Article Text
PDF
Abstract
Background Tobacco prevalence in Colombia is small compared with other Latin America despite the nation’s tobacco taxes being among the lowest in the region. However, tobacco taxes have increased several times during the last decade, and large increases in 2010 and 2016 impacted consumer prices.
Objective This paper aims to estimate the price smoking participation elasticity (PPE) in Colombia, with specific reference to regional increases in consumer prices after 2010 tax policy changes.
Methods The PPE is computed using logistic regression based on individual-level data from the National Psychoactive Substances Consumption Survey for 2008 and 2013. Our specific focus is state-level variation in Colombian cigarette prices between 2008 and 2013 induced by the tax hike in 2010.
Results The estimated PPE in Colombia is around −0.66 (p value=0.046). We find almost no differences across socioeconomic level, but price sensitivity was greater for women than men, and for relatively older individuals (ages 51–64).
Conclusions PPE for Colombia is above estimates for comparable middle-income countries such as Mexico. As a result, current estimates for health gains of tax policies are likely to be underestimated. Moreover, in contrast with the literature, we find that the PPE for the youth (≤25 years) is lower than older age groups, and there is no evidence of a prominent socio-economic status (SES) gradient.
- economics
- low-income and middle-income countries
- price
- socioeconomic status
- taxation
Statistics from Altmetric.com
Introduction
Tobacco consumption is a leading cause of ill-health globally, and although the prevalence of smoking in Colombia is among the lowest in the region, associated morbidity nevertheless is significant.1 In recent decades, the government of Colombia has implemented policies to prevent, discourage and reduce smoking. Particularly, increases in tobacco taxes have increased consumer prices.2 Tobacco consumption is an important public health issue, and preventive regulatory actions can substantially influence aggregate smoking in the long-term.3–6 Despite a diverse range of tobacco control policies as the MPOWER measures recommended by the WHO, excise taxes remain the most effective strategy to discourage tobacco consumption but not at the recommended rates.7 8 However, excise taxes may affect different populations differently.9
Promotion of the tobacco control agenda in public policy requires tools that simulate the impact of tax hikes. However, current exercises are based on estimates of the number of cigarettes consumed per day (intensive margin) for the total population.1 10 While such estimates are appropriate in a general sense, it is impossible to provide specific-group impacts that are desirable for inequality analysis and projections. Moreover, current national estimates are restricted to the intensity of consumption, namely the reduction in total cigarette consumption in response to price changes. This limits the ability of researchers to assess potential impacts on the prevalence of tobacco use (extensive margin).
The goal of this paper is to estimate of the cigarette price smoking participation elasticities (PPEs) for Colombia by age, income and gender groups, using household data from 2008 and 2013, while simultaneously exploiting state-level price variations driven by a tax hike in 2010. For this research, we use the National Psychoactive Substances Consumption Survey 2008 and 2013 (NPSCS) and average annual cigarette prices at the state level from the National Department of Statistics. Tobacco price elasticity is computed using average marginal effects from logistic regression.
