The Impact of Prices and Taxes on the Use of Tobacco Products in Latin America and the Caribbean

Abstract

We examined the impact of tobacco prices or taxes on tobacco use in Latin America and Caribbean countries. We searched MEDLINE, EconLit, LILACS, unpublished literature, 6 specialty journals, and reviewed references. We calculated pooled price elasticities using random-effects models.

The 32 studies we examined found that cigarette prices have a negative and statistically significant effect on cigarette consumption. A change in price is associated with a less than proportional change in the quantity of cigarettes demanded. In most Latin American countries, own-price elasticity for cigarettes is likely below  −0.5  (pooled elasticities, short-run: −0.31; 95% confidence interval = −0.39, −0.24; long-run: −0.43; 95% CI = −0.51, −0.35).

Tax increases effectively reduce cigarette use. Lack of studies using household- or individual-level data limits research’s policy relevance.

Among the many challenges facing health systems in low- and middle-income countries (LMICs) is the increasing burden of noncommunicable diseases. In 2010, more than one third of the 34.5 million deaths attributed to noncommunicable diseases occurred in LMICs.¹ Tobacco use—a major risk factor of noncommunicable diseases—is worryingly high in many Latin American countries. Chile, for example, has one of the highest smoking prevalences in the world (in 2010, 44% of men and 38% of women were current smokers).² The tobacco health toll is evident: in Chile, deaths attributable to tobacco use exceeded 15% of all deaths in 2009.³

Increasing tobacco prices has been found to be the single most effective method to reduce smoking.^4–6 Yet, it appears that relatively little work has been conducted using data from countries of Latin America and the Caribbean: a recent comprehensive review that the International Agency for Research on Cancer conducted identified only 6 studies.⁶ Additionally, reviews^4–11 that examine the impact of prices and taxes on the use of tobacco products provide limited quality assessment of the data and methods used and have generally weaker generalizability to LMICs. There are exceptions (examples include Rice et al.,⁸ Bader et al.,⁹ and Guindon,¹¹ who attempt to conduct some quality assessment of individual studies).

We systematically searched for and critically reviewed studies that examined the impact of tobacco prices or taxes on tobacco use in countries of Latin America and the Caribbean. We paid particular attention to the data and statistical approaches used.

METHODS

In the development and operation of the review, we used as a methodological guide the Assessment of Multiple Systematic Reviews (AMSTAR) assessment measurement tools developed by Shea et al.^12,13 Although AMSTAR was not designed to assess the quality of individual studies, we felt it offered useful guidance in identifying key attributes that require clear reporting or assessment. Studies that examined the effects of prices or taxes on health behaviors, such as tobacco use, for the most part use methodological approaches that are overlooked in quality assessment or reporting tools, such as the Quality Assessment Tool for Quantitative Studies,¹⁴ and Strengthening the Reporting of Observational Studies in Epidemiology,^15,16 or tools designed with economic evaluation in mind, such as the checklist of Drummond et al. for assessing economic evaluations,¹⁷ the Consolidated Health Economic Evaluation Reporting Standards statement,¹⁸ and the checklist of Philips et al. for assessing decision-analytic modeling in health technology assessment.¹⁹

Although we did not use a priori methods of assessment, we extracted from each study detailed data and methodological information and generally assessed the quality of the data and methods used in each study. We did not use quality scales for assessing quality or risk of bias, as empirical evidence does not support them (different scales often result in different conclusions, scales may include criteria that are not related to risk of bias, weighting may be ill justified, and the interpretation of numerical scores can be difficult); the Cochrane Collaboration explicitly discourages quality scales.²⁰

Criteria for Considering Studies

We considered all studies that quantitatively examined the relationship between prices of or taxes on tobacco products and tobacco use. We included all studies regardless of the publication type (e.g., peer-reviewed journals, book chapters, reports from government agencies or nongovernmental organizations, working articles).

We included all studies regardless of date of publication or data collection.

We included all studies from Latin America and the Caribbean regardless of the geographic coverage (e.g., state, province, municipality). We excluded studies conducted outside Latin America and the Caribbean. We included cross-country studies that included countries of Latin America and the Caribbean.

We included all studies regardless of the language of publication.

We included all measures of tobacco use: initiation or onset, participation, consumption, cessation, substitution, escalation, or persistence. We also included studies that examined aggregate outcome measures, such as national cigarette consumption or sales.

We excluded studies that did not clearly report SEs, t tests, or statistical significance; that did not clearly present the data, methods, or results; that arbitrarily manipulated the data; or that assumed in their statistical approach that all else was held constant, including income.

