LITERATURE REVIEW

According to the Chinese Association on Tobacco Control (1), China produces 42% of the tobacco products in the world, consumes 31% of the world's cigarettes (or 1.8 trillion cigarettes per year), and has 320 million smokers. Second-hand smoke affects 600 million people in China, including 200 million children (2). MacKay (3) cited Oxford epidemiologist Richard Peto who predicted that "of all the children alive today in China under the age of 20 years, 50 million of them will eventually die from tobacco." Recent literature suggests that these figures will worsen.

Researchers have identified a number of correlates of current smoking status such as gender, education, marital status, types of employer, Communist Party membership (4, 5), maternal smoking, paternal smoking, low paternal educational level, and a poor self-reported academic record (6, 7, 8). Researchers have also identified correlates of smoking frequency such as marital status, household size, financial assets (4), stress and attitudes (9, 10), curiosity, loneliness (11), etc. However, most of the studies use local cross-sectional data that do not address longitudinal trend and environmental factors beyond the household or peers.

Past research has identified additional interesting characteristics of the Chinese smoking epidemic: (a) Very high prevalence among males and very low prevalence among females. Levy (11) indicated a prevalence of 66.9% among males and 4.9% among females while Yang et al (12) reported a prevalence of 66.0% for males and 3.1% for females, (b) Rapid decline in the average age of initiation (13). The prevalence of smoking among people aged 15–24 has increased, although the overall ever-smoking rate has decreased by 1.8% (12), (c) A relatively positive attitude towards cigarette smoking among youths (6, 7), (d) Presence of the so-called "social smokers" (4), (e) Strong awareness of smoking as a bad habit (88% of adult smokers believe that it is harmful to health), but a persistently very low desire to quit among smokers (14%) (5,) although recent research (12) has observed an increase in the quitting rate from 9.42% in 1994 to 11.5% in 2002. Low awareness of the health risks of smoking was noted, with only 60% in several provinces knowing that smoking can cause lung cancer, and with fewer than 30% knowing that smoking may also cause coronary diseases (12), (f) Finally, in China, smoking serves as a social necessity in building and reinforcing relationships (4) where connections or personal relationship can frequently be the deciding factor in job allocation and promotion (14).

Despite alarms raised by almost every published study on the smoking problem in China, science-based tobacco use prevention and intervention programs remain rare and desperately needed. A few prior interventions such as Project EX (15), "Quit and Win" intervention (16), advice by obstetricians (17), individualized motivation (18), and smoking cessation intervention (19) have produced unimpressive results.

Major gaps in previous literature may be bridged, in part, by this study. Although almost every previous study has called for prevention/intervention programs in China, the limited literature still has not reached a consensus on the factors that should be targeted by these programs. Absence of consensus is partly due to a lack of quality data, and partly to unreliable analyses of the mostly regional case study data, plus small samples with conflicting findings. Our study is designed to identify the individual, household, and community factors associated with the prevalence and frequency of cigarette smoking in China using nationally representative, replicated cross-sectional data from the Carolina Population Center (CPC) at the University of North Carolina at Chapel Hill.

METHODS

Data

Study data were collected by the Carolina Population Research Center in collaboration with the Chinese Center for Disease Control and Prevention. A multistage, random cluster process was used to draw the original sample in the following provinces: Guangxi, Guizhou, Heilongjiang, Henan, Hubei, Hunan, Jiangsu, Liaoning, and Shandong. We attempted to survey over 4,000 households across a 13-year span (1991–2004) at approximately 3-year intervals. Various factors, however, caused the project team to drop and add households each survey year to maintain national representativeness and for other considerations. To be consistent with past literature on smoking research, only those 15 or older are included in the analysis. Our preliminary analysis showed that approximately 3,000 respondents were surveyed each year, while the remaining respondents have missing data for 1, 2, 3, or 4 years. The final sample in this study includes 9,447 individuals 15 years or older in 1991 sample, 8,882 in 1993 sample, 11,241 in 1997 sample, 10,206 in 2000 sample, and 10,330 in 2004 sample. Missing data prompted the research team to run the same analytical models for each survey year, so that the consistency of associations between independent variables and dependent variables across time could be assessed. Although some cases are present in the sample in more than one survey year, the research team believes that the nationally representative data across years do provide an excellent chance to explore longitudinally across a 13-year span the factors associated with cigarette smoking.

