Background In El Salvador, 8.8% of adults 15 years and older smoke cigarettes. Little is known about the sensitivity of cigarette consumption among the adults in El Salvador to tax and price increases and income growth.
Methods Elasticities are estimated using Deaton’s Almost Ideal Demand System model applied to data from the National Household Income and Expenditure Survey 2005/2006 for the total population and separately for income groups. The estimates are then used to simulate the effects of a proposed change in tobacco tax policy on cigarette consumption and tax revenue.
Findings The estimated price elasticities (−0.77 for the total population) are within the range of price elasticity estimates available for low and middle-income countries. Given the estimated elasticities, a tobacco tax increase is expected to reduce the number of smokers (by almost 20%) and increase tobacco tax revenue (by more than 50%).
Conclusions Increasing tobacco taxes has the potential to decrease consumption in El Salvador and raise fiscal revenues. The tobacco tax burden in El Salvador is one of the lowest in Latin America and the social costs of tobacco consumption largely exceed the tobacco tax revenues. An increase in tobacco tax could significantly decrease the number of smokers and reduce the burden of tobacco-related diseases and deaths.
- socioeconomic status
- public policy
Data availability statement
Data are available in a public, open access repository. Data are available upon reasonable request. Secondary publicly available data were used.
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.
Globally, tobacco use caused about 8 million deaths in 2017; of which, at least 0.9 million deaths were in the Americas region.1 2 As of 2015, tobacco consumption ranked as the second major risk factor that accounted for the most healthy years of life lost in 109 countries.3 Tobacco consumption also crowds out household resources which could have been better used on human capital, such as health and education.4–7 The WHO Framework Convention on Tobacco Control (FCTC) recommends a set of policies, among which the single most effective and cost-effective measure to control tobacco consumption is increasing tobacco taxes.8–10 Currently, approximately two-thirds of the ratifying countries have implemented at least one FCTC policy.11
El Salvador was the last country in Central America to ratify the FCTC in 2014, ten years after its signing.12 Although a tobacco control law was passed in 2011, it led to no increase in the tobacco tax. Consequently, cigarette tax burden was unchanged between 2012 and 2016 (around 52% of the most popular brand) and decreased to 47% by 2018.1 13 14
In 2014, monthly prevalence of cigarette consumption in El Salvador was 8.8% among adults over 15 years of age, with higher tobacco use prevalence among men (16.9%) than among women (2.2%).15 For youth smokers, the scenario was more worrisome; in 2015, the monthly prevalence was 9.9% among children aged between 13 and 15 years, with higher prevalence among boys (11.4%) than among girls (8.2%).16
A key parameter to assess tax effectiveness is the own price elasticity of demand, defined as percentage change in consumption in response to a given change in price keeping everything else constant. A larger price elasticity (in absolute values) means that a relatively small price increase implies a relatively large decrease in consumption, and vice versa. Another important parameter for policy purposes is the income or expenditure elasticity defined as the percentage change in consumption in response to a given change in income or total expenditure (a proxy for income).
Global evidence suggests that tobacco products, especially cigarettes, are price inelastic goods (ie, price elasticity is below 1, in absolute value). The price elasticity is approximately −0.4 in high-income countries and around −0.5 for low and middle-income countries (LMIC).8 A recent systematic review showed that Latin American and the Caribbean countries demonstrate an average price elasticity of −0.4.17 Far less is known about income or expenditure elasticities, though in most cases they are positive (eg, an increase in incomes and/or total expenditures increases cigarette demand).8 In a previous study on El Salvador, the price elasticity of cigarette demand was estimated at −0.93 (ie, a 10% increase in cigarette prices would decrease cigarette consumption by 9.3%), and the income elasticity was estimated at 0.9.12 This study used time series data from 2000 to 2014 on tax-paid cigarette sales and average cigarette prices. Tax-paid sales exclude illicit cigarette sales and can therefore induce upward bias in the estimation of price elasticity of cigarette consumption. An increase in licit cigarette prices (because of a tax increase or due to industry’s decision) will decrease consumption of licit cigarette (and that is captured by estimates of price elasticity when using tax-paid sales), but also might cause some consumers to switch to illicit cigarettes. Hence, only using the decrease in tax paying sales would overestimate the decrease in total consumption and would overestimate price elasticity, as supported by global evidence from a number of countries.9
The primary objective of this study is to estimate price elasticity and expenditure elasticity of cigarette demand in El Salvador using household-level survey data on self-reported cigarette consumption from the latest National Household Income and Expenditure Survey conducted in 2005/2006 (ENIGH 2005/2006).18 Unlike the estimates based on time series data, estimates based on household level allow one to control for observable household-level characteristics, such as sex, age, education of household head, household size, household location, and others, in order to study possible differentiated effects across different sociodemographic and economic determinants of tobacco demand. They also consider consumption of illicit cigarettes, as the survey collects information on all expenditures.
