Beliefs about smoking cigarettes among adolescents in Yunnan Province, China

ABSTRACT

Objective

Adolescence is an important time period in smoking experimentation and initiation. This study examined heterogeneity in key beliefs about smoking intention among Chinese adolescents.

Methods

Survey data came from 951 students (18 and 19 years) in two high schools in Kunming, Yunnan Province, China. The survey questions assessed smoking beliefs and perceptions based on the Theory of Planned Behavior. Regression and latent class analysis were utilized to identify key beliefs (i.e., beliefs that are most influential in smoking intention) and their heterogeneity.

Results

Emotion-related beliefs were reported by both genders, highlighting the role of anxiety and stress in smoking intention. Female and male adolescents had different sets of key beliefs. There were distinct subgroups of beliefs in the sample (two among female adolescents, and three among male adolescents) characterized by disparate patterns of behavioral beliefs, normative beliefs, and control beliefs and smoking status.

Conclusions

Considerable heterogeneity in belief profiles, which differs based on gender and smoking status, furnishes a more complete understanding of smoking intention among Chinese adolescents. Gender-specific anti-smoking interventions can be targeted to the beliefs of specific subgroups of adolescents. Stress management for students could also be a useful tool to prevent smoking uptake.

KEYWORDS:

• China
• key beliefs
• smoking
• adolescent
• Theory of Planned Behavior
• latent class analysis

Introduction

Most smokers, especially Chinese men, experiment with smoking in adolescence. Research is needed to understand beliefs about smoking among Chinese adolescents. This study identifies the most influential beliefs (i.e., key beliefs) underpinning intention to smoke in a sample of Chinese adolescent nonsmokers and smokers, based on the Theory of Planned Behavior (TPB; Ajzen, 1991).

The TPB is a social psychological model of cognitive beliefs and pathways for performing behavior, in which intention is guided by three key antecedents: attitude (positive and negative evaluations of smoking), subjective norm (perceptions of social expectations to smoke), and perceived behavioral control (PBC; perceptions of personal abilities to smoke). In this model, intention and PBC are theorized to predict smoking behavior.

Three types of beliefs form these antecedents. Specifically, attitude is based on behavioral beliefs about the benefits and costs of smoking, and their perceived value; subjective norm is comprised of normative beliefs about the importance of social approval or disapproval of smoking, and motivation to comply; and, PBC is based on control beliefs about the likely occurrence of barriers or facilitators of smoking, and perceptions about their influence (Ajzen, 1991). While a few studies have provided useful information about these beliefs in relation to smoking among Chinese adolescents (Davey et al., 2014; Zhao, White et al., 2018), a large number of beliefs are usually reported – e.g., Davey et al. (2014) reported 39 beliefs. A smaller number of beliefs could help interventionists to develop cost-effective and targeted anti-smoking strategies, although quantitative studies to reduce the number of beliefs are thus far limited. Moreover, as shown in previous studies (Davey et al., 2020; Zhao, White et al., 2019), people may have different patterns of beliefs regarding behavior, although such person-based analyses have not been widely tested in research on beliefs about smoking.

Based on this rationale, the present study aimed to identify the beliefs that are most influential (i.e., key beliefs) in intention to smoke cigarettes in a sample of Chinese adolescents, and to examine heterogeneity in these key beliefs.

Method

Measures

Measures in this study included items about the TPB and demographic characteristics. The TPB survey measured smoking-related perceptions in participants’ daily life situations sometime in the future. Its development was based on an elicitation study of 25 students with mixed smoking experience [for detail, see Davey et al. (2014)].

Each belief is a multiplicative composite of belief-expectancy and belief-value components (Francis et al., 2004), with a theoretical range of −21 to +21 (higher scores in behavioral beliefs, normative beliefs, and control beliefs mean that one is more in favor of smoking, perceives more social pressure to smoke, and feels less control over smoking, respectively). All items were scored on 7-point Likert scales.

Behavioral beliefs were measured with 13 items about the benefits or costs of smoking ([1] strongly disagree to [7] strongly agree) and 13 corresponding items about their desirability ([−3] extremely undesirable to [+3] extremely desirable). Due to the low internal consistency (α = .69), confirmatory factor analyses of the belief items revealed a model with two factors (advantages [α = .92] and disadvantages [α = .97] of smoking), which had a good fit (Table 1).