Search Methods for Identification of Studies

We searched the following computerized bibliographic databases: MEDLINE via PubMed, EconLit via ProQuest, and LILACS. We searched LILACS in English, Spanish, and Portuguese. We searched unpublished literature via Google and Google Scholar. We searched 6 specialty journals by hand (Addiction, Health Economics, Journal of Health Economics, Nicotine & Tobacco Research, Revista Panamericana de Salud Pública, and Tobacco Control), and we examined references of reviews.^{4–11,21–24} We searched 2 series of conference programs and abstracts: the World Conference on Tobacco or Health and the Society for Research on Nicotine and Tobacco Annual Meetings.

We last conducted searches on November 12, 2013. We contacted 20 key informants in March 2013 and sought help from participants at a workshop on tobacco control economic research in Latin America in December 2013. (The search strategy we employed for bibliographic databases is available as a supplement to the online version of this article at http://www.ajph.org.)

Review Methods

The review process had 4 stages, each of which at least 2 independent researchers conducted:

1. We screened studies identified in the electronic database and by hand search for relevance.

2. We assessed relevant studies for inclusion.

3. We extracted data using a standardized form.

4. We analyzed the extracted data and synthesized them into user-friendly tables.

We extracted the following study characteristics, where applicable: (1) authors, year of publication, country, journal; (2) methods (statistical analyses, model type, functional form); (3) data (type, sample size, population, missing data or data adjustments, source); (4) a description of the dependent variable; (5) a description of the price or tax measure and, where applicable, how the price or tax measure was adjusted for inflation; (5) covariates; (6) testing for misspecification; (7) sensitivity analyses; (8) results, price or income effects, and statistical significance; (9) a brief overview of potential limitations; and (10) whether the sources of support were clearly acknowledged.

Studies that examined the effect of prices or taxes on demand typically reported effect sizes in the form of elasticities (e.g., price and income elasticities). Elasticities are unitless and represent a measure of the responsiveness of a variable to a change in the value of another variable; specifically, an elasticity is the ratio of the percentage change in the former to the percentage change in the latter. An own-price elasticity of demand measures the responsiveness of the demand for a good or service to a change in its own price (i.e., the ratio of the percentage change in quantity demanded to the percentage change in price), holding all else equal. Similarly, a cross-price elasticity of demand represents a measure of the responsiveness of the demand for a good or service to a change in the price of another good or service.²⁵ In economic models of addiction, short- and long-run price elasticities are often computed. The short-run elasticity holds past consumption constant, whereas the long-run elasticity allows past consumption to vary. As a result, the long-run effect of a change in price will exceed the short-run effect.²⁶

We used random-effects models to pool results across studies. Relative to fixed-effects models, random-effects models allow interstudy variability and are more conservative (i.e., confidence intervals [CIs] are wider).²⁷ We calculated an I² (the ratio of true heterogeneity to total observed variation) as a measure of heterogeneity.^28,29 We generated forest plots using tools developed by Neyeloff et al.³⁰ We weighted intrastudy and interstudy effect sizes on the basis of the same or very similar data so that their contribution to the overall effect size was equivalent to a single study. We excluded effect sizes obtained using similar statistical approaches but fewer data from the meta-analyses.

RESULTS

Our search of bibliographic databases yielded 215 potential articles (PubMed: 54; EconLit: 15; LILACS: 146), 25 of which we selected for further investigation. Key informants provided an additional 28 potential articles, 10 of which we selected for further investigation. We identified a further 16 studies via hand and gray literature searches. The review of full articles yielded a total of 32 studies, a substantially larger number than reviewed in any other single study or review. Of the 32 studies, we identified 10 as having poor methodology or reporting. Of the remaining 22 studies, 5 used global cross-country data (i.e., they were not limited to Latin America and the Caribbean), 15 used country-specific aggregate-level data, and 2 used country-specific household-level data. We did not identify any study that used individual-level data from Latin America and the Caribbean. The flow diagram of study selection is presented in Figure 1.

FIGURE 1—

Flow diagram of study selection: Latin America and the Caribbean, 2013.

Figure 2 plots price elasticity estimates from studies that used country-specific aggregate time series and cross-sectional household-level data. We present only estimates for which enough information was provided to construct 95% CIs. Price elasticity estimates are grouped by data type (first cross-sectional, then time series) and studies are chronologically ordered vertically by country name and within country by date of publication. Short- and long-run estimates are presented separately.

FIGURE 2—

Estimates of own-price elasticity for cigarettes: Latin America and the Caribbean, 2013.

