Study design and data
In this study, data of all 28 EU countries were collected to construct a cigarette demand structure model. One dependent variable and five independent variables were considered. Per capita cigarette consumption for those aged 15 and over was chosen as the dependent variable. Independent variables comprised cigarette prices, cigarette prices in Eastern European countries, gross national income (GNI), rural population, and the number of MPOWER measures implemented at the highest level of achievement.
Data regarding cigarette consumption, cigarette prices, and cigarette prices in Eastern European countries were extracted from the 2005–2014 Euromonitor International market research database [
21]. Euromonitor International is recognized as a leading independent provider of global business intelligence, specialized in creating worldwide data and analysis on consumer products and services. Consumption of cigarette products was calculated based on annual cigarette consumption per capita for those aged 15 and over. The retail price for a pack of cigarettes in each country was calculated by dividing cigarette sales revenues by cigarette consumption, which was further deflated using consumer price indexes.
Cigarette price in Eastern European countries refers to the combined average cigarette price in Estonia, Latvia, Lithuania, Poland, Czech Republic, Slovakia, Hungary, Romania, Slovenia, Croatia and Bulgaria.
GNI per capita data were converted to US dollars using the World Bank Atlas method [
22], divided by the midyear population, and deflated based on consumer price indexes. Required data were retrieved from the World Bank’s database. In this study, the ratio of the rural population to the total population was used in the analysis. Rural population refers to the number of people living in rural areas as defined by the National Statistical Offices and was calculated as the difference between the total population and the urban population. Data on the ratio of the rural population to the total population are World Bank estimates [
22] and based on the United Nations, World Urbanization Prospects [
23].
As to the number of MPOWER measures implemented at the highest level of achievement by each country in each year, figures for the years 2007 to 2014 were taken from the 2015 WHO report on global tobacco epidemic [
17]. Data for the years 2005, 2006 and 2015 were unavailable and treated as missing data in the analysis.
Data characteristics
Table
1 shows the list of variables used in the analysis and the data characteristics. In 2014, per capita cigarette consumption in the 28 EU countries for adults aged 15 years and over was the highest in Slovenia at 2098 cigarettes, followed by the Czech Republic (1720 cigarettes), Austria (1631 cigarettes), Greece (1552 cigarettes), and Romania (1501 cigarettes); those of the other EU countries were all less than 1500 cigarettes. In 2014, the average real retail price was the highest in United Kingdom at US$9.48 per pack, followed by Ireland (US$9.04). In addition, the average highest number of MPOWER measures implemented in Ireland, Spain, and the United Kingdom was 3, followed by Belgium, Bulgaria, Denmark, Greece, Malta, and the Netherlands at 2; whereas in the remaining EU countries the number of measures were below 2.
Table 1
Comparison of cigarette consumption, retail prices and gross national income from 2005 to 2014 in the European Union
Austria | 1688 | 1631 | −3.38% | 3.27 | 4.59 | 40.37% | 38,500 | 41,765 | 8.48% | 1 | 74.00% |
Belgium | 1156 | 966 | −16.44% | 4.5 | 5.24 | 16.44% | 37,850 | 39,561 | 4.52% | 2 | 75.92% |
