Introduction
Cigarette and alcohol policies in China face a dilemma [25]. A large number of studies have shown that cigarette consumption and excessive alcohol consumption are very harmful to health. Cigarettes account for more than 13% of male deaths in China [32, 37]. Excessive alcohol consumption has been linked to liver cancer, female choriocarcinoma mortality, increased admission rates to psychiatric hospitals, and traffic accidents in China [15, 21, 22, 34]. Furthermore, addiction to cigarettes or alcohol can crowd out other household expenditures, having a negative impact on living standards, particularly for low-income households [24]. Therefore, the Chinese government is trying to adopt policies to reduce alcohol and tobacco consumption, especially the consumption of cigarettes. As shown in Table 1, as per capita income has increased in rural China, the expenditure share of both cigarettes and alcohol have decreased continuously from 1994 to 2003. The expenditure share of cigarettes dropped from 2.51% in 1993 to 1.18% in 2003, with per capita consumption decreasing from 24.6 packs to 21.6 packs although there are some fluctuations during this period. The expenditure share of alcohol decreased from 1.35% in 1994 to 0.97% in 2003; however, per capita consumption increased from 6.5 to 7.8 kg.
Table 1
Per capita cigarette and alcohol consumption in rural China (1994–2003)
Year | Net disposable income (current Yuan) | Average household size (persons) | House area per capita (m2) | Cropland area per capita (mu) | Fraction of population with > primary education (%) | Total expenditure (Yuan) | Cigarettes | Alcohol | ||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Consumption (packs) | Average price (Yuan/pack) | Expenditure (Yuan) | Expenditure share (%) | Consumption (kg) | Average price (Yuan/kg) | Expenditure (Yuan) | Expenditure share (%) | |||||||
1994 | 1,220.98 | 4.73 | 19.52 | 2.53 | 47.4 | 1,016.81 | 24.60 | 1.04 | 25.51 | 2.51 | 6.53 | 2.10 | 13.74 | 1.35 |
1995 | 1,577.74 | 4.65 | 20.32 | 2.54 | 48.9 | 1,310.36 | 27.39 | 1.07 | 29.43 | 2.25 | 7.11 | 2.36 | 16.81 | 1.28 |
1996 | 1,926.07 | 4.58 | 20.59 | 2.69 | 51.92 | 1,572.08 | 25.00 | 1.13 | 28.37 | 1.80 | 7.13 | 2.51 | 17.89 | 1.14 |
1997 | 2,090.13 | 4.52 | 21.34 | 2.42 | 53.42 | 1,617.15 | 23.64 | 1.13 | 26.80 | 1.66 | 6.98 | 2.56 | 17.84 | 1.10 |
1998 | 2,161.98 | 4.45 | 22.18 | 2.40 | 54.8 | 1,590.33 | 23.09 | 1.12 | 25.94 | 1.63 | 6.98 | 2.53 | 17.68 | 1.11 |
1999 | 2,210.34 | 4.40 | 22.96 | 2.42 | 56.18 | 1,577.42 | 23.86 | 1.09 | 25.97 | 1.65 | 7.02 | 2.50 | 17.52 | 1.11 |
2000 | 2,253.42 | 4.34 | 23.61 | 2.32 | 58.49 | 1,670.13 | 23.28 | 1.06 | 24.71 | 1.48 | 7.10 | 2.47 | 17.52 | 1.05 |
2001 | 2,366.40 | 4.29 | 24.33 | 2.33 | 59.77 | 1,741.09 | 22.75 | 1.06 | 24.07 | 1.38 | 7.50 | 2.45 | 18.37 | 1.05 |
2002 | 2,475.63 | 4.27 | 25.11 | 2.34 | 60.48 | 1,834.31 | 22.10 | 1.06 | 23.43 | 1.28 | 7.70 | 2.43 | 18.73 | 1.02 |
2003 | 2,622.20 | 4.23 | 26.18 | 2.30 | 61.38 | 1,943.30 | 21.60 | 1.06 | 22.88 | 1.18 | 7.80 | 2.43 | 18.94 | 0.97 |
On the other hand, cigarettes and alcohol are important industries in China. This is particularly true for some less developed regions in China, such as Guizhou Province and Yunnan Province, where cigarettes and alcohol are important income sources for farmers and major revenue sources for the government [23‐25]. In 2004, China collected 210 billion Yuan of taxes from the cigarette industry,1 which represents nearly 8% of the Chinese government’s total annual revenue.