The estimates of price and expenditure elasticities of cigarette demand were obtained by using the Almost Ideal Demand System (AIDS) method that controls for the quality choice of households across product categories.19 20 The estimates were then used to simulate the effects of a proposed change in tobacco tax policy in El Salvador on cigarette consumption and tax revenue.
Methods and data
Data from the ENIGH 2005/2006 which was carried out between September 2005 and August 2006 were used for the analysis.18 In the ENIGH 2005/2006, the households were asked to report the quantities and expenditures on every item they incur during the data collection of the survey. For the design of the sample, the municipalities were divided into 227 segments. Subsequently, the segments were used as primary sampling units (PSU) in a stratified (urban and rural), two-stage sampling design. The full sample consisted of 4381 households, distributed in 468 PSUs with a response rate of 95.2%. All expenses incurred by households were identified according to purpose: household purchases, business, agriculture, and others. For the analysis in this paper, only expenditures for household purchases were considered; 32 households did not register any expenses and were eliminated, reducing the full analytical sample size to 4349, including both tobacco user and non-user households.
The expenses and quantities consumed of cigarettes were identified as ‘cigarettes without filter,’ ‘cigarettes with filter,’ ‘unspecified cigarettes’ and ‘imputed expenditure on cigarettes with or without filter’. Because it was impossible to know the method by which the Dirección General de Estadísticas y Censos, El Salvador’s statistics office, imputed the expenses in the fourth category of ‘imputed expenditure on cigarettes with or without filter’, these cigarette expenditures were not included in the analysis. Thus, 638 purchases of 267 tobacco user households were eliminated from the sample, leaving the final database with 2522 cigarette purchases registered in 386 tobacco user households.
Once the total expenditure on cigarettes per household was identified, a series of demographic variables of households, including sex, age and educational level of the head of household, area of residence (urban or rural), number of people residing in the household, the proportion of individuals over 15 years residing in the household, total household expenditure and budget share for cigarettes, were identified. With the total quantity of cigarettes consumed and household expenditures on cigarettes, cigarette unit values, defined as the ratio of expenditures on cigarettes to the quantity of cigarettes purchased within each household, were calculated.
To avoid outliers influencing the results, households that reported more than 3 SDs from the average in budget share and/or unit values were eliminated, a procedure that has been followed elsewhere.21–23 In this case, only 17 households, representing 4.4% of those consuming cigarettes, were discarded, leaving the final full analytical sample to 4332 households.
In table 1 the main demographic and socioeconomic variables of households and the purchases of cigarettes by quintiles of total household expenditure are shown. 34.2% of households were headed by women, this being greater in households with the lowest total expenditures. The average age of heads of household was 48 years, with higher age for households with lower total household expenditure. 37% of households were located in rural areas, mostly concentrated in the lowest total household expenditure quintile. The average number of members in the household was 4.2 and households were larger in size in the highest quintile. 65.9% households had heads with no education or incomplete primary education; 13.1% completed primary education or did not complete secondary education; 10.4% completed secondary education; and 10.3% had above secondary education (completed or not). There is significant inequality in education and total expenditure across quintiles. Households in the highest income quintile spent nine times more than the total expenditure of the households in the lowest quintile. The highest total expenditure on cigarettes (among the households that purchased tobacco) took place in quintile 4 at approximately $14 (equivalent to 200 cigarettes on average per household), followed by quintile 3. The households in the first quintile had the lowest total expenditure on cigarettes and also paid less unit value per cigarette. These households also devoted the highest budget share to cigarettes (4.73%).