Note. 2SLS = two-stage least-squares; 3SLS = three-stage least-squares; ECM = error correction model; GARCH = generalized autoregressive conditional heteroskedasticity; GMM = generalized method of moment; IV = instrumental variables; OLS = ordinary least squares; VECM = vector error correction model.

^aOverall short-run estimate excludes González-Rozada (2006), Martinez et al (2008) and Iglesias, Nicolau (2006), OLS myopic.

^bOverall long-run estimate excludes Martinez et al (2008).

On the whole, the studies we reviewed indicated that cigarette prices have a negative and statistically significant effect on cigarette consumption. Effect sizes, however, vary substantially across studies and, at times, within studies. Evidence from aggregate time series analyses suggests that there is a statistically significant negative association between cigarette prices and cigarette consumption in Latin America. Studies from Argentina,^31–34 Chile,³⁵ and Mexico³⁶ pointed to short- and long-run price elasticity estimates in relatively narrow ranges (−0.10 to −0.30 and −0.25 to −0.45, respectively). Studies from Brazil^37–39 and Uruguay⁴⁰ found somewhat higher short-run price elasticity estimates: −0.20 to −0.60; long-run price elasticity estimates, however, vary widely across studies and specifications: −0.4 to −1.4. Studies from Panama,⁴¹ Guatemala (Gutiérrez M, Lic unpublished data, 2010), Colombia (Ariza M, MSc Llorente B, Lic Curti D, Lic unpublished data, 2010), and Bolivia⁴² found relatively high short- and long-run price elasticities in relatively wide ranges (−0.60 to −0.8 and −0.40 to −1.0, respectively). The only study from a Caribbean country (Jamaica)⁴³ generally found that prices have a statistically significant and negative effect on cigarette consumption, but effect sizes varied widely across specifications; these results, however, should be interpreted with caution, as the authors ignored the potential for nonstationarity and spurious regression.

Overall, we found that higher prices reduced total cigarette consumption. Our pooled analyses yielded short- and long-run own-price elasticities for cigarettes that were negative and statistically significant (short-run: −0.31; 95% CI = −0.39, −0.24; long-run: −0.43; 95% CI = −0.51, −0.35).

We identified only 2 studies that used country-specific household-level data. These studies used very similar methods and data. Sáenz de Miera Juárez et al.⁴⁴ used 11 waves of the Mexico Encuesta Nacional de Ingresos y Gastos de los Hogares (Mexico National Household Income and Expenditure Survey) conducted between 1994 and 2012 and a 2-part model to estimate both participation and consumption own-price elasticities. Sáenz de Miera Juárez et al. found prices to have a relatively small but significant effect on participation (−0.17) and a larger effect on consumption (−0.40). Households of lower socioeconomic status did not appear to be substantially more responsive to changes in prices. Jimenez-Ruiz et al.⁴⁵ used similar methods and 7 waves of the Mexico Encuesta Nacional de Ingresos y Gastos de los Hogares conducted between 1994 and 2005. They found that prices had a small but significant effect on participation (−0.06) and a larger effect on consumption (–0.45).

Table 1 presents a brief description of each study that we included. Studies are presented by data type and methodology, grouped by countries in alphabetical order; when we identified more than 1 study for a given country, we presented the studies in chronological order on the basis of the publication year. (A detailed synthesized overview of each study that we included in this review is available as a supplement to the online version of this article at http://www.ajph.org.) The descriptions provided represent our interpretation and are not necessarily the interpretations of the authors of the primary studies.