The original survey instruments were designed by an interdisciplinary group of social scientists and biomedical researchers from both China and the US with extensive experience in survey research on these topics. Frequent training of data collection teams and systematic cleaning by the principal investigators ensured that excellent data quality.

Measures

Table 1 presents the descriptive statistics of all the variables included in this study. Two dependent variables were included in this study: current cigarette smoking prevalence (1=yes, 0=no) and smoking frequency (number of cigarettes smoked per day).

Table 1 - Means of all variables: 1991–2004.

Full table

Dichotomous or dummy independent variables (1=yes or prevalent, 0=no) included are: gender, urban residence, currently working, having a secondary occupation, and working for state-owned or collective enterprise. Previous literature (4) suggests that gender and types of occupation are associated with smoking prevalence and these variables have been included in models for statistical control. Urban residence was included to test whether the difference in residential location predicts smoking prevalence after the effects of other variables are controlled for statistically.

Continuous and ordinal variables included in this study include: age in years, level of education (0=none, 6=master degree or more), level of physical activity at work (0=very light, 4=very heavy), the number of smokers in a household, and household size. Previous literature has established that education is associated with smoking. Other variables, such as physical activity at work, have rarely been analyzed, although studies on Western populations have documented its association with higher prevalence of smoking. The number of smokers in household was introduced based on previous literature, which suggests that family (especially parents) members' smoking status represents a strong exposure to smoking behavior, and is associated with smoking prevalence of others in the family (6, 7, 8).

Included community-level variables: proxy community economic development indicator (0=mostly dirt roads, 2=mostly paved roads), the local free market cigarette price in Yuan (Chinese currency), community size in thousands of households, and the average daily wage of a typical worker in a community. The road conditions in a community are included in the model because dirt or gravel roads are not suitable for sedan-sized cars. If a community has mostly dirt or gravel roads that would normally indicate lower economic development stage because there are not enough sedan-sized cars that create the demand for paved road. The local free market price of cigarettes is a key variable included in the model as the literature has long established the importance of cigarette price and its connection with the prevalence of smoking. Finally, the average daily wage of a typical worker is the average amount of Yuan a regular worker is expected to earn for each day of work in a community. This variable is included in the model to adjust for the variation of wages across communities and also to see the impact of community prevailing wages on cigarette smoking measures.

Analysis

Because the data are nested in nature, with individuals nested in communities, Hierarchical Linear Modeling (HLM) and its generalized models were used in this study (20). We conducted collinearity diagnostics as one statistical validity check. Linear regression with all the independent variables in the model suggests all tolerance levels are above 0.50, indicating no collinearity among the independent variables. Because no available theory suggests which of the level 1 variables should be set to vary randomly across communities, we first allow all coefficients of individual-level variables to vary randomly at community level to estimate a random component for each variable. Then only the statistically significant random components are left to vary randomly at level 2, while the others are fixed or assumed to be zero (ie, not varying randomly across communities).

FINDINGS

Table 2 presents the odds ratio and level of statistical significance for the Generalized Hierarchical Linear Modeling (GHLM) models with current smoking prevalence as the dependent variable. Only odds ratios are presented because the logit coefficients are only intuitively meaningful, while odds ratios can show not only the direction of the impact or association, but also the extent of the association. They can be interpreted as effect size in logit models that analyze dichotomous dependent variables. If an odds ratio is larger than 1, it would indicate positive impact of the predictor on the probability of being a smoker while a smaller than 1 odds ratio indicates a negative impact.

