Background China is the world's largest producer and consumer of cigarettes. The status of tobacco as both a contributor to China's economy and a liability for the health of its population may complicate the use of taxes for addressing smoking in the country. Understanding how cigarette prices affect transitions in smoking behaviour in China can increase understanding of how China's high smoking rates can be influenced by tax policy.
Methods In order to estimate the effect of cigarette prices on smoking initiation and cessation in China, we construct pseudo-longitudinal samples for duration analysis using data from the Global Adult Tobacco Survey China 2010. We use the historical variation in prices representative of 4 China regions over a 20-year period to identify the average price effect on the hazards of initiation and cessation while controlling for unobserved fixed and time-varying region characteristics.
Findings We find that initiation rates fall in response to higher prices (with a price elasticity of initiation estimated at −0.95 for men and −1.07 overall).
Conclusions The effect of prices on smoking in China is likely to occur through averting initiation over time. At the population level, cessation behaviour may be less responsive to price increases as the wide range of cigarette prices in China may provide relatively high opportunity for switching to lower priced brands.
- Low/Middle income country
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
China is the world's largest producer and consumer of cigarettes. It derives 7–10% of central government revenue from the tobacco sector,1 while also experiencing considerable health setbacks from the wide consumption of its products. Smoking in China is credited with lowering life expectancy by raising the risk of premature death by 23–28%,2 causes a quarter of all adult male cancers3 and generates morbidity costs that approach $30 billion per year.4
Raising cigarette prices through tobacco taxation has been shown to be a leading policy-based mechanism for reducing smoking, the biggest cause of non-communicable disease worldwide.5 However, the present status of tobacco as both a contributor to China's economy and a liability for the health of its population suggests a complex environment for tobacco tax policy. On the one hand, recent implementation of healthcare reforms in China has emphasised policy efforts to reduce chronic disease rates, strengthening the motivation for tobacco control;6 there has been increased adoption of measures like advertising bans and smoke-free areas legislation, even though these tend to have limited geographic scope and are not always enforced.7 On the other hand, the introduction of more potent policy tools, such as tax increases on cigarettes, has been complicated by the presence of government monopoly in tobacco production—a situation where tobacco production and tobacco taxation are determined within the same agent. The difficulty of implementing tobacco tax policy in this context was illustrated in 2009, when the cigarette excise tax was formally increased, but in practice the increase was absorbed at the production side before reaching retail, lowering its potential to reduce consumption.8 Although understanding of the negative health impact of smoking in China is increasing, and a range of estimates for the price elasticity of cigarette demand in China exists, understanding of the link between prices and the underlying behavioural mechanism through which smoking rates shift over time—the decision to initiate or cease smoking—remains limited.
The contribution of this paper is to evaluate the role of cigarette prices in initiation and cessation of smoking during the life course in the context of China. Employing duration analysis, we find significant effects of prices on reducing smoking initiation. The paper builds on a prior study by Kenkel et al,9 which was the first to introduce the question of how prices relate to life-course transitions in smoking behaviour in China. Though the prior study highlighted the relevance of the question, the authors noted that they were not able to discuss causal inference of the price effect due to lack of region-level variability in prices: “…our analysis was limited by the paucity of time series cigarette price data for different regions in China. Were such data available, it would be possible to estimate much more robust and richer models of initiation and cessation with the CHNS life-course data.” In other words, incorporating region-level prices would represent a step towards causal interpretation of price effects, by allowing regions to serve as within-country counterfactuals while also netting out region-level differences using the within-region variation in prices over time. The present study undertakes this step. We use the variation in prices across and within four regions of China from 1990 to 2010, obtained at the macro level from region-representative retail outlets and thus plausibly exogenous to individual smoking, to model individual transitions in smoking behaviour over time. In contrast to Kenkel et al,9 we find that higher prices deter smoking initiation. We interpret our findings in a policy context that is becoming increasingly relevant as China deepens its consideration of the environmental and health impact of its economic policies.