TABLE 1—

Summary of Studies: Latin America and the Caribbean, 2013

Study	Country	Data	Methods: Statistical Approach; Model Type	Limitations and Risks of Bias
Country studies: household-level data; 2-part model
Jimenez-Ruiz et al., 200845	Mexico	Repeated cross-sectional (1994, 1996, 1998, 2000, 2002, 2004, and 2005)	2-part model:	No testing for misspecification; no account for measurement error or endogeneity
		Sample size = 109 089 households	Participation: probit
			Consumption: weighted OLS
Sáenz de Miera Juárez et al., 201344	Mexico	Repeated cross-sectional (1994, 1996, 1998, 2000, 2002, 2004, 2005, 2006, 2008, 2010, and 2012)	2-part model:	No testing for misspecification; no account for measurement error or endogeneity
		Sample size = 196 089 households	Participation: probit
			Consumption: weighted OLS
Country studies: aggregate data; time series analyses
Gonzáles-Rozada, 200631	Argentina	Time series, monthly (1996–2004)	OLS	Short temporal dimension (9 y); limited set of control variables; unclear description of consumption measure
			Static, dynamic OLS
Martinez et al., 200833	Argentina	Time series, monthly (1994–2004)	VECM	Short temporal dimension (11 y); unclear description of cigarette consumption data; limited set of control variables
Martinez et al., 201332	Argentina	Time series, monthly (1994–2010)	VECM	Limited set of control variables
Gonzáles-Rozada and Rodriguez Iglesias, 201334	Argentina	Time series, monthly (1996–2012)	OLS; ECM	Limited set of control variables; unclear description of consumption measure
Carvalho and Lobão, 199839	Brazil	Time series (1983–1994)	OLS; 2SLS	Short temporal dimension (12 y); limited set of control variables; unclear if any tests for misspecification were conducted (e.g., unit root or cointegration); unclear if instruments are appropriate; significance level of long-run estimates not reported
			Myopic addiction, rational addiction
Iglesias and Nicolau, 200638	Brazil	Time series, quarterly (1991–2003)	OLS; 2SLS	Short temporal dimension (13 y); limited set of control variables; unclear if instruments are appropriate; significance level of long-run estimates not reported
			Myopic addiction
Iglesias et al., 200737	Brazil	Time series, quarterly (1991–2005)	OLS	Short temporal dimension (15 y); limited set of control variables; unclear if instruments are appropriate; significance level of long-run estimates not reported
			Myopic addiction
Alcaraz, 200642	Bolivia	Time series, annual (1988–2002)	2SLS	Small sample size (df < 15); limited set of control variables; unclear description of price and consumption measures; unclear if instruments are appropriate; significance level of long-run estimates not reported
			Static, myopic, and rational addiction
Debrott Sánchez, 200635	Chile	Time series, quarterly (1993–2003)	GARCH	Short temporal dimension (11 y); limited set of control variables; limited testing for misspecification (e.g., no testing for the presence of unit roots or cointegration); unclear if instruments are appropriate; significance level of long-run estimates not reported
			Static and myopic addiction
Ariza M, MSc, et al., unpublished data, 2010	Colombia	Time series, quarterly (1994–2009)	ECM; 2SLS; GMM	Limited set of control variables; significance level not consistently reported; unclear if instruments are appropriate
			Myopic addiction
Gutiérrez, Lic, unpublished data, 2010	Guatemala	Time series, monthly (1997–2009)	OLS	Short temporal dimension (13 y); limited set of control variables; price and income effects not robust to alternative specifications; unclear if instruments are appropriate; significance level of long-run estimates not reported
			Static
van Walbeek et al., 200543	Jamaica	Time series, annual (1974–2001)	OLS	Small sample size (df < 30); unclear if any tests for misspecification were conducted (e.g., unit root or cointegration); limited set of control variables; unclear if instruments are appropriate; significance level of long-run estimates not reported
			Static, myopic addiction
Olivera-Chavez et al., 201036	Mexico	Time series, quarterly (1994–2005)	OLS	Short temporal dimension (12 y); limited set of control variables
			Static, dynamic OLS
Herrera Ballesteros, 201241	Panama	Time series, quarterly (1999–2009)	OLS; 2SLS; ECM	Short temporal dimension (11 y); limited set of control variables; unclear if instruments are appropriate; significance level of long-run estimates not reported
			Myopic, rational addiction
Ramos and Curti, 200640	Uruguay	Time series, quarterly (1991–2003)	SURE with IV; 2SLS; 3SLS	Short temporal dimension (13 y); unclear if instruments are appropriate; significance level of long-run estimates not reported
			Myopic addiction
Cross-country studies
Blecher, 200846	51 LMICs, including Chile, Colombia, Ecuador, Guatemala, Mexico, Panama, Paraguay, Peru, and Uruguay	Time series–cross-section (1990–2003, unbalanced);	OLS	No testing for misspecification (in particular, no testing for unit roots or cointegration)
		Sample size = 617; t = 14 y; i = 51 countries
Kostova et al., 201147	17 LMICs, including Brazil, Chile, Costa Rica, Mexico, Peru, and Venezuela	Repeated cross-sectional; GYTSa (1999–2006)	2-part model	No testing for misspecification; unclear how consumption variable was constructed; unclear how price and survey location were matched
		Sample size = 315 353	Participation: logit
			Consumption: GMM (distribution and link not reported)
Kostova and Blecher, 201248	30 LMICs, including Argentina, Brazil, Chile, Colombia, Ecuador, Guatemala, Mexico, Panama, Peru, Uruguay, and Venezuela	Repeated cross-sectional; GYTSa (1999–2006)	2-part model	No testing for misspecification; unclear how consumption variable was constructed; unclear how price and survey location were matched
		Sample size = 342 926	Participation: IV-logit
			Consumption: IV-GLM (negative binomial distribution, log link)
Nikaj, 201249; Nikaj and Chaloupka, 201350	30 LMICs, including Argentina, Brazil, Chile, Colombia, Ecuador, Guatemala, Mexico, Panama, Peru, Uruguay, and Venezuela	Repeated cross-sectional; GYTSa (1999–2008)	2-part model	Unclear how consumption variable was constructed; unclear how price and survey location were matched
		Sample size = 518 009	Participation: logit
			Consumption: GLM (γ distribution, log link)
Kostova, 201251	48 countries (40 LMICs, including Argentina, Chile, Costa Rica, Guatemala, Mexico, Panama, Paraguay, Peru, Uruguay, and Venezuela)	Retrospective; GYTSa (1999–2006)	Split population duration models	No covariates in participation component of duration models; no interpretation of price elasticity estimates provided; unclear how smoking histories were constructed; unclear how prices were matched with smoking histories
		Sample size = 476 304; mean number of periods modeled: onset = 6.7; cessation = 3.8	Probability of ever starting or quitting: logit
			Duration: log-normal