The estimation of price and expenditure elasticity was carried out using the AIDS that adjusts for the household choice of product quality.19 20 This method uses the spatial variation of prices between clusters (PSUs) to obtain the price elasticity estimate. The consumption and quality choices are modelled and then estimated simultaneously using the seemingly unrelated regression of the following system of equations:
In equation (1), is the proportion of good G in total household expenditure of household i belonging to the cluster (PSU) c. The variable is the total household expenditure (as proxy for household income) and the variable is a vector of demographic variables, such as the natural logarithm of the number of people in the household, number of people over 15 years of age, proportion of women over 15 years of age, and sex, age and educational level (primary, secondary and above secondary) of the head of household. The variable corresponds to the unobserved prices of good H for household i of cluster c. represents the cluster fixed effects and is the idiosyncratic error term.
Equation (2) models the quality choice of households, where they choose the unit value of the good to be consumed, that is, the price per unit (the ratio of total expenditure on cigarettes to the number of sticks purchased). Unlike prices, which are established by the market, the unit value is determined by households. Given household budget and certain product characteristics (eg, filter, packaging, length, flavour, brand, and so on), they can choose cigarettes from a wide range of options reflected in the unit value of cigarettes. This choice is modelled as a function of the same variables as equation (1), except . Once the quality choice has been estimated, the coefficient represents the elasticity of quality with respect to total expenditure: the percentage change in the unit value in response to a one percentage change in total household expenditure. is the idiosyncratic error term.
It is important to note that in both equations the unobserved prices are constant at the cluster level, a crucial assumption for the present model. Price variations occur only between clusters. All households belonging to the same cluster observe the same price and make their consumption and quality decisions based on that price. Unlike unit values, we have no information on the prices observed by households. Therefore, the equations are not directly estimated, and an indirect method is used in three stages with cluster-level fixed effects described in detail elsewhere.7 The SEs of estimates were estimated using a bootstrap of 1000 replications. Additional analysis was conducted by household socioeconomic status of interest. The estimation was done for two subsamples: bottom 40% and top 20% households ranked in order of total household expenditure.
Demand identification in Deaton’s AIDS model comes from price’s spatial variation, as explained elsewhere.7 24 In the case of El Salvador, running an analysis of variance test between the natural logarithm of cigarette unit values and PSUs shows that this assumption is reasonable, as the F value associated with a null hypothesis of no spatial variation has a value of 86.98 (p<0.001). The R2 is over 0.50 (equal to 0.514), the threshold defined as reasonable for such a variation.7
The estimated elasticities are presented in table 2. The price elasticity of cigarettes for the total population is −0.77, statistically significant at 1% level. This means that, given a 10% increase in the price of cigarettes, consumption would decrease by 7.7%. The expenditure elasticity for the total population is 0.43, also statistically significant at 1%. This implies that a 10% increase in total expenditure would lead to an increase in cigarette expenditure by 4.3%. The quality elasticity with respect to the total expenditure is statistically significant at 1% and equal to 0.05. This estimate indicates that for a 10% increase in total spending, households increase the chosen unit value of cigarettes by 0.5%. The inclusion of outliers does not alter significantly these results (not shown but available from the authors).
The subsample analysis shows that the cigarette price elasticity is statistically not significant for households in quintiles 1 and 2. It is statistically significant for households in quintile 5 and is slightly above the elasticity for the total population in absolute value (−0.88). The total expenditure elasticity is significant and higher in households with lower total expenditure, which implies that cigarette consumption increases in a greater proportion among lower income households in response to a given increase in income. Quality elasticity is only statistically significant for high-expenditure households.
Simulation of tobacco tax increase
In the baseline scenario, as of 2019, there is a specific excise of US$0.0225 per cigarette and an ad valorem excise of 39% of the final retail price without taxes, plus a value-added tax (VAT) of 13% of the retail price (excluding VAT). The total average taxes collected by the Salvadorian Treasury, including excise and VAT, amount to US$1.58 accounting for 45.7% of the final price of an average 20-cigarette pack (US$3.46), well below the 75% threshold recommended by the WHO. Total fiscal revenue (tobacco excise plus VAT) amounted to US$35.9 million in 2017.