Anzeige
A number of studies on the demand in China for alcohol [43, 47] and cigarettes [10, 26, 33] have been conducted. However, with the exception of Fan et al. [17], little attention has been paid to interdependencies between alcohol and cigarette consumption in China. The objective of this paper is to analyze the interactions between cigarette and alcohol consumption in rural China, using panel data covering 10 years (1994–2003) for rural areas of 26 Chinese provinces. In China, most people (59.6% of the total population in 2003) still live in rural areas,2 although urbanization is occurring rapidly.
The issue of interdependencies has important implications from both econometric and policy perspectives. From an econometric perspective, most existing empirical studies of cigarette demand in China exclude the price of alcohol, and vice versa. If there are interdependencies, the results from these studies will be biased [16].
From a policy perspective, taxes are often recommended as a tool to reduce alcohol and cigarette consumption. Becker and Murphy [9], using a theory of rational addiction, suggest that taxes may be a very useful tool with which to reduce harmful addictions, and a number of empirical studies have found that consumption of alcohol and cigarettes is responsive to price [4‐7, 10, 14, 35, 36]. If cigarettes and alcohol are complements, taxing one will reduce the consumption of both and thus achieve a double public health dividend. However, if they are substitutes, taxing one will induce consumers to increase consumption of the other, offsetting the public health benefits of the tax.
Studies in countries other than China have yielded conflicting results regarding the relationship between cigarettes and alcohol. Using panel data for individual states in the United States from 1959 to 1982, Goel and Morey [19] found that cigarettes and liquor are substitutes in consumption, while Jones [27], using aggregate quarterly expenditure data for the United Kingdom from 1964 to 1983, found that cigarettes are a complement to all types of alcoholic beverages. Decker and Schwartz [16], using individual-level data for the United States from 1985 to 1993, found an asymmetry in Marshallian cross-price elasticities of demand: higher alcohol prices decrease both alcohol consumption and smoking, while higher cigarette prices tend to decrease smoking but increase drinking. Busch et al. [13] reached the same conclusion using individual-level data for the United States from 1995 to 2001.
Anzeige
Medical and psychological research indicates that co-occurrence rates of alcohol and cigarette addiction are very high [8], which may be partly a result of genetic factors [39]. In a study of light smokers, King and Epstein [31] found that alcohol dose-dependency increases the urge to smoke.
To date, with some exceptions [19], studies of the interactions between cigarette and alcohol consumption do not consider the biological and psychological characteristics of addiction, such as dependence, reinforcement and tolerance, that imply that current consumption may be affected by past consumption. This paper constructs a dynamic model of consumption involving a theory of habit persistence in order to account for these factors.
A habit persistence model
Studies of the demand for cigarettes or alcohol typically begin with a Marshallian demand function that depends on income, prices, and a vector of household characteristics. Advertising is sometimes also included in the demand function [7, 41], although we lack data on advertising and thus cannot include it here. For an addictive product such as alcohol or cigarettes, dependence, reinforcement and tolerance are also important factors. Dependence, also known as withdrawal effects, means that consumption of a drug takes precedence over consumption of other goods. Reinforcement means that current consumption of a product increases future consumption of that product, and tolerance means that there is a progressively decreasing response to consumption of a drug [29, 45].