Note. 2SLS = 2-stage least squares; 3SLS = 3-stage least squares; df = degrees of freedom; ECM = error correction model; GARCH = generalized autoregressive conditional heteroskedasticity; GMM = generalized method of moments; GYTS = Global Youth Tobacco Survey; IV = instrumental variable; LMIC = low- and middle-income country; OLS = ordinary least squares; SURE = seemingly unrelated regression equations; VECM = vector error correction model.

^aGenerally nonrepresentative national or subnational sample of schoolchildren and adolescents.

We identified 5 studies that used cross-country data not limited to Latin America and the Caribbean (data available as a supplement to the online version of this article at http://www.ajph.org). Blecher⁴⁶ used time series–cross-section data from 51 LMICs, including Chile, Colombia, Ecuador, Guatemala, Mexico, Panama, Paraguay, Peru, and Uruguay, and found a small statistically significant association between prices and per capita cigarette consumption (−0.10). Kostova, Ross et al.,⁴⁷ Kostova and Blecher,⁴⁸ Nikaj and Chaloupka,^49,50 and Kostova⁵¹ used data from the Global Youth Tobacco Surveys. Kostova et al.⁴⁷ used data from 17 LMICs, including Brazil, Chile, Costa Rica, Mexico, Peru, and Venezuela, where the Global Youth Tobacco Surveys had been conducted more than once, and found very high participation and consumption elasticities (–0.75 to −1.1 and −1.4 to −1.7, respectively). Nikaj and Chaloupka^49,50 used data from 30 LMICs, including Argentina, Brazil, Chile, Colombia, Ecuador, Guatemala, Mexico, Panama, Peru, Uruguay, and Venezuela, and also found very high participation and consumption elasticities (–0.6 and −1.8, respectively). Nikaj and Chaloupka, however, did not find that prices are associated with smoking prevalence among girls. Kostova and Blecher⁴⁸ used data from 19 low-, middle-, and high-income countries, including Brazil, Chile, Costa Rica, Mexico, Peru, and Venezuela, where the Global Youth Tobacco Surveys had been conducted more than once. Although price elasticities were not reported, reported marginal effects suggest high and statistically significant participation and consumption elasticities. It is important to note that the youths’ own-price elasticities are typically not comparable to the price elasticity estimates that we have reported elsewhere in this review, as the definition of youths’ smoking prevalence (i.e., smoked at least 1 cigarette in the past month) often differs from that of adult populations.

Kostova⁵¹ used Global Youth Tobacco Survey data from 48 low-, middle-, and high-income countries, including Argentina, Chile, Costa Rica, Guatemala, Mexico, Panama, Paraguay, Peru, Uruguay, and Venezuela, and duration analyses to explore the impact of cigarette prices on smoking onset (the transition between never smoker and first experimentation) and cessation. Higher cigarette prices are generally found to delay first experimentation; however, no interpretation was given regarding the sizes of estimates found, so it was difficult to ascertain if effect sizes were meaningful because of the limited information provided. Cigarette prices were not found to affect youths’ smoking cessation.