Bulgaria | 2750 | 1282 | −53.38% | 0.8 | 1.6 | 100.00% | 3760 | 5261 | 39.92% | 2 | 82.65% |
Croatia | 1735 | 1330 | −23.34% | 1.69 | 2.74 | 62.13% | 9870 | 10,471 | 6.09% | 0 | 75.26% |
Czech Republic | 2183 | 1720 | −21.21% | 1.77 | 2.79 | 57.63% | 12,380 | 15,246 | 23.15% | 1 | 77.42% |
Cyprus | 1371 | 1400 | 2.12% | 2.06 | 2.72 | 32.04% | 21,490 | 22,597 | 5.15% | 0 | 77.47% |
Denmark | 1472 | 1080 | −26.63% | 4.06 | 5.15 | 26.85% | 49,620 | 52,632 | 6.07% | 2 | 74.75% |
Estonia | 1481 | 1480 | −0.07% | 1.25 | 2.27 | 81.60% | 9710 | 13,436 | 38.37% | 1 | 77.24% |
Finland | 954 | 789 | −17.30% | 4.05 | 5.44 | 34.32% | 40,100 | 40,954 | 2.13% | 1 | 81.53% |
France | 898 | 682 | −24.05% | 5 | 7.07 | 41.40% | 36,000 | 37,487 | 4.13% | 1 | 80.3% |
Germany | 1175 | 980 | −16.60% | 4.49 | 5.23 | 16.48% | 35,880 | 41,335 | 15.20% | 1 | 72.9% |
Greece | 3016 | 1552 | −48.54% | 2.46 | 3.3 | 34.15% | 22,510 | 18,212 | −19.09% | 2 | 79.95% |
Hungary | 1366 | 936 | −31.48% | 1.52 | 1.88 | 23.68% | 10,430 | 9202 | −11.77% | 1 | 77.26% |
Ireland | 1364 | 685 | −49.78% | 6.01 | 9.04 | 50.42% | 44,500 | 32,362 | −27.28% | 3 | 77.8% |
Italy | 1569 | 1147 | −26.90% | 3.57 | 4.9 | 37.25% | 32,390 | 29,248 | −9.70% | 1 | 75.68% |
Latvia | 2089 | 892 | −57.30% | 0.81 | 2.02 | 149.38% | 7360 | 10,293 | 39.85% | 1 | 76.89% |
Lithuania | 1079 | 965 | −10.57% | 1.26 | 2.42 | 92.06% | 7550 | 11,421 | 51.27% | 1 | 75.76% |
Luxembourg | 1110 | 854 | −23.06% | 4.57 | 4.63 | 1.31% | 70,340 | 62,685 | −10.88% | 1 | 70.24% |
Malta | 1661 | 1048 | −36.91% | 2.76 | 3.77 | 36.59% | 14,380 | 19,645 | 36.61% | 2 | 74.63% |
Netherlands | 837 | 659 | −21.27% | 3.13 | 5.92 | 89.14% | 42,390 | 43,804 | 3.34% | 2 | 73.4% |
Poland | 1926 | 1136 | −41.02% | 1.17 | 2.52 | 115.38% | 7330 | 10,858 | 48.13% | 1 | 80.29% |
Portugal | 1754 | 852 | −51.43% | 2.29 | 4.62 | 101.75% | 18,550 | 18,342 | −1.12% | 1 | 74.51% |
Romania | 1604 | 1501 | −6.42% | 0.23 | 0.55 | 139.13% | 3930 | 6190 | 57.51% | 1 | 75.41% |
Slovakia | 1136 | 1381 | 21.57% | 1.26 | 1.83 | 45.24% | 11,280 | 14,410 | 27.75% | 1 | 81.54% |
Slovenia | 2281 | 2098 | −8.02% | 1.54 | 2.47 | 60.39% | 18,440 | 19,217 | 4.21% | 1 | 80.41% |
Spain | 2212 | 911 | −58.82% | 2.79 | 4.09 | 46.59% | 25,930 | 24,424 | −5.81% | 3 | 78.09% |
Sweden | 2212 | 666 | −69.89% | 4.93 | 6.37 | 29.21% | 45,350 | 54,339 | 19.82% | 1 | 68.84% |
United Kingdom | 843 | 568 | −32.62% | 7.87 | 9.48 | 20.46% | 41,150 | 34,194 | −16.90% | 3 | 82.16% |
Empirical specification and analysis
To calculate cigarette price elasticity, a cigarette demand structure model was constructed using cigarette consumption as the dependent variable and cigarette price, cigarette prices in Eastern Europe countries, GNI, rural population, and the highest number of MPOWER measures implemented as explanatory variables. Cigarette price elasticity was estimated with income as the threshold variable using the threshold regression model of panel data from Hansen [
24].
The baseline cigarette demand structure model of the 28 EU countries is as follows:
$$ \ln {C}_{it}={\beta}_{1i}+{\beta}_2\ln {P}_{it}+{\beta}_3\ln {GNI}_{it}+{\beta}_4{Rural}_{it}+{\beta}_5{MP}_{it}+{\beta}_6\ln {NeiP}_{it}+{\varepsilon}_{it} $$
(1)
Where,
C
it
: the annual cigarette consumption per capita in the population aged 15 years and over in country i in year t.
P
it
: the cigarette price per cigarette in country i in year t.
GNI
it
: the per capita national income in country i in year t.
Rural
it
: the rural population percentage in country i in year t.
MP
it
: the highest number of MPOWER measures implemented in country i in year t.
NeiP
it
: cigarette prices per cigarette in Eastern European country i in year t.