Franks et al explored how tax impacts on tobacco consumption in the USA included both cessation and reduced intensity.11 Moreover, the impact on low-income populations was found to be less than expected, and some studies found low-income populations to be relatively insensitive. This study opened questions on the importance of separately studying the impact on smoking prevalence, and on doing so across population groups. In the same country, differences in PPE across populations are still relevant: Yao et al estimates such differences as being −0.26 for whites, −0.10 for African–Americans, −0.42 for Asians and −0.11 for Hispanics.12 Wilson et al reviewed impacts for several countries and found figures ranging from −0.45 to 0.1 for adults and −1.41 to −0.1 for youth.13 The higher responsiveness of youth is well known in the literature: in the USA estimates of PPE estimates for the USA range from −0.145 to −1.51.14–17 This is important because delayed initiation of smoking correlates with early cessation.3 4
For middle-income countries, while there are figures on the responsiveness of smoking participation to price for countries such as Russia and South Africa, actual estimates of PPE are scarce.18–20 An estimate of the PPE of −0.06 was obtained for China.21 A similar pattern occurs for Latin America (LA): there are several cigarette consumption price-elasticity estimates,22 but few exist for PPE. This is mainly because current estimates are based on supply-side information due to the lack of periodic household surveys that include information on individual smoking patterns. The closest references for us are two exercises for Mexico using household expenditure surveys. First, Miera-Juárez and Iglesias estimated PPE of −0.17, while Jiménez-Ruiz et al estimated PPE at around −0.06.23 24 An important caveat is that if one is interested in individual smoking choices, it is important to consider that household-based PPE estimates underestimate individual sensitiveness. For instance, if in a household, there are two smokers and only one-stop smoking as a result of the price increase, this would not be taken into account in the calculation of the estimate. As a local reference, total elasticities for the Colombian case haven been estimated on −0.36 by Santa María and Rozo, on −0.44 by James et al and on −0.78 by Maldonado et al.1 25 26 For Colombia, Santa María and Rozo estimate an elasticity of −0.36 using data from NIELSEN for the period between 2000 and 2007 for 15 cities, combined with price variation at brand-quarter level from 2000 to 2007.25 Maldonado et al estimated the same figure to be −0.78 using tax office sales records, and yearly national prices from the statistics department between 1994 and 2014.26 Using household data and the tobacco consumer price index (CPI), James et al estimate price elasticity on −0.44.1 Generally, sensitivity of cigarette consumption in LA mainly lies within 0 and −0.5.22
Context
To curb the tobacco epidemic, the Colombian government has implemented diverse control mechanisms, which have contributed significantly to decreasing tobacco use. Taxation is known to be the most cost-efficient policy, and the country undertook several tax reforms between 1997 and 2016. Several limited tax increases occurred under different regimes after 1995, taking such forms as specific-contributions to sports’ budget, custom tariffs and VAT and other consumption taxes. In 2006, as part of the WHO Framework for Tobacco Control, Law 1111 unified ad-valorem consumption tax, earmarked part of it towards sports promotion and created a specific tax that depended on the final consumer price: cheaper and expensive brands division, using a reference price for the entire country. In 2010 a major reform occurred when specific taxes for cheaper and expensive brands were unified. These legislative changes increased the tax burden on cheaper cigarettes, which historically have dominated the market, and simultaneously slightly reduced the tax burden on high-end tobacco products.27 Other antitobacco policies, such as advertising bans and smoke-free environments, were implemented between 2009 and 2011.
Data
The NPSCS is a cross-sectional study conducted by the Ministries of Health and Justice, focused on consumption of alcohol, tobacco and illicit drugs among individuals aged 12–65 years of age, living in either metropolitan areas or urban areas with populations exceeding 30 000. The results are representative of 27 administrative areas (departamentos) which are equivalent to states. We consider the 2008 and 2013 waves, which follow a standard structure defined by the Organization of American States. The Sistema Interamericano De Datos Uniformes Sobre Consumo De Drogas is a protocol for household surveys on legal and illegal drug consumption inLA. An individual is considered a tobacco user if they have consumed more than 100 cigarettes in their life, and the data further records age of initiation, as well as frequency and intensity of use during the last 12 months and the last 30 days.
As shown in table 1, by 2008 the prevalence of tobacco consumption during the last 30 days among the population was 17.3% (24.3% for men, 11.2% for women) with substantial state-level differences. The average age of initiation of tobacco use was 16.83 years. In terms of the analysis we consider socioeconomic groups using the variable estrato, which classifies the quality of life of different locations in the country, as a proxy of SES, according to six levels (ranging from 1 to 6, with 1 indicating the lowest SES and 6 the highest) to assign cross-subsidies in costs for public utilities. We group this variable into three categories corresponding to low, middle and upper SES.
Descriptive statistics
Figure 1 presents the average smoking prevalence and age of initiation for 2008 and 2013 by state. In terms of prevalence, large decreases occur in Bogotá, Caldas, Nariño and Valle, while the age of initiation remains constant.