The demand for tobacco products, such as cigarettes, may be more sensitive to changes in tobacco prices in LMICs because of relatively low individual and household incomes and low levels of consumption.⁵² In 1999, the World Bank concluded that the “estimates of elasticity vary from study to study, but there is reasonable evidence that in middle-income and low-income countries, elasticity of demand is greater than in high-income countries. . . . For LMICs as a whole, then, a reasonable estimate of the average elasticity of demand would be –0.8, based on current data.”^53(p41) A more recent nonsystematic but comprehensive review conducted by the International Agency for Research on Cancer concluded that there was limited evidence that the demand for tobacco products in low-income countries is more responsive to price than is the demand for tobacco products in high-income countries.^6,54 We have not provided evidence to support the claim that the own-price elasticity for cigarettes is higher in LMICs than the consensus own-price elasticity estimates for high-income countries of about −0.4.

Moreover, in Latin American and Caribbean LMICs, there does not appear to be a strong price effect gradient. As a measure of income, we used per capita expenditure-side real gross domestic product at chained purchasing power parity (2005 US dollars) averaged over the period used in the regression analyses.⁵⁵ (We plot the price elasticity estimates presented in Figure 2 ordered vertically according to income [from high to low] in a supplement to the online version of this article at http://www.ajph.org.) The price elasticity for cigarettes does not appear to be substantially lower in higher-income and higher-consuming countries of Latin America, such as Argentina, Mexico, Uruguay, and Chile, relative to lower-income, lower-consuming countries, such as Bolivia and Guatemala. However, the data presented may suggest the existence of a price effect gradient, albeit a weak one.

Only 1 study examined cross-price effects. Ramos and Curti⁴⁰ examined the effect of changes in the price of loose tobacco on the demand for cigarettes and found positive and statistically significant cross-price elasticities. Put differently, cigarettes and loose tobacco appear to be substitutes in Uruguay; an increase in the price of loose tobacco is expected to increase the demand for cigarettes. Income was found to have a positive and statistically significant effect on cigarette consumption. Income effect sizes varied substantially across studies, but on the whole, both short- and long-run income elasticity estimates fell in ranges that were well below unity (i.e., a change in income was associated with a less than proportional change in the quantity demanded of cigarettes). On the whole, income elasticities for cigarettes fell in a relatively narrow range clustered around 0.5, which suggests that, on average, a 10% increase in income would lead to a 5% increase in total cigarette consumption.

Studies With Poor Methods or Reporting and Excluded Studies

Numerous studies had important limitations, some serious enough to make the results uninterpretable or uninformative. Several studies did not clearly report SEs, t tests or statistical significance^56–60; 2 studies failed to clearly present the data, methods, and results (Llorente B, Lic, Maldonado N, PhD, unpublished data, 2010)⁶¹; 1 study arbitrarily manipulated the data⁵⁷; and 2 studies^62,63 assumed in their statistical approach that all else was held constant, including income. (We extracted the data for these studies, and they are presented as a supplement to the online version of this article at http://www.ajph.org.)

Three studies that did not meet the inclusion criteria merit discussion. First, Saenz de Miera et al.⁶⁴ attempted to assess the impact of a 2007 cigarette tax increase in Mexico. Using the first 2 waves of the Mexico International Tobacco Control Evaluation Project—a cohort survey of adult smokers conducted in 4 cities (Mexico City, Guadalajara, Tijuana, and Ciudad Juárez) between September and November 2006 and between November and December 2007—daily cigarette consumption and quit behavior were compared at baseline (September through November 2006) and follow-up (November through December 2007). Cigarette consumption was found to have decreased between waves 1 and 2 but only among users of more than 5 cigarettes per day, whereas 13% of smokers reported having stopped smoking for at least 1 month at follow-up. The lack of trend data made it impossible to attribute the tax changes to changes in consumption and cessation.

Second, Saenz de Miera Juarez,⁶⁵ using the third (October 2008 through January 2009), fourth (January through February 2010), and fifth (April through May 2011) waves of the Mexico International Tobacco Control Evaluation Project, examined whether and, if so, to what extent cigarette taxes were passed on to users. On the whole, tobacco taxes were generally accompanied by price increases, but price increases differed by brand categories (national vs international brands).

Third, 1 study, using data from Chile, estimated a differentiated product discrete-choice demand model.⁶⁶ Using cigarette brand–specific prices and market shares, Agostini found a total price elasticity of the market shares that was extremely high (–1.72; 95% CI = −1.94, −1.46), which would suggest that Chilean smokers are unusually responsive to changes in cigarette prices. Agostini’s estimate, however, is not comparable to the total own-price elasticity estimates we have reported and that are reported more broadly in the literature. Agostini’s estimate represented the effect of a change in price on market shares and not on total volume. Furthermore, consumers (i.e., smokers) were restricted to substituting other brands in proportion to market shares (i.e., smokers cannot quit or even reduce their consumption). In the differentiated product discrete-choice demand model, an “outside good” or “outside alternative” needs to be specified. In such models, consumers need not be restricted to cigarette smoking. For example, Tan,⁶⁷ Min,⁶⁸ and Pham and Prentice⁶⁹ used the decision not to smoke cigarettes as the outside alternative.