Table 2 - Generalized hierarchical linear modeling of cigarette smoking prevalence in China: 1991–2004.

Full table

It can be observed that most of the variables show very consistent association, with the same direction of impact, although the odds ratios are different across years for some variables. Age of the respondents shows a consistent positive (P0.01) impact on the probability of smoking prevalence, suggesting that as the respondents age, they are more likely to smoke. Consistent with previous literature, the large odds ratio of male gender suggests that after adjusting the effects of all other variables in the model, males are much more likely to smoke than females in China. Notably, across all five models between 1991 and 2004, the level of education is not a statistically significant predictor of smoking prevalence, contradicting research findings by Gong et al. (5). Moreover, the level of physical activity at work is only a statistically significant predictor of the prevalence of smoking in the 1997 sample, contrary to the studies of smoking behaviors in Western countries, which showed that higher physical activity or labor-intensive types of jobs are associated with higher probability of smoking. Table 2 also shows that urban residence does not have a consistent association with higher probability of smoking, suggesting that although at times urban residential location tends to predict higher probability of smoking, the distinction is not necessarily always clear. Rural residents tend to be as likely to be smokers, as indicated by the non-significant effects of urban residence for 1991, 2000, and 2004 samples.

Not surprisingly, either presently working or having a secondary occupation is associated with a much higher probability of smoking, suggesting that either the occupation or the stress associated with it is a strong predictor of smoking in the 13-year span between 1991 and 2004 in China. Interestingly, working for state-owned or collective enterprises is not a statistically significant predictor of the prevalence of smoking. This result is contrary to previous research (4,) which suggests that working for state-owned or collective enterprises has stronger requirement for building personal connections, thus leading to higher smoking prevalence among those workers because smoking and "cigarette offering" are cultural practices in Chinese society, used to build relationships among friends, colleagues, and business contacts. Finally, the number of smokers in a household is associated with higher probability of being a smoker while household size is negatively associated with smoking prevalence, suggesting that living with other smokers in the household increases the probability of smoking, while living in larger families tend to reduce the probability of becoming a smoker.

Most of the community variables included in the model show no significant impact on the probability of smoking. But the community size shows a positive impact on smoking prevalence in the 1991 sample (P0.10).

Table 3 presents the GHLM modeling of smoking frequency for the samples between 1991 and 2004. Because the number of cigarettes smoked is a count variable, we chose the Poisson distribution setting in HLM software when analyzing this dependent variable. To provide better comparison across different models in this study and across different studies in the literature, effect sizes in terms of Cohen's correlation r are presented instead of unstandardized coefficients (21).

Table 3 - Generalized hierarchical linear modeling of cigarette smoking frequency in China: 1991–2004.

Full table

Consistent with Table 2, age and male gender are associated with the frequence of smoking, but the extent of impact is, however, very different in that males smoked much more than females after the effects of all other factors in the model were adjusted for (P0.01). Also consistent with Table 2 is that the level of education and urban residence do not have any impact on smoking frequency. Table 3 further shows that those who were working or had a secondary job tend to smoke more than others. Similarly, household size has a negative impact on smoking frequency.

Finally, contrary to the results in Table 2, the community variables in Table 3 show some sporadic statistically significant results. For example, in the 1991 and 1993 sample, the free market cigarette price in a community has a negative association with smoking frequency, suggesting that the higher cigarette price is associated with lower frequency of cigarette smoking. Unfortunately, this association is not present in samples of later years.

Limitations of this preliminary look at data from as long as 16 years ago suggest strategies for the future. We were unfortunately unable to distinguish between those who never smoked and those who had quit smoking – perhaps a serious misclassification problem. Futher, there are limitations associated with regionally observed findings – heterogeneity in geography, economy, industry, and degree of urbanization among the nine chosen provinces – as the basis for generalizations about all of China. Repeated interviews with about 3,000 people might have generated information about changes over time, but also risk problems with correlated responses. The mix in this study increases the temptation to interpret cross-sectional data as if it were from a longitudinal panel, although both effects, age and cohort, are present and difficult to disentangle.