Table 1. Means and standard deviations of beliefs, Pearson correlations between beliefs and smoking intention, and F statistics of differences by gender.

Code	Belief	M	SD	r	M	SD	r	F
		Female			Male
Behavioral beliefs
BB1	Relaxation	−12.00	10.72	.54**	−7.88	11.99	.53**	30.94**
BB2	Entertainment	−14.64	9.77	.54**	−9.66	11.70	.55**	50.47**
BB3	Feeling alert/attentive	−13.20	11.02	.46**	−7.90	13.07	.47**	45.52**
BB4	Facilitates social interaction	−13.45	10.77	.48**	−8.01	12.50	.48**	51.46**
BB5	Anxiety/distress relief	−12.21	11.19	.48**	−7.00	12.78	.50**	44.34**
BB6	Unhealthy	−2.98	16.75	.03	−3.45	15.06	.02	.20
BB7	Lung damage	−3.08	18.16	.07	−4.04	16.96	.01	.70
BB8	Bothers others	−3.72	16.88	.07	−4.24	15.37	.01	.24
BB9	Harmful to the mind	−3.35	15.55	.06	−3.71	13.90	.03	.14
BB10	Bad smell	−4.00	16.98	.04	−4.55	15.28	.01	.27
BB11	Addiction	−4.26	16.35	.09	−4.23	14.85	.09	.00
BB12	Financial cost	−4.85	16.74	.10*	−4.64	15.43	.05	.04
BB13	Benefits of not smoking	−15.35	8.23	.51**	−12.98	8.95	.51**	17.60**
Normative beliefs
NB1	Close friend’s approval	−11.38	7.59	.46**	−7.95	9.51	.50**	37.76**
NB2	Parents’ approval	−15.14	7.53	.42**	−11.73	9.59	.37**	37.36**
NB3	Best friend’s approval	−12.91	7.75	.53**	−8.87	10.16	.54**	47.54**
NB4	Classmates’ approval	−10.69	7.15	.52**	−7.87	8.84	.51**	29.17**
NB5	Playmates’ approval	−11.43	7.29	.49**	−7.91	9.71	.57**	40.05**
NB6	Mother’s approval	−15.79	7.26	.37**	−12.40	9.58	.33**	37.89**
NB7	Father’s approval	−15.23	7.77	.39**	−11.60	10.12	.38**	38.23**
NB8	Family’s approval	−13.83	7.66	.44**	−10.70	9.22	.40**	32.37**
NB9	Partner’s approval	−12.31	7.94	.48**	−10.56	9.18	.37**	9.76**
NB10	Other students’ approval	−9.88	7.12	.46**	−6.93	8.89	.53**	31.81**
NB11	Teachers’ approval	−13.95	7.32	.39**	−11.06	9.02	.36**	29.42**
Control beliefs
CB1	Encouragement	−2.00	5.72	.43**	−.18	6.98	.53**	19.30**
CB2	Feeling depressed	−1.40	5.73	.52**	.52	7.76	.52**	18.85**
CB3	Encouragement from classmates	−2.51	4.78	.39**	−1.52	6.40	.37**	7.25*
CB4	Bars/KTV/mah-jongg	−1.66	5.63	.51**	.24	7.53	.49**	19.36**
CB5	School pressure	−1.56	5.21	.59**	.02	7.26	.54**	15.00**
CB6	Face-saving	−2.20	4.08	.45**	−1.27	5.57	.34**	8.67**
CB7	Cigarette gifting	−2.14	3.83	.50**	−1.72	4.84	.34**	2.23
CB8	Parental permission at home	−2.44	4.16	.37**	−2.25	5.33	.16**	.38
CB9	Permission at school	−2.43	4.29	.40**	−2.32	6.07	.14**	.10
CB10	Obeying parents	−9.12	11.86	.29**	−5.47	11.75	.33**	22.49**
CB11	Concerns about bad health	−8.28	14.80	.24**	−5.94	13.32	.32**	6.53*
CB12	Tobacco smell	−9.58	13.13	.26**	−7.31	10.60	.42**	8.41**
CB13	Unhappiness	−2.40	8.28	.50**	.09	9.25	.55**	19.01**
CB14	Objections of other people	−8.32	11.21	.27**	−5.96	10.49	.36**	11.01**
CB15	Financial cost	−6.82	10.09	.28**	−4.54	9.80	.40**	12.36**

*p < .05, **p < .01. r = Pearson product-moment correlation coefficient. Given the relatively large sample size, only beliefs highly associated with intention (p < .01) were entered into the analyses.