Dependence, reinforcement and tolerance imply that current consumption of cigarettes and alcohol may depend on the past consumption path, suggesting a habit persistence model of consumption. Following the habit persistence model [12, 42], per capita consumption of cigarettes or alcohol in province i of rural China at time t (Y
it
) is assumed to be determined by past consumption:where P
it–k is a vector of prices at time t–k, PCIit–k is per capita income, Z
it–k is a vector of household characteristics, and ρ
k
is a parameter reflecting the impact of past consumption on current consumption. Suppose ρ
k
is geometrically declining, i.e., ρ
k
= (1 − λ)λ
k
, 0 < λ < 1, where k = 0, 1, 2, …. Then Eq. 1 can be rewritten as
$$ \ln Y_{it} = \sum\limits_{k = 0}^{\infty } {\rho {}_{k}f\left( {P_{it - k} ,{\text{PCI}}_{it - k,} Z_{it - k} } \right)},$$
(1)
$$ \ln Y_{it} = \left( {1 - \lambda } \right)\sum\limits_{k = 0}^{\infty } {\lambda^{k} f\left( {P_{it - k} ,{\text{PCI}}_{it - k,} Z_{it - k} } \right)}.$$
(2)
Taking a one-period lag of Eq. 2, multiplying that one-period lag by λ, and substituting the resulting expression into the right-hand side of Eq. 2 yieldswhich is a dynamic model of consumption that can be estimated.
$$ \ln Y_{it} = \lambda \ln Y_{it - 1} + \left( {1 - \lambda } \right)f\left( {P_{it} ,{\text{PCI}}_{it,} Z_{it} } \right) ,$$
(3)
Of the current studies on interdependencies between alcohol and cigarette consumption, only Goel and Morey [19] use a dynamic equation.3 Other studies do not employ a dynamic model, which for some studies that used individual-level survey data can be explained by the absence of a panel component in those surveys. Since the present study uses provincial-level panel data for 26 provinces, a dynamic model is feasible.
In the Becker and Murphy [9] model of rational addiction, current-period consumption depends not only on lagged consumption but also on expected future consumption because rational addicts consider how much they are planning to consume in the future when making current consumption decisions. In our case we have panel data for 10 years (1994–2003). One of those years is used up by the lagged consumption term in Eq. 3 and another by the generalized method of moments (GMM) estimation procedure we employ [2, 3, 11]. Adding a future consumption variable would use up two additional years (one for the variable itself and one for the GMM estimation procedure), leaving our panel component rather short. We therefore do not include future consumption in the demand equation to be estimated.
Empirical model and data
The price vector P
it
in the model to be estimated includes prices of cigarettes, alcohol, grains, and meat. The price of grains is included because grains are both an important income source and an important food source in rural China. The price of meat is included for similar reasons—the expenditure share of meat has increased in recent years and now is close to that of grains [49]. The vector of household characteristics Z
it
in the empirical model includes average household size, house area (in square meters) per capita, average cropland area (in mu) per capita,4 and the fraction of the adult population with more than a primary school education.5 The empirical model also includes a time-trend variable to capture other factors that may be affecting cigarette and alcohol consumption over time.
Anzeige
The function f(P
it, PCIit,
Z
it) is assumed to be linear in the logs of all its arguments:
where, α*, the \( \beta_{j}^{*},\)
η*, the \( \gamma_{k}^{*},\) and the σ* are parameters.
$$ f\left( {P_{it} ,{\text{PCI}}_{it,} Z_{it} } \right) = \alpha^{*} + \sum\nolimits_{j} {\beta_{j}^{*} \ln P_{jit} } + \eta^{*} \ln {\text{PCI}}_{it} + \sum\nolimits_{k} {\gamma_{k}^{*} \ln Z_{kit} } + \sigma^{*} t,$$
(4)
Let α = (1 − λ)α*, \( \beta_{j} = \left( {1 - \lambda } \right)\beta_{j}^{*},\)
η = (1 − λ)η*, \( \gamma_{k} = \left( {1 - \lambda } \right)\gamma_{k}^{*},\) and σ = (1 − λ)σ*. Substituting Eq. 4 into Eq. 3, and appending a term to reflect unobserved heterogeneity among provinces (v
i) as well as an error term (ɛ
it), yields the model to be estimated:
$$ \ln Y_{it} = \alpha + \lambda \ln Y_{it - 1} + \sum\nolimits_{j} {\beta_{j} \ln P_{jit} } + \eta \ln PCI_{it} + \sum\nolimits_{k} {\gamma_{k} \ln Z_{kit} } + \sigma_{1} t + \nu_{i} + \varepsilon_{it} . $$
(5)
The unobserved heterogeneity term is assumed to be fixed over time.