Prevalence and Smoking Starting Age, by Year and StateIncluded departamentos (states) are ANT:Antioquia, ATL: Atlántico, BOG: Bogotá, BOL: Bolívar, CAL: Caldas, COR: Córdoba, HUI: Huila, MET: Meta, NAR: Nariño, NSA: Norte de Santander, RIS: Risaralda, STD: Santander, VAL: Valle del Cauca. Own calculations based on the National Psychoactive Substances Consumption Survey, for yeas 2008 and 2013.
State prices were constructed based on the National Income and Expenditure Survey (known by the Spanish acronym ENIG) 2006–2007. Using data from 2007q2 household reported prices and tobacco and cigarette CPI time series, we obtain state series for cigarette real prices from 2000q1 to 2016q4. Because our analysis runs until 2013q4, we restricted information for that quarter. We also controlled left and right outliers of the distribution, taking the sample from its 1st to its 99th percentile. The change in taxation implied an increase of the real inflation-adjusted average price per cigarette of nearly 60% during the considered period. Figure 2 shows the state-level evolution of prices per stick, which had a mean of COP (Colombian Peso) 133.6 (SD=16.32) in 2008 and of COP 160.7 (SD=20.1) in 2013. In US dollar, in 2008 a stick was around US$0.06, and in 2013 it became US$0.08. Roughly, a 20 pack move from US$1.2 to US$1.7. To show how prices evolve, we divide the graph into two subfigures. The first shows the evolution of tobacco prices for Antioquia, Atlántico, Bolivar, Meta and Risaralda, all of which have higher average cigarette prices during the period under observation. The second shows the increase in tobacco prices for Bogotá, Caldas, Córdoba, Huila, Nariño, Santander, Norte de Santander and Valle, all of which have lower average cigarette prices during the study period. There are differences in prices at the state-level because of several factors such as distribution and retailing costs, but also on the presence of brands that were categorised as expensive or cheap according to the 2006 tax scheme. Therefore, it is expected that the 2010 reform had a differential impact by state. While brand shares and industry decisions over the passthrough might be impacted by this reform, smoothing part of the potential price increases, overall, we observe variation on the final price at state-level. We observe larger price increases for some states, such as Meta, Antioquia and Bolivar, and hence these states represent the main source of variation.
Individual cigarette prices over time per stateSource: Own derivation of prices based on state level figures from 2007Q2 in the Colombian national income and expenditure survey (ENIG) which were projected using official cigarette CPI time series.
Empirical strategy
Our objective is to estimate how variation in consumer prices within departments affects the probability of an individual to be an active smoker at all. Particularly, we care about the probability of a person being classified as a smoker based on their characteristics. We compute PPE using a logistic regression, where the probability of being a smoker depends on a linear combination of local cigarette prices, socioeconomic status controls. As well, it includes dummies to control for municipality characteristics that differ across states but remain constant across time, and dummies for the month and year of the interview at the NPSCS survey, to control for any temporal aggregate shock.
(1)
Equation 1 presents the logistic regression that approximates the conditional probability that individual i living in municipality m at department j and surveyed in period t is classified as a current smoker. In this expression, 
represents the logistic transformation, and the matrix Ximjt
indicates that the probability varies with local-state prices Pjt
, individual and household characteristics 
. The estimate of the parameter of interest, 
, represents how a marginal increase in prices will be reflected in smoking probability and is identified by substantial variation in local prices among departments from 2008 to 2013. We argue that the main source of such variation is the differential impact of the 2010 taxation hike across states, which affected them differently based on the initial presence of brands.
In order to compute elasticities, we consider the average marginal effect estimated from equation 1, relative to the observed prevalence of smoking in 2008 and the average national price for that year as presented in equation 2. For specific groups (by age, gender and SES) we use the relevant prevalence for each case.
(2)
Results
Table 2 shows estimates for the average marginal effect of 
in equation 1. Column (1) consider the general prevalence model. We find that the smoking participation decreases in 0.8% points with a price hike of COP 10, which is around two-thirds of a SD of the 2008 price. From columns (2) to (4), we present heterogeneous effects on sociodemographic characteristics. In column (2) we present a specification that allows for differential effects for young (25 years old or less), adult (26–50 years old) and middle-age individuals (51 and older). We find that the effect of prices on smoking participation increases with age group. Column (3) shows the differential effect of price changes on smoking prevalence for men and women. Finally, column (4) allows for differential effects by SES. The next step is to assess if these differences are translated into elasticities.