We used the aforementioned baseline tax rates and average retail price of cigarette to derive the producer price as follows:
The base year PP was adjusted for expected rate of inflation (available from the World Economic Outlook database of the International Monetary Fund25) in every successive year beginning in 2020 to keep the PP constant in 2019 prices. Baseline tax-paid cigarette sales (S) and tax revenue (R) data were obtained from the Ministry of Finance.
Using the price elasticity (EP) and income (proxied using expenditure) elasticity (EY) parameters estimated here, we predicted the new cigarette sales (St) in each successive year as follows:
where ΔY represents expected rate of income growth as reflected in the expected rate of increase in per-capita gross domestic product in the corresponding year.25 The predicted sale was multiplied with the tax rate per unit to predict the tax revenue each year.
In addition, the effects of tax and price increases on smoking prevalence were estimated by applying prevalence price elasticity and prevalence income elasticity (assumed half of EP and EY, respectively, following ref 26) to the baseline male and female smoking prevalence data (interpolated from ref 27 for 2019). The number of smokers was estimated by multiplying the predicted smoking prevalence and projected adult (age 15+) population size.28
For simulating the effects of tobacco tax increase over 2020–2025 from the above base year level, we considered the following scenario:
Increasing the specific tax on cigarettes to US$0.10 per cigarette in 2020.
Increasing the specific tax by 5% every year from 2021 through 2025 with annual indexation.
Keeping the ad valorem tax at 39% on the price without taxes and VAT at 13% of the suggested retail price through 2025.
The new final retail price level (RPt) in each year from 2020 through 2025 was calculated by summing the inflation-adjusted PP, and the new tax levels of specific and ad valorem excise and VAT in corresponding years. Thus, full pass-through of the tax increase to consumers was assumed (the real PP is constant across time). The year-on-year percentage increase in the final retail price (ΔRP) was calculated after adjustment for the expected rate of inflation.
The total decrease in cigarettes sold as a consequence of a price increase may be decomposed in a decrease in cigarette consumption due to people who quit smoking, and a decrease in cigarette consumption due to smokers consuming a fewer number of cigarettes. Hence, total price elasticity (presented in table 2) is also formed by two components: a prevalence price elasticity and a conditional price elasticity. The first measures in which percentage does smoking prevalence fall when tobacco prices increase by 10%; while the second measures in which percentage cigarette consumption decreases among smokers. The same can be said for income (or expenditure) elasticity: an increase in income may result in an increase of smokers and/or an increase in the number of cigarettes smoked by smokers (if income elasticity is positive).
Following ref 26 we assume that prevalence price and income elasticities are half the value of the total price and income elasticities. Using the price elasticity estimate of −0.77 and the expenditure elasticity estimate of 0.43 (and its corresponding prevalence elasticities), we ran the simulation of cigarette tax increase specified above. The results of the year-on-year estimated effects of the tax increase on cigarette price, tax share, sales, smoking prevalence, number of adult smokers and tax revenue are reported in table 3. It shows that the tax increase is expected to lead to a large increase in cigarette price (50% after adjustment for inflation) in the first year 2020 followed by small increments in subsequent years up to 2025. Between 2019 and 2025, the retail price would increase by 63% after adjustment for inflation. The tax share in the retail price will jump from 45.7% to 63.7% in 2020 and would gradually reach 66.8% by 2025.
The sustained price increases are expected to reduce total cigarette sale from 454.3 million pieces in 2019 to 271.7 million in 2025 marking 40% decline over 6 years. The largest part of this decrease in sale (38%) would, however, take place in the first year following most of the price increase taking place in that year. Similarly, adult smoking prevalence would decrease from 9.2% in 2019 to 7.6% in 2020 and gradually to 7.5% by 2025. The expected relative decline in smoking prevalence due to tax increase over 2019–2025 is 18.6%. As a result, there will be about 74 453 fewer adult smokers by 2025. Following epidemiological studies that have shown that tobacco ultimately kills a third to a half of all of its users,26 29 30 we assumed that about 33%–50% of smokers die prematurely and about 70% of these premature deaths could be avoided by quitting smoking. Thus, we estimated that between 17 198 (74 453×33%×70%) and 26 058 (74 453×50%×70%) lives could be saved from smoking-attributable deaths by the tax increase.
The annual revenue collection from cigarette taxes (including excise and VAT) would increase from the current level of US$35.9 million to US$54.7 million by 2025. The tax increase would result in a 43% increase in annual revenue after adjustment for projected inflation over this period.