Clearly, the error term ɛ
it
is correlated with ln Y
it+1, ln Y
it+2, etc. in Eq. 5. Therefore, a fixed-effects model, a random-effects model, and the maximum likelihood estimator usually applied to static panel data models are all inconsistent [1, 2]. Taking the first difference of Eq. 5, Anderson and Hsiao [1] suggest a consistent estimator for this equation using \( \Updelta \ln Y_{it - 2} \)
\(( = \ln Y_{it - 2} - \ln Y_{it - 3}) \) as an instrumental variable for Δ ln Y
it−1. However, Arellano and Bond [2] and Judson and Owen [30] point out that the Anderson–Hsiao estimator is inefficient because it does not take into account all the available moment restrictions, and the performance is very poor when the sample size is small. Arellano and Bond [2] suggest a GMM estimator that is more efficient because it uses additional instruments whose validity is based on the orthogonality between lagged values of Y
it
and the error term ɛ
it
. This method is further extended by Arellano and Bover [3] and developed fully by Blundell and Bond [11]. Roodman [44] provides a pedagogic instruction to the practice of linear GMM with STATA. GMM methods for dynamic panel data are used to estimate the econometric model in the present study.
Anzeige
The panel dataset consists of data for 10 years (1994–2003) for rural areas of 26 Chinese provinces, with data being at the provincial level. Data are from the National Statistics Bureau of China (NSBC). The dataset begins in 1994 in order to avoid prior years in which prices were significantly distorted by government regulations. Even though China began food policy reforms in the late 1970s, price regulations were not abandoned until 1993 [38]. The descriptive statistics of the data can be seen in Table 1.
Prices for 1994 are derived from Rural Household Survey Statistics (RHSS), a NSBC publication, dividing total expenditure in each group by the total quantity consumed. Starting with the 1994 unit values, we use the provincial consumer price indices (CPI) for cigarettes, alcohol, grains, and meat for 1995–2003 to compute prices for those years. CPIs are obtained from the China Statistical Yearbook of Prices and Urban Household Survey (various editions), published by NSBC (http://www.stats.gov.cn/english/statisticaldata/yearlydata/). Data on the consumption of alcohol (measured in kilograms per capita, including spirits, beer and wine) and cigarettes (measured as packages per capita), as well as the household characteristic variables in Eq. 5, are from RHSS (various editions), which covers 27 provinces; Tibet is excluded from our analyses because of missing data, leaving 26 provinces. Nominal values are converted to real terms using the provincial overall rural CPI, with all prices expressed in 1994 Yuan.
The optimal matrix of instruments in GMM estimation depends on whether the explanatory variables are predetermined or strictly exogenous. We assume that prices and per capita income are predetermined while the household characteristic variables (as well as a time trend) are strictly exogenous.
Results and discussion
Table 2 presents the estimation results for cigarette and alcohol demand in rural China, using the fixed-effects model and the first-difference model. The results show that the first-difference model fits the data very poorly. The coefficients of lagged consumption for both cigarettes and alcohol are negative in the first-difference model, which does not make sense. Though the fixed-effects model is not consistent with a dynamic panel dataset, the results can be used for comparison with GMM methods such as the Arellano–Bond [2] and Arellano–Bover [3] methods.