Data on smoking and individual characteristics were obtained from the Global Adult Tobacco Survey (GATS), conducted in China in 2010. The GATS is a nationally representative survey on non-institutionalised men and women aged 15 and older, which collects data on individual smoking behaviour, demographics and a number of environmental factors related to tobacco use.10 It is not a longitudinal survey, but can be transformed into pseudo-longitudinal format by reconstructing each individual's smoking history from survey questions recording respondents' age of starting or quitting smoking. In the initial full sample of 13 354 respondents in 2010, ∼34% report smoking currently or in the past; among men the rate of ever-smoking is 63% (table 1). On average, initiation occurs at the age of 21, almost entirely among men. Among ever-smokers in the sample, 16% report having quit smoking, at the mean age of 49.
GATS China records respondents' geographic location using six regions: North, Northeast, East, Midsouth, Southwest and Northwest. We used these region identifiers to match respondents' smoking histories to historical series on cigarette prices available for the main city in four of the six regions: Beijing (North region, price series available for 1990–2010), Dalian (Northeast region, 2000–2010), Shanghai (East region, 1993–2010), Guangzho (Midsouth region, 1993–2010). Data on cigarette prices are obtained from the Economist Intelligence Unit (EIU) World Cost of Living survey, which tracks the prices of consumer goods, including cigarettes, in major cities globally. EIU has collected data on the retail prices of a standard international cigarette brand (Marlboro) and a leading local (domestically produced) cigarette brand (selected by EIU) from a representative retail shop and a supermarket in each of the Chinese cities and time periods listed above. We use the local brand retail shop price for this analysis as it is likely to be more representative of prices faced by consumers. Since EIU prices are independently collected at the retail level, the presumption is that their within-region temporal variation reflects market conditions and is exogenous to individual smoking behaviour. EIU price data are not available for cities in two of the six regions of China, Southwest and Northwest. For these regions, we created an imputed price by averaging the prices of the neighbouring regions: Northwest was assigned the price average of Midsouth and North, and Southwest was assigned the price average of Midsouth and Northwest. Sensitivity analyses alternately include and exclude regions with imputed prices. Prices are in Chinese yuan, and corrected for inflation using China's gross domestic product (GDP) deflator with base year 2000. Unimputed prices range from an average of ¥10 in Midsouth to ¥14 in the North region; within-region variability of prices is largest in the North, where prices range from ¥7 to ¥20 over time (table 2).
The final size of the pseudo-longitudinal data sets used in the initiation and cessation analyses is determined, and limited, by the availability of price data. Although we can theoretically derive each individual's smoking behaviour as far back as their year of birth, the full length of the reconstructed smoking history cannot be used in practice because price data are not available to match all years. In this case, the earliest years for which we are able to obtain cigarette prices range from 1990 for the North region to 2000 for the Northeast region. This nearly halves the original sample size for the men's initiation analysis: 3125 of the 6603 men in GATS 2010 are excluded from the initiation models because they report having initiated smoking before the year of the earliest available price in their region (table 3). The average length of follow-up for the men's baseline initiation analysis is 13 years. The follow-up period for each individual in the initiation sample starts at age 15, or at the age when price data become available for that individual's geographical region. It ends at the year of initiation, or is truncated at the year of the survey if initiation does not occur prior to the survey. The data set used for the cessation analysis consists of 4449 men with history of daily smoking, whose average length of pseudo-longitudinal follow-up is 16 years. The duration of follow-up in the cessation sample starts at the year of initiation and ends either at the year of quitting, or at the year of survey, whichever comes earlier. In smoking initiation and cessation samples, the length of individual follow-up can be left-truncated by the length of the available price series and right-truncated by the survey interview.
Demographic control variables include the age of the individual in each year of follow-up (Age), gender (Male), residence (Urban), education category (Education) and wealth index (Wealth). Male and Urban are binary variables indicating, respectively, whether the participant is a male and whether the participant resides in an urban area at the time of the interview. Education is a categorical variable describing the level of education attained by each individual at the time of the interview: no formal education/less than primary, completed primary/less than secondary, completed high school, completed college or higher. Wealth is a categorical variable based on five levels of personal wealth, constructed from GATS responses on possession of core household items following Palipudi et al.11 Urban, Education and Wealth are time-invariant individual factors obtained from responses at the time of the interview—we do not have enough information to reconstruct their variability in preceding periods. Therefore, they are used as proxies for unobserved fixed personal characteristics that may influence smoking behaviour, rather than as standard demographic controls contemporaneous to the behaviour observed in each year. For instance, in this analysis Education and Wealth cannot capture the schooling or income gradients, but may capture individual educational capacity or earning potential; similarly, Urban may not necessarily assign the correct residence type to each year of follow-up, but may nonetheless be useful to account for certain personal characteristics correlated with both urbanicity and smoking (such as social mobility).