Formula (1) is the traditional constant-elasticity log-linear demand model, but the influence of cigarette prices on cigarette consumption may not be limited to a single pattern. That is, there may also be non-linear structural relationships, such as income threshold effects on the demand for tobacco products [
20]. We thus used income as the threshold variable in the threshold regression model to estimate the elasticity of cigarette prices and to simulate the effects of price fluctuations. One feature of a threshold regression model is that threshold variables are ordered so as to be a structure breakpoint of regime variation. The estimated reference points are divided into different regimes by variable value, which is greater or smaller than the threshold value.
The panel threshold regression model is often “de-meaned” first, in order to eliminate the individual effect
β
i
[
24]. If our baseline model contains three regimes of national incomes that are conditional on two threshold values, Eq.
1 can be derived as
$$ {\displaystyle \begin{array}{l}\ln {C}_{it}^{\ast }={\beta}_{21}\ln {P}_{it}^{\ast}\left( GNI\le {\gamma}_1\right)+{\beta}_{22}\ln {P}_{it}^{\ast}\left({\gamma}_1\le GNI\le {\gamma}_2\right)+{\beta}_{23}\ln {P}_{it}^{\ast}\left( GNI>{\gamma}_2\right)+\\ {}\kern3.12em {\beta}_{31}\ln {GNI}_{it}^{\ast}\left( GNI\le {\gamma}_1\right)+{\beta}_{32}\ln {GNI}_{it}^{\ast}\left({\gamma}_1\le GNI\le {\gamma}_2\right)+{\beta}_{33}\ln {GNI}_{it}^{\ast}\left( GNI>{\gamma}_2\right)\\ {}\kern2.84em +\kern0.72em {\beta}_4{Rural}_{it}^{\ast }+{\beta}_5{MP}_{it}^{\ast }+{\beta}_6\ln {NeiP}_{it}^{\ast }+{\varepsilon}_{it}^{\ast}\end{array}} $$
(2)
Where
\( {\overline{C}}_i={T}^{-1}{\sum}_{t=1}^T{C}_{it} \);
\( {C}_{it}^{\ast }={C}_{it}-\overline{C_i} \);
\( {P}_{it}^{\ast }={P}_{it}-\overline{P_i} \);
\( {GNI}_{it}^{\ast }={GNI}_{it}-{\overline{GNI}}_i \);
\( {Rural}_{it}^{\ast }={Rural}_{it}-{\overline{Rural}}_i \);
\( {MP}_{it}^{\ast }={MP}_{it}-{\overline{MP}}_i \);
\( {NeiP}_{it}^{\ast }={NeiP}_{it}-{\overline{NeiP}}_i \);
\( {\varepsilon}_{it}^{\ast }={\varepsilon}_{it}-{\overline{\varepsilon}}_i \); and γ
1 and γ
2 are the two threshold values that control
GNI
it
. We use ordinary least squares (OLS) to estimate Eq.
2. The selection of threshold variables in the empirical model can be determined by economic theory or statistical testing. In the process of statistical testing, the null hypothesis (H
0:
β
i
are all the same) maintains that a traditional log-linear model is sufficient. In this article, we applied the likelihood ratio to test the nonlinear relationship. In such cases, further testing is required to determine single, double or triple thresholds. In order to avoid heteroskedasticity, the inconsistency of standard errors caused by serial correlations and heterogeneity of the residual terms, consistent correction was performed according to White (1980) [
25]. Statistical software packages Gauss and Stata were used to perform the analysis.
To determine the effects of cigarette price hikes on cigarette consumption, cigarette consumption in 2015 was set as the baseline for this study. We introduced 10% increments in cigarette prices to simulate changes in future cigarette consumption based on the cigarette price elasticity estimated in this study. Changes in tobacco tax revenues were calculated based on changes in consumption due to price increases.
The number of averted smoking-attributable deaths (SADs) derived from the simulated impact of price increments on the reduction in smokers and was adjusted for the fact that smoking cessation still carries considerable risks of early death [
26]. The applied mortality adjustment factors were calculated for each country surveyed, assuming that 95, 75, 70, 50 and 10% of those who ceased smoking when aged 15 to 29, 30 to 39, 40 to 49, 50 to 59 and at least 60 years, respectively, would remain unaffected by their previous smoking habits [
27]. Data on population stratification were extracted from the Eurostat database.