Table 2
Estimation results I
Fixed-effects model | First-difference model | First-difference/SURE modela
| ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
ln (cigarette consumption) | ln (alcohol consumption) | ln (cigarette consumption) | ln (alcohol consumption) | ln (cigarette consumption) | ln (alcohol consumption) | |||||||
Coefficient |
t
| Coefficient |
t
| Coefficient |
t
| Coefficient |
t
| Coefficient |
t
| Coefficient |
t
| |
ln (lagged consumption) | 0.215 | 3.45*** | 0.121 | 3.27*** | −0.353 | −6.63*** | −0.055 | −1.77* | −0.349 | −6.99*** | −0.028 | −0.96 |
ln (cigarette price) | 0.147 | 1.11 | 0.043 | 0.20 | −0.105 | -0.54 | 0.255 | 0.89 | −0.105 | −0.56 | 0.198 | 0.71 |
ln (alcohol price) | −0.743 | −3.51*** | −0.772 | −2.36** | −1.401 | −5.57*** | −0.329 | −0.91 | −1.399 | −5.72*** | −0.308 | −0.88 |
ln (grain price) | −0.022 | −0.19 | 0.487 | 2.66*** | −0.049 | −0.43 | 0.266 | 1.61 | −0.048 | −0.44 | 0.298 | 1.86 |
ln (meat price) | −0.058 | −0.47 | 0.130 | 0.65 | −0.115 | −0.95 | 0.149 | 0.84 | −0.116 | −0.98 | 0.119 | 0.69 |
ln (per capita income) | −0.138 | −1.07 | 0.272 | 1.31 | 0.347 | 2.50** | 0.002 | 0.01 | 0.345 | 2.56 | 0.039 | 0.20 |
ln (average household size) | −1.806 | −3.83*** | −1.542 | −2.13** | −1.280 | −2.22** | −0.075 | −0.09 | −1.280 | −2.28 | −0.160 | −0.20 |
ln (house area per capita) | −0.148 | −1.44 | 0.026 | 0.15 | −0.077 | −0.70 | 0.218 | 1.31 | −0.077 | −0.71 | 0.177 | 1.10 |
ln (cropland area per capita) | −0.177 | −2.56*** | 0.035 | 0.33 | −0.065 | −1.05 | 0.003 | 0.03 | −0.065 | −1.08 | 0.007 | 0.08 |
ln (fraction of population with > primary education) | 0.055 | 0.20 | 1.199 | 2.71*** | 0.429 | 1.60 | 1.140 | 2.91*** | 0.431 | 1.66* | 1.216 | 3.19*** |
Time | −0.037 | −3.87*** | −0.022 | −1.50 | −0.079 | −6.13*** | −0.004 | −0.24 | −0.079 | −6.29*** | −0.007 | −0.36 |
Intercept | 7.490 | 5.50*** | 2.829 | 1.34 | ||||||||
Significance test for the model | F(11,197) = 12.06*** | F(11,197) = 7.37*** | F(10,197) = 110.40*** | F(10,197) = 2.36** | Chi(10) = 110.40*** | Chi(10) = 22.51** | ||||||
Anzeige
Using GMM, Table 3 shows the estimation results for cigarette and alcohol demand in rural China. We report the estimation results for both the Arellano–Bond method and the Arellano–Bover method. The Arellano–Bover method proposes an orthogonal deviations transformation, subtracting the mean of all available future observations, rather than first-order differences (subtracting the previous observation as in Arellano–Bond method). In this way the lagged observations of a variable do not enter the formula for the transformation; they remain orthogonal to the transformed errors and are valid as instruments [3]. However, like differencing, taking orthogonal deviations still removes fixed effects.