DISCUSSION

We found cigarette prices to have a negative and statistically significant effect on cigarette consumption. Estimates of own-price elasticity for cigarettes varied substantially within and between studies. On the whole, both short- and long-run estimates fell in ranges that were almost certainly below unity (in absolute value); that is, a change in price was associated with a less than proportional change in the quantity demanded of cigarettes. In higher-income Latin American countries, own-price elasticity for cigarette is likely below  −0.5 ; a 10% increase in price would be expected to reduce the demand for cigarettes by less than 5%. The price elasticity for cigarettes does not appear to be substantially higher in lower-income and lower-consuming Latin American countries relative to higher-income and higher-consuming countries. The evidence base, however, is too thin to rule out the possibility that the demand for cigarettes in low-income countries is more responsive to price than is the demand for cigarettes in high-income countries.

Assessment of Risk of Bias and Limitations

In interpreting results, numerous risks of bias and limitations should be kept in mind. An important limitation of studies that use aggregate time series data (all but 2 country studies in our review) is the impossibility of disentangling the effect of price changes on smoking onset, participation, consumption, or cessation. Additionally, such studies do not permit the examination of price responsiveness by individual characteristics such as gender, age, and socioeconomic status. Socioeconomic status is of particular importance, as the lack of sustained increases in taxation is often in part owing to misconceptions about economic harm to the poor.

Several studies used data of short temporal dimensions. Studies that used monthly, quarterly, and annual time series data had temporal dimensions between 5 and 17 years, 11 and 17 years, and 10 and 28 years, respectively. Only 4 studies used data that covered more than 15 years.^32,34,43,56 The use of monthly and quarterly data augments the sample size, but more complicated dynamics may become important.

Nearly all studies that used aggregate time series data used a very limited set of control variables. For example, only 1 study⁴⁰ included in this review examined cross-price effects, even though such data are often readily available (i.e., have the same source as do cigarette price data).

Cigarette tax evasion can bias price elasticities estimates obtained using aggregate time series data derived from legal sales or consumption of licit cigarettes; the substitution toward tax-evaded cigarettes will be wrongly recorded as a drop in sales or consumption.^70,71 The direction of the bias is not necessarily upward (as the tobacco industry argues). For example, a cigarette tax increase combined with an increase in tax enforcement may well bias elasticity estimates downward (i.e., there would be substitution from tax-evaded cigarettes toward licit cigarettes). It is important to note that the size of the cigarette tax evasion market is less relevant than are changes in its size as a result of price changes.

Some studies^35,43,46 that use aggregate time series data ignore the potential for nonstationarity and spurious regression (i.e., misleading correlations that occur because of common trends in the data and not because of a true economic relationship). If the cigarette price variable or the cigarette consumption variable is nonstationary, a regression of cigarette consumption on prices may be spurious and, hence, yield invalid results. If both cigarette consumption and prices are nonstationary, there may be a stationary linear combination of these variables, implying that they are cointegrated. Only if both the price and consumption variables are each stationary or if they are cointegrated can one be confident that the results are not spurious.

Price elasticity estimates may be biased because of the endogeneity of the price variable. A variable is deemed endogenous if it is a function of parameters or of variables in the model. Econometrically, a variable is endogenous if it is correlated with the disturbance (i.e., the error term of the regression).⁷² Endogeneity may occur because of omitted variables, measurement error, or simultaneity. Simultaneity can arise when the independent variable, price, is jointly determined with the dependent variable, consumption, through an equilibrium mechanism (e.g., supply and demand). For example, the use of cigarette price data could introduce an endogeneity problem if heavier smokers tended to purchase cheaper cigarettes (e.g., by purchasing cigarettes by the carton).⁷³ In a situation in which price is an endogenous regressor, instrumental variable estimation can be used as a solution to the endogeneity bias. However, to use an instrumental variable estimator, one must find a suitable instrument. A good instrument is correlated with the endogenous regressor (e.g., price) for reasons that are intuitive and verifiable but uncorrelated with the outcome variable (e.g., cigarette consumption or smoking) for reasons beyond its effect on the endogenous regressor.⁷⁴ Lagged variables are often used as instruments, but their use can be problematic (see Angrist and Krueger⁷⁴ for more details). Six studies used an instrumental variable approach (Ariza M, MSc, et al., unpublished data, 2010). ^38,41,42,66 Instruments used, on the whole, were a mix of lagged cigarette prices, cigarette taxes, exchange rates, and tobacco producer prices.