Average marginal effects of cigarette price variations on the probability to report being a smoker after logistic regressions
Table 3 presents the estimated price elasticities according to equation 2. Panel A presents the prevalence for each group, its elasticity and respective p value. Panel B presents the statistical test for the equivalence of elasticities between subpopulations. The general population average PPE is around −0.66. Restated, a 10% increase in the price of cigarettes would result in a 66% decrease in the prevalence of smoking. Concerning age, we find an elasticity of −1.33 for middle-age individuals. For young and adult individuals, the value is considerably smaller, around −0.47. It is important to notice that these figures are imprecisely estimated; nevertheless, the difference between the mature age group and the younger ones is statistically significant at the 99% confidence level. One possibility is the role of the minimum age included in the dataset (10 years old). If it is too low, the low variation obtained for younger individuals might be to the very low prevalence for the youngest individuals. Online supplementary figure A1 in the appendix shows that the point estimates are robust to the minimum age considered in the dataset. As for gender, there is a substantial difference between the estimates, showing an elastic response for women ass opposite to men (−1.32 vs −0.33). Finally, by SES, there is no statistical difference between the estimates.
Supplemental material
PPE estimates
In the online supplementary appendix, we explore differences in smoking initiation and cessation with the following auxiliary analysis. First, we split the sample between teenagers and young adults (age ≤25) versus adults (age >25). Second, we restricted the youth group to those who are neither smokers nor started smoking during the last 5 years; and the adults’ group to those who were active smokers at least 5 years ago. Repeating the analysis described above, estimates with the first group allow us to compute initiation elasticities (online supplementary table A2), while the second does it for cessation elasticities (online supplementary table A4). As a general rule, cessation elasticities exceed the main estimate, and initiation elasticities are smaller. These results complement the analysis above by emphasising that, in our sample, young individuals are less responsive to prices than older ones.
Limitations
There are at least three essential points that we should consider before discussing the results presented above.
Even if this is the first exercise to estimate PPE at an individual-level in Colombia, NPSCS dataset imposes important challenges for this exercise. The first consideration is that we only have two points in time, which do not allow us to control for differential demand state-level trends which might bias the estimate.
Second, the lack of actual prices in the NPSCS dataset required the usage of state-level aggregate prices. This imputation means that we do not observe the exact price that each respondent paid for their cigarettes, a figure which might change according to SES. This lack of price variability within states might be one of the reasons behind the lack of a gradient over the SES dimension if price increases were larger for poorer individuals and smaller for richer ones. A similar pattern can also apply to the age dimension, as younger individuals might choose cheaper cigarettes than older smokers. This differential pattern on price changes according to initial prices should be expected from the tax reform that we are considering. The reason for this is that cheap cigarettes prices might have increased more as a result of the unified specific tax, which might even have benefited some expensive brands. However, it is relevant to mention that having individual-level data would not be enough to estimate the gradient, and an endogeneity problem will arise given the choice of consumers.
Third, SE from our estimates are large. These results might conceal non-prominent gradients as it becomes harder to reject the specific null hypothesis. It also might be behind the large difference observed between men and women. While it might be true that women are more sensitive to prices than men, the point estimate for women is very large (−1.36) while for men, it is very small (−0.35). Nevertheless, in both cases, a PPE of −0.66 (the overall PPE) is contained in the 95% CIs of both estimates, For men the 95% CI is (−0.81 to 0.125), for women it is (–2.37 to –0.34) indicating us that the actual differences could be smaller than what appears based only on the point estimates.
Conclusions
The estimated tobacco PPE in Colombia is around −0.66. Similar figures from Mexico, which have less affordable cigarettes than Colombia,9 has a lower PPE (−0.06 and −0.17), which suggest that Colombian smokers remain sensitive to tobacco prices, showing that there exists room for further excise tax increases.