Table 3
Estimation results II
ln (cigarette consumption) | ln (alcohol consumption) | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Arellano–Bond one-step | Arellano–Bover one-step | Arellano–Bover two-step | Arellano–Bond one-step | Arellano–Bover one-step | Arellano–Bover two-step | |||||||||
Coefficient | t ratio | Coefficient | t ratio | Coefficient | t ratio | Corrected t ratio | Coefficient | t ratio | Coefficient | t ratio | Coefficient | t ratio | Corrected t ratio | |
ln (lagged consumption) | 0.207 | 2.45** | 0.385 | 3.12*** | 0.426 | 8.13*** | 3.16*** | 0.054 | 1.16 | 0.148 | 2.74*** | 0.146 | 5.31*** | 1.82* |
ln (cigarette price) | 0.115 | 0.56 | 0.093 | 0.69 | 0.089 | 1.32 | 0.63 | 0.072 | 0.28 | 0.005 | 0.02 | 0.049 | 0.31 | 0.15 |
ln (alcohol price) | −1.366 | −4.38*** | −0.590 | −2.55*** | −0.619 | −5.56*** | −2.33** | −0.782 | −2.29** | −0.728 | −2.23** | −1.526 | −4.20*** | −1.94* |
ln (grain price) | 0.014 | 0.11 | 0.016 | 0.14 | 0.017 | 0.27 | 0.11 | 0.656 | 4.24*** | 0.510 | 2.81*** | 0.635 | 5.83*** | 3.02*** |
ln (meat price) | −0.159 | −1.26 | −0.063 | −0.52 | −0.068 | −1.48 | −0.65 | 0.081 | 0.53 | 0.095 | 0.47 | −0.029 | −0.36 | −0.18 |
ln (per capita income) | −0.008 | −0.05 | −0.182 | −1.40 | −0.188 | −2.33** | −1.17 | 0.150 | 0.78 | 0.306 | 1.47 | 0.125 | 0.86 | 0.43 |
ln (average household size) | −1.670 | −2.44** | −1.531 | −3.07*** | −1.377 | −4.80*** | −1.93* | −0.473 | −0.61 | −1.551 | −2.20** | −1.203 | −2.88*** | −1.28 |
ln (house area per capita) | −0.150 | −1.10 | −0.139 | −1.35 | −0.088 | −1.25 | −0.56 | 0.018 | 0.10 | −0.003 | −0.02 | 0.161 | 1.46 | 0.57 |
ln (cropland area per capita) | −0.088 | −1.03 | −0.157 | −2.25** | −0.143 | −4.51*** | −1.96** | −0.105 | −1.10 | 0.032 | 0.30 | 0.053 | 0.82 | 0.42 |
ln (fraction of population with > primary education) | 0.086 | 0.25 | 0.175 | 0.61 | 0.269 | 1.83* | 0.85 | 1.723 | 4.32*** | 1.246 | 2.86*** | 1.848 | 5.69*** | 2.46** |
Time trend | −0.046 | −3.99*** | −0.030 | −2.94*** | −0.032 | −8.80*** | −3.36*** | −0.017 | −1.22 | −0.023 | −1.62 | −0.031 | −4.29*** | −2.23** |
Test of over-identifying restrictions | ||||||||||||||
χ2 (35) = 100.315*** | χ2 (14) = 44.40*** | χ2 (14) = 20.51 | χ2 (35) = 59.736*** | χ2 (14) = 25.20** | χ2 (14) = 17.52 | |||||||||
Tests of exogeneity of instrument subsets | ||||||||||||||
– | χ2 (10) = 31.65*** | χ2 (10) = 8.91 | χ2 (10) = 20.63** | χ2 (10) = 13.46 | ||||||||||
Test of first-order non-autocorrelation among residuals | ||||||||||||||
Z = 6.120*** | Z = −5.270*** | Z = −1.880* | Z = −1.941* | Z = −1.980** | Z = −2.200** | |||||||||
Test of second-order non-autocorrelation among residuals | ||||||||||||||
Z = 0.835 | Z = 1.350 | Z = 1.490 | Z = −1.096 | Z = −0.750 | Z = −0.42 | |||||||||
The tests for over-identifying restrictions reject the null hypothesis of over-identifying restrictions for the one-step method for both the alcohol and cigarette equations in both the Arellano–Bond model and the Arellano–Bover model, but not for the two-step method in the Arellano–Bover model. This implies that there is an identification problem for the one-step GMM methods.
The consistency of GMM estimators hinges heavily upon the assumption that \( E[\varepsilon_{it} \varepsilon_{it - 2} ] = 0 \) rather than \( E[\varepsilon_{it} \varepsilon_{it - 1} ] = 0 \) [2]. The Arellano–Bond test of the null hypothesis of no second-order correlation cannot be rejected for any of the GMM models, which indicates that the estimators are consistent.