Most, but not all, studies used conventional demand models as well as models that take into account addiction, such as the myopic addiction model. These models produce both short- and long-run price elasticity estimates. None of the studies, however, reported enough information to assess the uncertainty about long-run price elasticity estimates (i.e., it is not possible to assess the significance level of long-run estimates). Similarly, 1 study interacted the price variable,⁵⁶ whereas another study used a lag of the price variable⁷⁵; neither of these 2 studies, however, reported the necessary information to assess significance level.

Some studies estimated rational addiction demand models. Estimates from these models should be interpreted with caution. The rational addiction model in general and 1 of its key assumptions, perfect foresight, have been criticized.^5,76,77 Additionally, Auld and Grootendorst⁷⁸ argue that aggregate time series data are particularly ill-suited for the empirical analysis of the rational addiction model.

Nearly all studies we included used a ln–ln functional form, which implies constant elasticities, and such an assumption may not be valid for time series data. Because of the time span in most of these studies, elasticities are typically identified by small year-to-year changes in prices. Consequently, these estimates may not accurately predict the likely effects of large price changes. It is important to note, however, that own-price elasticity for cigarettes tends to be constant over a fairly wide range of prices and price increases.⁶

Limitations

Our review has limitations. First, because of the lack of quality assessment tools and the heterogeneity in the methods used, we did not use a priori methods of assessment. Second, numerous studies we reviewed provided limited data and methodological information, which rendered quality assessment difficult. Although at least 2 reviewers performed data extraction and assessments, readers of our review are urged to refer to original studies and not to rely uncritically on the descriptive information we have provided online.

Fourth, several studies used similar data and methods, so the number of independent estimates was smaller than was the number of studies. For example, using data from Argentina, Gonzáles-Rozada,³¹ Martínez et al.,³³ Martinez et al.,³² and Gonzáles-Rozada and Rodriguez Iglesias³⁴; using data from Brazil, Iglesias and Nicolau³⁸ and Iglesias et al.³⁷; and using data from Uruguay, Ramos and Curti⁴⁰ and Ramos Carbajales and Curti⁵⁸ employed very similar data and methods. Fifth, our pooled analysis of short-run elasticities showed substantial variability in the magnitude of effects (I² = 88%). Although it is common practice to pool results in the presence of significant heterogeneity, inferences associated with pooled estimates are weaker.²⁷

Conclusions

Our findings confirm the effectiveness of higher cigarette prices in reducing cigarette use. Increasing the price of cigarettes by increasing taxes can also be expected to increase cigarette tax revenue. In fact, in countries of Latin America, any politically feasible tax increases can be expected to generate increased tax revenues. As the economies of Latin America and the Caribbean are expected to grow in the short to medium term,⁷⁹ the positive association between income and cigarette consumption suggests that, everything else being equal, higher incomes will more than likely lead to higher cigarette consumption.

Our results point to numerous lessons. First, our review demonstrates the benefit of searching systematically. We identified more studies than any other single study or review. Systematic reviews (especially those with a focus on LMICs) should not restrict the search to English nor should the search be restricted to peer-reviewed journals; both published and unpublished sources of literature ought to be explored. Funding agencies should encourage their recipients to publish in peer-reviewed journals. The use of incentive or requirements should be considered.

Second, many reviews pay little or no attention to quality assessments. Efforts by organizations such as the International Agency for Research on Cancer and the World Health Organization to synthesize the literature are to be commended, but such efforts should rely less on inventories of studies and expert opinions and more on accepted systematic review methods.^20,80 The importance of assessing statistical and economic significance independently from authors’ interpretations cannot be overemphasized. For example, all studies included in our review that used a myopic addiction approach failed to present enough information to assess the statistical significance of long-run elasticity estimates. The same is true for all studies that interacted the price variable.

Third, there is a lack of studies that use household- or individual-level data. Studies that use aggregate time series data cannot disentangle the effect of price changes on smoking onset, participation, consumption, or cessation, nor can they examine price responsiveness by individual or household characteristics. This is important because the health benefits of reduced cigarette use occur primarily through cessation rather than reduced consumption in smokers.^81,82 Similarly, the evidence base for Latin American and Caribbean countries is far too limited to make any assertions regarding differences in price responsiveness across socioeconomic status. The contribution of additional studies that use similar aggregate time series data are far outweighed by the potential contribution of studies that use household- or individual-level data and that explore price responsiveness across socioeconomic status.