For behavioral beliefs, confirmatory factor analyses following the TPB were conducted using the 13 belief items resulting in a final model with two factors. The first model with beliefs about advantages of smoking (i.e., relaxation, entertainment, feeling alert/attentive, facilitates social interaction, anxiety/distress relief) and disadvantages of smoking (i.e., unhealthy, lung damage, bothers others, harmful to the mind, bad smell, addiction, financial cost, benefits of not smoking) showed a mediocre fit, ${\chi }_{\left(64\right)}^2$= 847.44, p < .01, CFI/TLI = .94/.93, RMSEA = .11, SRMR = .09, with a modification index suggesting that the beliefs about benefits of not smoking had a strong relationship with the factor regarding advantages of smoking. As a result, the final model containing beliefs about advantages of smoking (Cronbach’s α = .92; relaxation, entertainment, feeling alert/attentive, facilitates social interaction, anxiety/distress relief, benefits of not smoking) and disadvantages of smoking (Cronbach’s α = .97; unhealthy, lung damage, bothers others, harmful to the mind, bad smell, addiction, financial cost) showed a good fit, ${\chi }_{\left(64\right)}^2$= 653.35, p < .01, CFI/TLI = .96/.95, RMSEA = .10, SRMR = .04

Normative beliefs included 11 items ([−3] strongly disagree to [+3] strongly agree; α = .90) assessing beliefs about social referents’ expectations to smoke (injunctive norms), or perceptions of actual behavior (descriptive norms), and 11 corresponding items assessing motivation to comply with each belief ([1] not at all to [7] very much).

Control beliefs included 14 items ([−3] extremely unlikely to [+3] extremely likely; α = .90) measuring the likelihood of agreement with facilitators or barriers of smoking, and 14 corresponding items measuring their perceived importance ([−3] much more difficult to [+3] much easier).

Intention. Three items were assessed on a Likert scale ([1] strongly disagree to [7] strongly agree; α = .93); a higher mean score of these items indicated a greater intention to smoke.

Demographic characteristics included age, ethnicity (Han/non-Han), best friend smoking status (yes/no), maternal and paternal smoking status (yes/no), gender (female/male), and current smoking experience (yes/no).

Procedure

A Chinese version of the survey was prepared with forward and back translation procedures, and pilot testing resulted in minor changes such as rewording of words and sentences. Survey data were collected from 951 students in 2012 from two high schools in Kunming, Yunnan Province, following approval by a university ethics committee and high school administrators, and following informed consent. A mixed cohort of nonsmokers and smokers was included because smoking intention during adolescence is in a transitional stage, with different adolescents at different stages of their smoking careers, rather than at a necessarily manifest stage (Davey & Zhao, 2020). For example, many Chinese adolescents smoke cigarettes but do not regard themselves as smokers (Davey & Zhao, 2020).

Data analysis

Normality and linearity were visually inspected; variables were relatively normally distributed with linear trends between beliefs and intention. Independence of the observations was screened in residual plots and showed that studentized residuals were generally distributed between −2 to +2. As smoking among Chinese adolescents dramatically differs between females and males (Han & Chen, 2015), and differences in beliefs by genders were empirically identified (Table 1), analyses were conducted separately for each gender. Apart from descriptive statistics and stepwise regressions processed in IBM SPSS 25, Mplus 7.4 was used for data analysis (Muthén & Muthén, 1998–2015).

Key beliefs were identified using a methodology by Von Haeften et al. (2001). It began with Pearson’s correlation to identify beliefs significantly correlated with intention, stepwise regression of significant beliefs from Pearson correlation matrices (p < .01) to identify behavioral beliefs, normative beliefs, or control beliefs that independently contributed to intention (Table 2), and a multiple stepwise regression analysis of all significant beliefs from the stepwise analyses, to identify key beliefs which significantly contributed to smoking intention (Table 2).

Table 2. Summary of stepwise regression results identifying key beliefs regarding intention to smoke.