CONCLUSION

The Chinese Congress approved the Framework Convention on Tobacco Control on August 28, 2005, formally declaring the country's intention to enforce tobacco control nationwide. Several tentative conclusions of our study may provide insights needed to start combating this serious public health problem in China. First, future prevention or intervention programs should aggressively target males because the much larger probability of males being smokers in China. The positive impact of age suggests that the prevention programs should start when people are still young to prevent initiation. The average age of smoking initiation is declining, from 28 to 25 (13) in the 1990s and from 22.4 to 19.7 in recent years (1), suggesting that seniors in high school might be the best target population for prevention programs.

Second, the non-significant effects of education and urban residence on both the prevalence and the frequency of cigarette smoking suggest that smoking is a widespread problem in China across education levels and the urban–rural regions. The consistent positive impact of occupational variables suggest that reducing stress or teaching stress management techniques should be one of the key focuses in future prevention or intervention programs to combat the widespread smoking problem in China.

Third, the conflicting effects of the number of smokers in a household and household size indicate that future health promotion programs should have a family focus to reduce smokers in the families and to seek stronger support from family members. The one child policy practiced in the past two decades means that most of the teenagers in urban China do not have any competition from any brother(s) or sister(s). Because parents wish to do their best for the child and because of strong traditional family values in China, the investment of time and resources by parents can be easy to obtain for their children in Chinese families. This could be an attractive opportunity for family-based interventions to prevent teenagers from starting to smoke. The literature has long established the association between parental smoking status and the onset of smoking among their children (6).

Finally, the non-significant impact of community variables on smoking prevalence and smoking frequency suggests that regardless of the economic development stage, the daily wage of a typical worker in a community, the size of the community or the free market price of local cigarettes, the prevalence of smoking is the same after other variables such as gender, occupation, and household factors are adjusted. We argue, therefore, that these findings do not imply that policy intervention through higher taxation of tobacco products will not be effective to reduce smoking in China. A study by Wang (22) has offered early evidence that expenses on cigarettes have produced significant impact on other household expenditures. In addition, a study by Yang et al. (12) suggests that if there is a financial incentive to be gained through not smoking, some people in China are willing to quit. Our study argues that raising cigarette prices through taxation may still be a very effective intervention for the Chinese government along with other measures such as the establishment of a minimum age cigarette purchase law, stricter enforcement of no-smoking laws in public places (12), and an expansion of the bans on cigarette advertisements (23).

