearity, homoscedasticity, normality of residuals) were checked and deemed satisfactory. Multicollinearity was assessed using Variance Inflation Factors (VIF), with all VIF values below 2.5, indicating no significant multicollinearity.

The study protocol was approved by the Institutional Review Board of Ji’nan University, Guangzhou, China (Approval no: JNUKY-2021-018). All participants provided electronic informed consent before participation.

A total of 11,031 valid questionnaires from 23 provinces, 5 autonomous regions, and 4 municipalities in mainland China were analyzed. The sample consisted of 5,033 males (45.6%) and 5,998 females (54.4%). The age distribution was as follows: 12–18 years (9.7%), 19–40 years (48.3%), 41–60 years (31.6%), 61–75 years (7.9%), and over 75 years (2.5%). Detailed sociodemographic characteristics are presented in Table 1.

The mean score for support of smoke-free environments was 78.75 (SD = 26.996), indicating a generally high level of support. More than half of the respondents showed a support score above 50 (Figure 1). Geographic distribution revealed variations across provinces, with central China regions (including Shanxi and Henan) demonstrating relatively higher support scores (Figure 2).

The multiple linear regression model (R² = 0.13) identified several significant correlates of support, which explained 13% of the variance (R² = 0.13) in Table 2.

Sociodemographic and behavioral factors: female gender (β = 5.505, 95% CI [4.409, 6.601]), higher education (Undergraduate or above: β = 4.110, 95% CI [2.475, 5.746]), and having children (β = 1.554, 95% CI [0.276, 2.833]) were associated with stronger support. In contrast, being married (β = −3.375, 95% CI [−4.849, −1.900]), frequent drinking (e.g., Weekly: β = −5.275, 95% CI [−6.851, −3.699]), and higher tobacco dependence (e.g., Severe: β = −21.968, 95% CI [−27.671, −16.265]) were associated with lower support. It is noteworthy that this positive association emerged only after adjusting for confounders in the multivariable model, whereas the univariate analysis showed the opposite trend (see Table 1), highlighting the role of confounding factors.

Psychosocial factors: stronger family health (FHS-SF: β = 0.645, 95% CI [0.543, 0.746]) and greater perceived social support (PSSS: β = 0.142, 95% CI [0.092, 0.193]) were positive correlates. Conversely, a higher BFI-10 total score (BFI-10: β = −0.559, 95% CI [−0.853, −0.266]) and unhealthier eating habits (EBS-SF: β = −0.328, 95% CI [−0.444, −0.212]) were negative correlates.

Stratified analyses revealed that the influence of these factors was not uniform across populations (See Supplementary Tables S3–S5).

By age: the strongest predictors varied across life stages, from parental influence in adolescents, to education in young adults, and the pronounced impact of tobacco dependence in middle-aged and older adults (see Table 3).

By gender: the positive association of having children and family health was stronger among females, while the negative impact of tobacco dependence and drinking was more pronounced among males (see Table 4).

By smoking status: the negative association of tobacco dependence was most pronounced among current smokers. In contrast, former and never-smokers’ support was more strongly influenced by psychosocial factors like family health and social support (see Table 5).

The generally high level of public support (mean score: 78.75) indicates a fertile ground for advancing smoke-free policies in China. Consistent with global evidence (3, 5–8), being female and having a higher educational attainment were significant positive correlates. The stronger support among women may be linked to their heightened health awareness and role as family health guardians, particularly regarding secondhand smoke exposure (10–14).

More importantly, our study extends beyond demographics to reveal critical psychosocial mechanisms. The independent positive associations of family health (FHS-SF) and perceived social support (PSSS) with policy support suggest that interventions strengthening family dynamics and social networks could be highly effective. This finding bridges the gap between literature on social support in smoking cessation (15–20) and public policy endorsement, proposing a novel avenue for mobilizing community-level support. This finding addresses a key gap in the literature by demonstrating that psychosocial determinants, independent of demographics, play a crucial role in shaping policy support. Furthermore, the observed negative association between the global BFI-10 score and policy support suggests that broader personality dispositions may also influence an individual’s receptiveness to public health regulations. While the use of a total score limits specificity, this finding is consistent with studies that have successfully used this concise instrument to link personality to health-related behaviors (9). It resonates with research linking certain personality profiles (e.g., those lower in agreeableness or conscientiousness) to a greater propensity for health-risk behaviors, including smoking (21, 22). This indicates that individuals with such inherent tendencies may be less supportive of policies they perceive as restricting personal freedom.

The clustering of lower support with frequent drinking and unhealthy eating patterns points to a underlying health-risk behavior syndrome. This synergy, supported by existing research on co-occurring risky behaviors (23–33), advocates for integrated public health initiatives that address these behaviors collectively rather than in isolation. Our stratified analyses further provide a blueprint for precision public health. The dominant influence of nicotine dependence among current smokers underscores that policy acceptance in this group is contingent on effective cessation support. Conversely, never-smokers are more responsive to educational and family-protection messaging. These patterns, along with notable gender differences, affirm that tobacco control strategies must be subgroup-specific to resonate with distinct motivations and barriers (34–36).

The interpretations of our findings should be considered in light of several limitations. First, the cross-sectional design precludes causal inference. Second, the non-probability sampling method may limit the generalizability of our results. Third, the model’s R² of 0.13 indicates that a substantial portion of the variance remains unexplained, highlighting the complexity of this phenomenon and the influence of unmeasured factors. Additionally, the use of a total score for the BFI-10, while parsimonious, limits our ability to identify which specific personality traits underlie the observed association with support.