Transitions in smoking behaviour are modelled in a duration framework, where the timing of each transition, represented by the hazards of initiation and cessation of individual i in region j at analysis time t (hijt), is evaluated as a function of cigarette price (Pricejt), a vector of time-invariant individual characteristics Xi, region fixed effects (γ0j), calendar year fixed effects (τt) and region-specific linear time trends (γ1jT): 1 2
In this framework, analysis time t does not represent calendar year T but is specific to each individual: in equation (1), t indicates the number of years since age 15, and in equation (2) t indicates the length of the smoking spell defined as the number of years since initiation. Individual attributes in Xi include Age, Urban, Education and Wealth in men-only specifications, plus Male in combined-gender specifications. The baseline models are based on men only, although we present results from combined-gender analyses as well. Separation of the models by gender is warranted because of the wide differences in outcomes between women and men—only 49 initiations occurred among the 6530 qualifying women during the initiation analysis period, compared with 908 initiations among 3478 qualifying men. The low variability in initiation outcomes among women does not allow for statistically sound estimation of separate price effects, but results from combined-gender analyses in comparison to the baseline models for men may be suggestive of the magnitude of women-specific effects relative to men. SEs are clustered by primary sampling unit (PSU), following the design of complex survey data.
The identification of price effects is based on the variation in cigarette prices within regions of China over time. The application of macrolevel (ie, region- level) prices addresses the possibility of confounding from simultaneity between prices and smoking, since smoking choices at the individual level are unlikely to influence region-level pricing. However, confounding of the naïve relationship between prices and smoking can arise from unobserved regional characteristics that simultaneously determine both smoking behaviour and region-level prices. This source of bias has been demonstrated in research on smoking in the USA, where unobserved differences in state-level factors such as antismoking sentiment can drive smoking decisions at the individual level as well as cigarette taxes at the state level, inflating the relationship between the two.12 ,13 We address this issue by employing region fixed effects (γ0j) to account for unobserved region characteristics that do not vary with time, and we control for time-variant region unobservables by including interaction terms between region dummy variables and a linear time trend (γ1jT).14 ,15 The sensitivity of the results to the use of imputed prices for two of the six regions is examined by performing all analyses excluding the two imputed-price regions.
Traditional duration models are estimated by assigning a distribution to the functions in equations 1 and 2, using the assumption that all persons are latent smokers and that all smokers are latent quitters—that is, all individuals who do not experience a smoking transition during the period of observation would eventually undergo the transition if observed long enough. Such an assumption may not be appropriate in the case of smoking, where some individuals will never smoke, or some smokers will never quit, regardless of change in external circumstances. To allow for the possibility that never-smokers or never-quitters exist beyond the follow-up period, we use a modified duration approach that has found frequent application in the analysis of life-course smoking, the split-population duration model.16–21 The split-population duration model first estimates the individual probability of ever experiencing a smoking transition, then weights the hazard function by this probability. We use a lognormal distribution for the hazards of initiation and cessation, and logit for the probabilities of ever smoking or quitting.
Though the price series used in this study provide within-region variation for employing region fixed effects, they are subject to measurement error. First, we are not able to observe respondents' geographic movement over time, and the match between some respondents and region prices may not be fully accurate due to internal migration across regions. As in all retrospective studies, recall bias may introduce error in the outcome variables by producing uncertainty about the correct start and end dates of smoking. These sources of mismeasurement can interfere with the detection of price effects, potentially resulting in underestimation.22–24
The models of smoking initiation in men indicate that cigarette prices played a significant role in initiation decisions in specifications that alternately include and exclude regions with imputed price data (table 4). The price elasticity of initiation in the six-region specification is estimated at −0.95 for men and −1.07 for both genders, increasing to −1.6 and −1.4, respectively, in the four-region specification (table 6). In the baseline specification for men, these results indicate that for a 10% increase in price, the risk of smoking initiation would fall by 9.5%. The role of prices in cessation in China is not statistically significant (tables 5 and 6). Since prior research has shown that the price effect on smoking rates reflects the weighted average of the elasticities of initiation and cessation,12 our findings suggest that the effect of prices on smoking in China may be driven primarily by the initiation component.