Beliefs predicting intention	Key beliefs of each construct		Final key beliefs regressed on intention
	B		B
	Female	Male	Female	Male
Behavioral beliefs
BB1. Relaxation	.03**		.03**
BB2. Entertainment	.03**	.05**		.02*
BB5. Anxiety/distress relief		.03**		.01*
BB13. Benefits of not smoking	.05**	.07**	.04**	.04**
Normative beliefs
NB3. Best friend’s approval	.05**		.03**
NB4. Classmates’ approval	.03*
NB5. Playmates’ approval		.07**		.03**
NB6. Mother’s approval	−.04**		−.03**
NB8. Family’s approval	.03*
NB9. Partner’s approval	.03**		.03**
NB10. Other students’ approval		.05**		.03**
Control beliefs
CB1. Encouragement		.06**		.04**
CB5. School pressure	.08**	.06**	.08**	.03**
CB7. Cigarette gifting	.07**
CB11. Concerns about bad health		.02**
CB13. Unhappiness	.03**	.05**		.03**

*p < .05, **p < .01. B = unstandardized coefficient.

Next, latent class analysis (LCA) was conducted to identify subgroups (heterogeneity) among female and male participants. Maximum likelihood estimation with robust standard errors (MLR) was used as the estimator (Muthén & Muthén, 1998–2015). Finally, based on the optimal models identified, associations of belief patterns and demographic variables were examined by logistic regression for each subgroup to better understand their composition. Both binary logistic and multinomial logistic regressions were conducted due to the latent class numbers.

Results

Demographic characteristics

The sample has 862 (90.8%) non-current smokers and 87 (9.2%) current smokers. There were almost equal proportions of female (490; 51.6%) and male (459; 48.4%) participants. Most participants (81.8%) were 18 years old and the rest were 19 years old. Overall, most participants had a nonsmoking mother (872; 96.1%), and a nonsmoking best friend (605; 67.1%), but had a father who smoked (603; 66.1%).

Key beliefs undergirding smoking intention

Pearson correlation tests showed that most of the salient beliefs were correlated with intention (Table 1). However, beliefs about disadvantages of smoking (e.g., unhealthy, lung damage) generally were not associated with intention (criterion is p < .01). Stepwise regression was conducted separately for each gender to identify key beliefs (Table 2), showing that three behavioral beliefs (benefits of not smoking, relaxation, entertainment) predicted intention among females, and three beliefs (benefits of not smoking, entertainment, anxiety/distress relief) predicted intention among males. In the normative beliefs model, five beliefs (approval by best friend, mother, classmates, family, partner) were found as the predictors among females, whereas only approval by playmates or by other students were significant predictors among males. In the control beliefs model, three beliefs (school pressure, receiving cigarettes as gifts, unhappiness) predicted intention among females, whereas four beliefs (encouragement, school pressure, unhappiness, concerns about bad health) predicted intention among males. Using these significant beliefs generated above as independent variables in the subsequent stepwise regression model, significant contributors to intention were identified, as shown in Table 2. The final model consisted of six predictors (key beliefs) among females (school pressure, benefits of not smoking, relaxation, best friend’s approval, mother’s approval, partner’s approval), and eight key beliefs among males (benefits of not smoking, encouragement, playmates’ approval, other students’ approval, school pressure, unhappiness, entertainment, anxiety/distress relief).

Smoking intention-related belief profiles

Three models (2-class to 4-class) were conducted for both genders to identify the optimal LCA solution (for detail, see Table 3). Two and three distinct subgroups were discerned among females and males respectively (Figure 1). Specifically, the majority of females (83.1%) were ‘socially conscious girls,’ characterized by stronger anti-smoking beliefs including normative beliefs of social disapproval by best friends, partners, and mothers; whereas a smaller proportion (16.9%) were ‘hesitant girls’ characterized by an overall undecided stance toward smoking, as shown in more neutral belief scores, especially for social approval by friends and mothers. In both subgroups, the strongest pro-smoking belief was ‘school pressure,’ although ‘socially conscious girls’ felt more in control over the decision not to smoke, whereas ‘hesitant girls’ felt less in control due to pressure from school work.

Figure 1. Distinct subgroups and patterns of key beliefs contributing to smoking intention among female and male Chinese adolescents.