This estimate of PPE could be compared against recent cigarette tax increases between 2016 and 2017. Real prices have increased by around 28.2% for a packet of 20 cigarettes, and 23.1% for loose cigarettes.2 Given this increase, we should expect a decrease in smoking prevalence between 14.55% and 18.05%. The National Quality of Life Survey (NQLS) estimates the smoking prevalence at 10% in 2016 and 8.7% in 2017 for individuals 18 years and older. These figures represent a reduction of around 12% in smoking prevalence. In this sense, our estimates of the PPE are close to the proportional change calculated from the NQLS. The small difference can be explained by the fact that the reduction in prevalence between 2016 and 2017 not only includes the pressure from the reform but also of additional factors like demand trends or income changes for the recent periods.
Our estimates found a slightly lower PPE for young people in Colombia. This finding differs from the literature, which has shown a higher elasticity for young people. As well, we do not find evidence of a gradient of the elasticity over SES, which is another established patter in the developed countries literature. There are two potential reasons for this. The first one, when discussing PPE, the gradient should not be as prominent as in the intensity of consumption elasticities as intensity tends to be more responsive than participation. In other words, individuals will first reduce the number of cigarettes consumed than stop smoking at all. As a result, poorer households might react more than richer ones on the number of cigarettes than they consumed, but their decision to stop smoking might be as sensitive for both groups. A second explanation for this result is the high affordability of tobacco for all income groups in Colombia at the time of price change in 2010, as consequences of the low prices of cigarettes. Even if the poorest smokers exhibit high affordability, it makes them react to prices in the same way that their richest counterpart.
Our findings above have implications on policy simulations that are based on these gradients. For instance, James et al assumed a sharp total elasticity gradient over income quintiles (−0.61 to −0.26). In this case, the gradient is for total elasticity, and it uses quintiles of income instead of our measure of SES which is aggregated in three categories. While under those conditions, gradients should be more prominent by construction, part of the distributional benefits provided by tax policy is based on such assumption over the gradient. Nevertheless, under our estimates total elasticities should be larger and with a gradient on SES. Therefore, James et al are probably underestimating the total health gains of the policy while overestimating its distributional effects. This implication motivates the need for improving research on equality studies of tobacco responsiveness in the region.
These results above indicate that on top of taxes, which work well with experienced smokers, it would be required to use other strategies that target the youth and the poor. For instance, strengthening and fully implementing the cessation programme designed by the Ministry of Health and Social Protection, and regulating the advertising of e-cigarettes which are targeted to the youth and might increase usage of common tobacco products.28 29
What this paper adds
Evidence, mainly from high-income countries, documents that a gradient exists for the effectiveness of price increases in reducing smoking over both age and socioeconomic status.
Few studies have estimated price smoking participation elasticity (PPE) in middle-income countries, and estimates based on individual or household data are especially lacking.
Current simulation methods for predicting policy efforts extrapolate results on the sensitiveness of smoking consumption to price for subpopulations using estimates from high-income countries.
We estimate a PPE of −0.64 for Colombia. As a result, predicted health gains associated with tax policies are currently underestimated.
Our estimates of price sensitivity do not differ by SES, and hence currently predicted gains on health distribution due to tax policies might be overestimated.
Acknowledgments
We thank Edanz Group (www.edanzediting.com/acwww.edanzediting.com/ac) for editing a draft of this manuscript. We also acknowledge the valuable research assistance given by Leonel Eduardo Criado Meneses.
References
Supplementary materials
Supplementary Data
This web only file has been produced by the BMJ Publishing Group from an electronic file supplied by the author(s) and has not been edited for content.
Footnotes
Twitter @JuanMGallego1, @androdri
Contributors Each author contributed in the same amount to this project.
Funding This project was funded under the GADC project by the CIHR/IDRC [grant number 108442-001]. Data and code for replicating these exercises can be obtained at: https://github.com/androdri1/PPEColombia.
Competing interests None declared.
Patient consent for publication Not required.
Provenance and peer review Not commissioned; externally peer reviewed.
Data availability statement Data are available in a public, open access repository. Scripts and final datasets are available in the repository: https://github.com/androdri1/PPEColombia.