Unlike Kenkel et al,9 we estimate that higher cigarette prices reduce smoking initiation in China. Our findings that the effect on initiation dominates that on cessation in China contrasts with findings from similar analyses for Western countries, where prices have been estimated to play a larger role in cessation (for instance, cessation elasticities gave been estimated at 0.62–1.09 for the USA,12 0.46–0.60 for the UK18 and 1.3–1.5 for Spain).20 A possible explanation for the attenuated cessation effect in China may be the presence of an especially wide range of brands in the China cigarette market. Since the price variable used in this study represents the retail price of a single representative brand per region, the estimates are based on the assumption that the temporal variation in the prices of other cigarette brands within regions is proportional to the variation in the representative brand. If the assumption does not hold within regions, the model's inclusion of region-specific time trends may mitigate the issue, but it remains possible that a downward bias would be imparted on the price elasticity of cessation. This is because an increase in the price of the representative brand that is not reflective of an increase for other brands may lead some smokers to switch to other brands, leading to a lower overall response of cessation to prices.
The above explanation for the absence of a cessation effect is consistent with growing evidence of brand downswitching in China. (Since switching between brands can occur in existing smokers only, it would not interfere with estimating the effect that price increases would have on initiation.) Using individual-level longitudinal data, White et al 25 find that a non-trivial portion of smokers in China switch to lower priced cigarettes as a result of a price increase; such brand downtrading was subsequently found to be higher in lower income smokers.26 Huang et al 27 estimated that price-reducing behaviour like increased purchasing in bulk (ie, cartons instead of packs) can lower the average price paid for cigarettes by a smoker in China by up to 15%, potentially offsetting the effect of price increases on smoking. The increasing evidence of price-reducing behaviour occurring at the expense of consumption-reducing behaviour in China is consistent with our findings of low cessation response, and may help explain previous findings of relatively weak price effects on the number of cigarettes consumed among smokers in China.28 ,29 Lance et al,29 for example, find very low price responsiveness of smoking participation in China. Based on the theoretical model developed in DeCicca et al,12 which predicts that short-run participation effects must be smaller than long-run initiation effects, such estimates of low price responsiveness of smoking participation can be aligned with our estimates of significant price responsiveness of smoking initiation.
The implications of this study are timely as China increasingly considers policy options, including tobacco taxes, for improving national health outcomes. Chronic disease, already an established factor for productivity loss and healthcare costs in developed countries, is increasingly affecting China. It crowds out spending, particularly for rural households, potentially widening the urban–rural development gap in the country.2 Smoking may especially contribute to this gap because men, who smoke and experience tobacco-related illnesses disproportionately more than women, serve as a primary labour resource in farming communities; furthermore, the lowest income populations in China have been shown to be most likely to benefit financially from the behavioural changes associated with increased tobacco taxation.30 Increasing longevity in China overall is estimated to yield broad economic benefits: previous projections have suggested that a 1% drop in premature mortality from cardiovascular disease annually over a period of 30 years could generate economic value equivalent to 68% of China's 2010 GDP.31 Tax policy raising cigarette prices and reducing the width of the brand price range in the country can play a role in shaping the trajectory of its future health and economic outcomes.
What this paper adds
Not much is known about how cigarette price increases affect initiation and cessation of smoking in China.
We estimate price elasticities of initiation and cessation of smoking for China using duration analysis on a pseudo-longitudinal data set with historical variation in region-level prices over time.
We find that initiation rates fall in response to higher prices, indicating that the effect of prices on smoking in China is likely to occur through averting initiation over time. Cessation of smoking may be less responsive to price increases as the wide range of cigarette prices in China may provide relatively high opportunity for switching to lower priced brands.
Contributors DK conducted data analysis and drafted the initial manuscript; DK, MJH and FJC contributed to data interpretation, and revised the manuscript.
Disclaimer The findings and conclusions in this manuscript are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention.
Competing interests None declared.
Provenance and peer review Not commissioned; internally peer reviewed.