Higher scores in behavioral beliefs (BB), normative beliefs (NB), and control beliefs (CB) mean that one is more in favor of smoking, perceives more social pressure to smoke, and feels less control over smoking, respectively. The beliefs displayed on the x-axis were generated from stepwise regression analyses of key beliefs.

Table 3. Fit indices for latent class analysis (LCA).

		Female
	2-class		3-class		4-class
SSBIC	19420.29		19015.79		18824.96
Entropy	.95		.98		.95
Adj. LMR-LRT (p)	748.02 (.016)		416.05 (.184)		207.19 (.466)
Subgroup size (%)
C1	407	83.1%	121	24.7%	315	64.3%
C2	83	16.9%	354	72.2%	95	19.4%
C3			15	3.1%	47	9.6%
C4					33	6.7%
		Male
	2-class		3-class		4-class
SSBIC	25534.85		25003.05		24893.05
Entropy	.93		.95		.91
Adj. LMR-LRT (p)	1016.16 (<.001)		548.46 (<.001)		134.16 (.480)
Subgroup size (%)
C1	233	50.8%	225	49.0%	223	48.6%
C2	226	49.2%	45	9.8%	63	13.7%
C3			189	41.2%	130	28.3%
C4					43	9.4%

SSBIC = Sample size adjusted Bayesian information criteria. Adj. LMR-LRT = Adjusted Lo-Mendell-Rubin Likelihood Ratio Test. The optimal models are shown in bold.

To decide the optimal number of latent classes with similar patterns of smoking-related beliefs, multiple fit indices were considered (Wickrama et al., 2016). First, Sample-size Adjusted Bayesian Information Criterion (SSABIC) statistics were compared, and the model with the smallest SSABIC was considered best. Second, entropy and likelihood ratio were evaluated, as higher entropy suggests better separation between classes, and significant results from the likelihood ratio test indicate the k-class model has a better fit than the k-1-class model; the Lo-Mendell-Rubin Likelihood Ratio Test (LMR-LRT) was used as it is regarded as a robust fit index. Third, a model was regarded as problematic if its smallest subgroup size was too small (proportion < 5.0%; Nylund et al., 2007).

Males reported different smoking-related belief profiles than females, shown in three subgroups: ‘anti-smoking boys’ (49%), characterized by perceived benefits of not smoking, less emphasis on smoking for entertainment to overcome boredom and anxiety, and stronger anti-smoking beliefs; ‘anxious boys’ (41.2%), characterized by behavioral beliefs about smoking to relieve anxiety and mental distress; and ‘stressed-out boys’ (9.8%) who were more in favor of smoking, had more social influence to smoke, and regarded smoking as a solution for pressure from school work and for feeling unhappy (Figure 1).

Predictors of subgroup membership

Based on membership of the 2-class and 3-class models identified above, logistic regression was conducted separately for each sex to identify relationships between demographic information and subgroup membership. ‘Socially conscious girls’ and ‘anti-smoking boys’ were the reference subgroups for each gender as their members generally did not show a pro-smoking propensity in their profiles of beliefs, and they also had the largest proportions.

Among females (Table 4), ‘hesitant girls’ versus ‘socially conscious girls’ were more likely to be current smokers. Similarly, ‘stressed-out boys’ and ‘anxious boys’ were more likely to be current smokers than ‘anti-smoking boys.’ Compared with ‘socially conscious girls,’ ‘hesitant girls’ were more likely to have best friends who smoked. However, in contrast to ‘anti-smoking boys,’ ‘anxious boys’ and ‘stressed-out boys’ were less likely to have best friends who smoked even though they held more favorable beliefs about smoking and were more likely to be current smokers. Another unexpected finding was that ‘anxious boys’ versus ‘anti-smoking boys’ had significantly less Han Chinese, although this statistic was nonsignificant when compared to ‘stressed-out boys’; by contrast, ethnicity and parents’ smoking status did not differ between subgroups of girls.

Table 4. Covariates of latent class membership.

Variable	Female		Male
	Hesitant girls		Stressed-out boys			Anxious boys
	Estimate	S.E.	Estimate	S.E.		Estimate	S.E.
Han ethnicity	−.22	.32	−.40	.65		−.87*	.37
Current smoker	1.07*	.49	1.41**	.21		.65**	.15
Mother who smoked	.49	.54	−.69	.73		.33	.53
Father who smoked	.25	.30	−.71	.49		−.08	.26
Best friend(s) who smoked	1.49**	.29	−1.80**	.50		−1.53**	.26

*p < .05, **p < .01. Reference group: socially conscious girls (female); anti-smoking boys (male). All variables are binary (Yes = 1; No = 0).