The over-identification tests cannot reject the null hypothesis of over-identifying restrictions for the two-step Arellano–Bover method; neither can the test of exogeneity of instrumental variables. Furthermore, the estimation results of the two-step Arellano–Bover method are consistent with that of the fixed-effects model. Therefore, the following discussion is based on these results. Arellano and Bond [2] and Blundell and Bond [11] point out that though the two step method is asymptotically more efficient, the reported two-step standard errors tend to be severely downward biased. Windmeijer [46] developed a finite-sample correction to the two-step covariance matrix, which makes the two-step method more efficient than the one-step method. Table 3 reports both the uncorrected t ratios and the corrected t ratios for the two-step Arellano–Bover method. The discussion here is based on the corrected t ratios.
Dependence, reinforcement and tolerance
The lagged terms for both cigarettes and alcohol are statistically significant and positive, implying that the effects of dependence, reinforcement and tolerance are important for both products. The coefficients for cigarettes and alcohol are 0.426 and 0.146, respectively. This implies that the consumption of cigarettes has stronger dependence, reinforcement and tolerance effects than alcohol, and current consumption of cigarettes has stronger impacts on future consumption of cigarettes than on alcohol consumption.
Prices
The most interesting result of this study is that the demands for both alcohol and cigarettes are highly sensitive to the price of alcohol. The estimated own-price elasticity of demand for alcohol is about −1.53 in the short run, substantially higher than the estimate of −0.34 in Fan et al. [17]. Our estimated short-run cross-price elasticity of the demand for cigarettes with respect to the price of alcohol is about −0.62, which indicates that cigarettes are a Marshallian complement to alcohol. This is also substantially larger in absolute value than the estimate of −0.19 in Fan et al. [17]. Our estimated long-run cross-price elasticity is approximately −1.08.
Our results imply that taxes on alcohol may be a very effective tool for reducing consumption of both alcohol and cigarettes. Similar results can be found in Decker and Schwartz [16] and Busch et al. [13]. A possible explanation based on the medical and psychological literature, as mentioned above, is that alcohol dose-dependency increases the urge to smoke, at least among light smokers [31].6
Interestingly, our results fail to show a statistically significant effect of the price of cigarettes on either cigarette consumption or alcohol consumption. The point estimate of the own-price elasticity of demand for cigarettes is positive and very small (about 0.09 in the short run and 0.16 in the long run). The policy implication is that an increase in cigarette taxes may not be effective in curbing cigarette or alcohol consumption in rural China, although it could be an efficient way of raising government revenue. Similar results concerning the impacts of prices of cigarettes and alcohol can be found in the fixed-effects model.
The estimated short-run cross-price elasticity of demand for alcohol with respect to the price of grains is about 0.64 and statistically significant. The estimated impact of grain prices on cigarette consumption is also positive, but it is not statistically significant. The results for alcohol consumption may be explained by the income effect of an increase in grain price: as grain prices increase, rural incomes increase and hence the demand for alcohol increases. The estimated short-run cross-price elasticities of demand for cigarettes and alcohol with respect to the price of meat are not statistically significant.
Income
The per capita income variable is not statistically significant for either cigarettes or alcohol. The point estimates of the income elasticities are 0.19 and 0.13, respectively. The results imply that demands for alcohol and cigarettes are not sensitive to income. This finding is consistent with some studies on cigarette consumption [10, 48] which argue that income may not play a significant role in the consumption of addictive products due to dependence effects.
Household size, house area, and cropland area
The estimated impact of average household size on the demand for cigarettes is negative and marginally significant, while the estimated impact on the demand for alcohol is not statistically significant. A possible explanation is that an increase in household size raises the pressure on smokers within the household to quit, because smoking often crowds out expenditures for other goods [13]. Smoking can also have serious health effects on other members of the household through second-hand smoke, and the number exposed to second-hand smoke increases as household size increases.