References

1. CATC(Chinese Association on Tobacco Control). http://www.cash-prc.com.
2. Best Foundation. Secondhand Smoke Affects 600 Million in China Retrieved April 15, 2006, from http://www.jointogether.org/news/headlines/inthenews/2006/se
   condhand-smoke-affects-600.html.
3. MacKay J. China's Tobacco Wars. Retrieved 22 April 2006, from http://multinationalmonitor.org/hyper/issues/1992/01/mm0192_06.html.
4. Pan Z. Socioeconomic predictors of smoking and smoking frequency in urban China: evidence of smoking as a social function. Health Promotion Int. 1992;19(3):309–315. | Article |
5. Gong YL, Koplan JP, Feng W, Chen CH, Zheng P, Harris JR. Cigarette smoking in China: prevalence, characteristics, and attitudes in Minhang district. J Am Med Assoc. 1995;274:1232–1235. | Article | ChemPort |
6. Hesketh T, Ding QJ, Tomkins A. Smoking among youths in China. Am J Public Health. 2001;91(10):1653–1655. | PubMed | ChemPort |
7. Li X, Fang X, Stanton B. Cigarette smoking among schoolboys in Beijing, China. 1999. J Adolescence. 1999;22(5):621–625. | Article | ChemPort |
8. Zhang L, Wang W, Zhao Q, Vartiainen E. Psychosocial predictors of smoking among secondary school students in Henan, China. Health Educ Res. 2000;15(4):415–423. | Article | PubMed | ChemPort |
9. Sun W, Shun J. Smoking behavior among different socioeconomic groups in the work place in the People's Republic of China. Health Promotion Int. 1995;10(4):261–266. | Article |
10. Xiang H, Wang Z, Stallones L, Yu S, Gimbel HW, Yang P. Cigarette smoking among medical college students in Wuhan, People's Republic of China. Prev Med. 1999;29(3):210–215. | Article | PubMed | ChemPort |
11. Levy D. The Role of Public Policies in Reducing Smoking and Deaths Caused by Smoking in China: Results from the China Tobacco Policy Simulation Model. New York, NY: World Health Organization; 2006.
12. Yang Y, Jiang Y, Yang X, Deng Y, Li J, Guo Z, et al. Analysis on main factors for successful quitting: study on the one-year follow-up for Chinese "Quit and Win" in 2002. Wei Sheng Yan Jiu. 2004;33(4):478–480. | PubMed |
13. Yang G, Fan L, Tan J, Qi G, Zhang Y, Samet JM, et al. Smoking in China: findings of the 1996 National Prevalence Survey. JAMA. 1999;282(13):1247–1253. | Article | PubMed | ChemPort |
14. Bian Y. Guanxi and the allocation of urban jobs in China. The China Quarterly. 1994;140:971–999.
15. Zheng H, Sussman S, Chen X, Wang Y, Xia J, Gong J, et al. Project EX – a teen smoking cessation initial study in Wuhan, China. Addict Behav. 2004;29(9):1725–1733. | Article | PubMed |
16. Yang Y, Jiang Y, Yang X, Deng Y, Li J, Guo Z, et al. Analysis on main factors for successful quitting: study on the one-year follow-up for Chinese "Quit and Win" in 2002 (Chinese). 2004. Wei Sheng Yan Jiu. 2004;33(4):478–480. | PubMed |
17. Loke AY, Lam TH. A randomized controlled trial of the simple advice given by obstetricians in Guangzhou, China, to non-smoking pregnant women to help their husbands quit smoking. Patient Educ Counseling. 2005;59(1):31–37. | Article |
18. Chan SS, Lam TH, Salili F, Leung GM, Wong DC, Botelho RJ, et al. A randomized controlled trial of an individualized motivational intervention on smoking cessation for parents of sick children: a pilot study. 2005. Appl Nursing Res. 2005;18(3):178–181. | Article |
19. Abdullah AS, Mak YW, Loke AY, Lam TH. Smoking cessation intervention in parents of young children: a randomised controlled trial. Addiction. 2005;100(11):1731–1740. | Article | PubMed |
20. Bryk AS, Raudenbush SW. Hierarchical Linear Models: Applications and Data Analysis Methods. Newbury Park: Sage; 1992.
21. Cohen J. Statistical Power Analysis for the Behavioral Sciences. Hillsdale, NJ: Lawrence Erlbaum Associates; 1988.
22. Wang HK. The impact of tobacco expenditure on household consumption patterns in rural China. Soc Sci Med. 2006;62(6):1414–1426. | Article | PubMed |
23. Fielding R, Chee YY, Choi KM, Chu TK, Kato K, Lam SK, et al. Declines in tobacco brand recognition and ever-smoking rates among young children following restrictions on tobacco advertisements in Hong Kong. J Public Health. 2004;26(1):24–30. | Article | ChemPort |

About the Authors

Zhenfeng Pan, Ph.D., is an Associate Research Scientist at the Louisville Center of Pacific Institute for Research and Evaluation, with headquarter at: 11720 Beltsville Drive, Calverton, MD 20705, USA. E-mail: terry3232@pire.org

Dongsheng Hu, Ph.D. is a Professor of Epidemiology, Dean of College of Public Health, Zhengzhou University, Zhengzhou, Henan, P.R. China. E-mail: dongsheng-hu@zzu.edu.cn.