Discussion

This study identifies the key beliefs that are important in intention to smoke cigarettes in a sample of Chinese adolescents, and their heterogeneity characterized by disparate patterns of key beliefs and smoking status. The findings also provide a novel person-based perspective for understanding people’s beliefs about smoking.

The differences in key beliefs between genders have important implications for understanding smoking intention. The key normative beliefs reported by females about social approval from best friends, partners, and mothers, especially by those more likely to be current smokers (‘hesitant girls’), in contrast to the key normative beliefs reported by males about playmates and other students, especially among those more likely to be current smokers (‘stressed-out boys’ and ‘anxious boys’), reveal different sources of social influence and pressure to smoke. Based on these findings, assumptions in the literature about the influence of peers and social acquaintances on smoking among Chinese adolescents (Kobus, 2003; Zhu et al., 1996) require further interrogation. Perhaps males, especially ‘stressed-out boys’ and ‘anxious boys,’ smoke more as a social gesture with common friends or a part of social groups, whereas females, especially ‘hesitant girls,’ smoke with close friends in private and hide their smoker identity in public (Davey & Zhao, 2020). The greater number of key control beliefs among males highlights the greater importance of perceptions about the ability to smoke; and differences in these beliefs between subgroups reveal the varying importance of barriers and facilitators of smoking (peer influence was most important among ‘anti-smoking boys,’ feeling unhappy was most important among ‘anxious boys,’ and pressure from school work was most important among ‘stressed-out boys’). Importantly, ‘school pressure’ is a key belief for both genders; and emotion-related beliefs were also reported, for example, behavioral beliefs about smoking for relaxation (females), and smoking for anxiety and stress relief (males), especially in subgroups whose members were more likely to be current smokers. This highlights the importance of anxiety and stress in school as a trigger of smoking in school students, which is most likely connected to the national entrance college exam (Zhao, Young et al., 2018). Perhaps surprisingly, most beliefs about the health effects of smoking were not critical, meaning that school-based health education curriculum in China in its current form (Zhao et al., 2017) will not affect intention to smoke. In contrast, the benefits of not smoking have a strong impact on intention, indicating that positive messages about being a nonsmoker are important. Although ethnicity seems to shape adolescent beliefs about smoking intention, further research with representative and generalizable samples is needed.

That Chinese adolescents apparently holding similar beliefs underpinning intention to smoke actually have considerable heterogeneity in key belief profiles, and also have differences in beliefs based on gender and smoking status, provides a more complete understanding of smoking intention. The findings will also aid the development of interventions targeted to the beliefs of specific subgroups. This is especially important given that previous school-based smoking interventions in China report mixed success (Zhao, Young et al., 2019).

This study provides a useful methodology for understanding key beliefs, with useful implications for anti-smoking interventions in China, the world’s largest tobacco consumer. However, the survey was not conducted nationally, and Yunnan province has the largest tobacco production in China, meaning that the sample is not representative. Subgroup patterns based on data from only two schools may also have amplified the sample’s peculiarities. These limitations are closely rooted in the study’s TPB-based design wherein the specific context is an integral part (Ajzen, 1991). The study is also limited by its social-cognitive focus and self-reported smoking behavior, even though recent theorization of smoking emphasizes the entirety of tobacco and its symbolism in society and in people’s lives in numerous dimensions (e.g., behavior, individual, culture, social processes, society), an integrated and totalizing phenomenon coined as ‘rhizomatic smoking’ (Davey & Zhao, 2021). Since the majority of the sample were nonsmokers, further research can exclusively recruit smokers to determine whether their beliefs regarding smoking possess heterogenous patterns. Nevertheless, our study provides a novel approach for profiling the patterns of beliefs underpinning a health behavior. This is especially important for the TPB research field as the value of underlying beliefs is thus far under-utilized. Based on this method, future interventionists could identify subgroups and develop tailored programs in a more granular approach.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

The author(s) reported there is no funding associated with the work featured in this article.