The estimated price elasticities, −0.77 for the total population, are within the range of price elasticity estimates available for LMICs.8 The price inelastic demand for cigarettes evident in the less than unitary absolute value indicates that while price increases cigarette consumption decreases less than proportionately. As a result, total expenditure on cigarettes increases. If a tax increase induces the price increase, tax revenue is expected to increase.
The total expenditure elasticity is 0.43, in line with evidence from other LMICs. It suggests that cigarettes are a ‘normal good’: with increases in total spending, households increase their spending on cigarettes. In addition, the analysis by level of total expenditure indicates that households with lower total expenditure increase their spending on cigarettes to a greater extent.
The 5-year roadmap used in the simulation of tobacco tax increase was worked out by the authors in consultation with the Ministry of Finance of El Salvador keeping three considerations of reform in the tobacco tax system in El Salvador in view:
Introducing an autoadjustment process of increase in the specific component of the mixed excise system in keeping with the expected annual rate of inflation and income growth to prevent the real value of the specific excise component from falling behind the growth of people’s income and affordability of tobacco products.
Delivering a shock to cigarette consumption in the economy through a significant tax-induced price increase in the first year, followed by smaller increments in subsequent years to keep the momentum of the shock.
Approaching two-thirds’ tax share of the retail price of cigarettes by the end of the 5-year roadmap.
While two-thirds’ tax share falls short of the WHO-recommended level of 70% excise tax share or 75% total tax share, achieving the higher tax share was deemed politically infeasible to implement in a short time period. Hence, the tax proposal adopted a moderate approach for tax reform keeping the long-term goal of achieving 75% total tax share in perspective.
It is worrying that since the signing of the FCTC, no tax increases have been implemented on tobacco products and El Salvador is missing the opportunity to reduce smoking and smoking-related deaths and diseases. A tax increase would also generate an important increase in tax revenues, as shown here, which in turn could be used for nationwide tobacco control programmes and help avert the social costs associated with tobacco consumption. A recent study estimated the health costs attributed to tobacco use in El Salvador at over US$263 million, more than seven times the tax revenues currently generated by tobacco.31
This study has limitations. First, data used for the analysis came from 2005/2006, as a more recent household expenditure survey is not available. Second, the limitations of Deaton’s AIDS model hold in the present study, such as potential biases introduced by using unit values instead of actual prices.7 The absence of prices in the survey for each brand consumed in El Salvador makes it impossible to test if working with unit values introduced such biases and, if so, to determine their direction and magnitude. Third, there are unavoidable uncertainties around projections of inflation and growth rates. We used well-known, respected sources that may produce different projections compared with what finally occurs. If, for instance, actual economic growth is lower (higher) than it has been projected, one can expect a higher (lower) reduction in the projected prevalence, as a consequence of the tax increase. This is due to the fact that, given an income elasticity, lower economic growth would imply a demand that grows less.
Increasing tobacco taxes has the potential to decrease consumption in El Salvador and raise fiscal revenues. The tobacco tax in El Salvador is among the lowest in Latin America and the social costs of tobacco consumption largely exceed tobacco tax revenues. An increase in tobacco tax could significantly reduce smoking and the associated burden of tobacco-caused diseases and deaths.
What this paper adds
This is the first study that estimates cigarette demand elasticities using household-level data for a Central America and the Caribbean country. In addition, it is one of the first in Latin America that provides such elasticities by socioeconomic groups.
This study adds to the growing body of literature that uses Deaton’s Almost Ideal Demand System model to estimate cigarette demand price, expenditure and quality elasticities.
The results of simulation of a tobacco tax increase and its expected impact on public health and government revenue in the short to medium term are provided.
Data availability statement
Data are available in a public, open access repository. Data are available upon reasonable request. Secondary publicly available data were used.
Contributors GP and ADP conceptualised the study. GP, DA and NN conducted the statistical analysis and contributed substantially to the interpretation of the findings. GP, DA, ADP and NN wrote the article. GP is the guarantor of the study.
Funding This work has been funded by the International Union Against Tuberculosis and Lung Disease (The Union) and the Pan-American Health Organization.
Competing interests None declared.
Provenance and peer review Not commissioned; externally peer reviewed.