House area and cropland area per capita do not have a statistically significant effect on the demand for alcohol. House area also is not statistically significant for cigarette consumption. Cropland area per capita has a negative and statistically significant impact on cigarette consumption, while the estimated impact on the demand for alcohol is not statistically significant. Mullahy and Sindelar [40] suggest that addictive goods, such as alcohol, are complementary in consumption with leisure. As cropland area per capita increases, there is more farm work to do; hence leisure decreases and the demand for cigarettes decreases in turn.7
Education
Another interesting result of this study is that education in rural China increases cigarette and alcohol consumption. The estimated impacts of an increase in the fraction of the population with more than primary education on consumption of these two goods are positive, and statistically significant for alcohol consumption. In the short run, an increase in this fraction by 1% increases alcohol consumption by about 1.84%.
Wu [47] and Pan et al. [43] also find that education is positively associated with consumption of alcoholic products in China, except for the demand for wine coolers [43]. Hu and Tsai [26] find that education is positively correlated with consumption of cigarettes.
In rural China, higher education is often associated with more social activities where cigarettes and alcohol are available and consumption is encouraged. Peer pressure [18] in these settings may induce more people to consume cigarettes and alcohol, and those already consuming to consume more. The policy implication for reducing the consumption of alcohol and cigarettes is to promote alternative, healthier lifestyles and other types of social activities that do not involve cigarettes or alcohol.
Studies for other countries suggest more ambiguous effects of education on cigarette and alcohol consumption. For the United States, Grossman et al. [20] found that education has a positive impact on alcohol consumption. However, another United States study by Decker and Schwartz [16], and a study in Russia by Baltagi and Geishecker [4], suggested that the relationship between education and alcohol consumption might have an inverted U-shape: as education increases, alcohol consumption at first increases, and then decreases, with the peak occurring typically at a high school education level. Testing an inverted U-shaped model for rural China is not really feasible because the percentage of the population in rural China with more than a high school education is still very low, only 1.1% in 2005.8
Studies for the United Kingdom [28] and the United States [16, 48] suggest that education is negatively associated with the demand for cigarettes. A possible explanation in these countries is that education may increase the cognitive skill of an individual regarding the health risks of smoking to both the individual and others in the household through second-hand smoke.
Time trend variables
The time trend is statistically significant and negative for both cigarettes and alcohol consumption in rural China. The estimated values of the coefficients are such that, other factors being equal, the demand for cigarettes and alcohol per capita declined continuously during the period 1994–2003.
Conclusions
Using aggregate data for 26 provinces from 1994 to 2003 to estimate a habit persistence model of the demand for cigarettes and alcohol, this paper explored the interactions between cigarette and alcohol consumption in rural China. The main findings are that:
1.
Dependence, reinforcement and tolerance effects are statistically significant for both cigarette consumption and alcohol consumption, with a stronger effect for cigarettes than alcohol.
2.
Demand for both cigarettes and alcohol is very sensitive to the price of alcohol, but not to the price of cigarettes or to income.
3.
The consumption of alcohol is positively associated with education. In rural China, higher education is often associated with more social activities where alcohol is consumed.
Our results imply that taxes on alcohol could have a double dividend—they may be a very effective tool for reducing consumption of both alcohol and cigarettes. On the other hand, an increase in cigarette taxes may not be effective in curbing cigarette or alcohol consumption in rural China, although it could be an efficient way of raising government revenue. Our results for education suggest the promotion of alternative, healthier lifestyles and other types of social activities that do not involve alcohol. Our results indicate that cigarettes are a Marshallian complement to alcohol, so that reducing alcohol consumption may also reduce consumption of cigarettes.
Open Access
This article is distributed under the terms of the Creative Commons Attribution Noncommercial License which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.
Open AccessThis is an open access article distributed under the terms of the Creative Commons Attribution Noncommercial License (https://creativecommons.org/licenses/by-nc/2.0), which permits any noncommercial use, distribution, and reproduction in any medium, provided the original author(s) and source are credited.