Smokers’ Strategic Responses to Sin Taxes: Evidence from Panel Data in Thailand

Abstract

In addition to quitting and cutting consumption, smokers faced with higher cigarette prices may compensate in several ways that mute the health impact of cigarette taxes. This study examines three price avoidance strategies among adult male smokers in Thailand: trading down to a lower-priced brand, buying individual sticks of cigarettes instead of packs, and substituting roll-your-own (RYO) tobacco for factory-manufactured cigarettes. Using two panels of microlevel data from the International Tobacco Control Southeast Asia Study, collected in 2005 and 2006, we estimate the effects of a substantial excise tax increase implemented throughout Thailand in December 2005. We present estimates of the marginal effects and price elasticities for each of five consumer behaviors. We find that, controlling for baseline smoking characteristics, socio-demographics, and policy variables, quitting is highly sensitive to changes in cigarette prices, but so are brand choice, stick-buying, and use of RYO tobacco. Neglecting such strategic responses leads to over-estimates of a sin tax’s health impact, and neglecting product substitution distorts estimates of the price elasticity of cigarette demand. We discuss the implications for consumer welfare and several policies that mitigate the adverse impact of consumer responses.

1. INTRODUCTION

1.1. Motivation

Despite the overwhelming evidence that tobacco taxes affect smoking behavior (Chaloupka and Warner, 2000; IARC, 2011), the policy has not been without its detractors. One commonly raised objection is the potential regressivity of tobacco taxes, whereby poor smokers bear a disproportionate share of the financial burden (Colman and Remler, 2008). A less studied challenge to cigarette taxes is the compensatory behavior of smokers. A rational consumer will weigh the expected benefits of quitting against the expected physical, psychic, and social costs of ending the dependence, both discounted to the present.¹ Many studies confirm that a price increase alters this calculus, spurring some smokers to cut consumption or quit altogether, consistent with the tax’s intended public health impact (Tauras and Chaloupka, 1999; Laxminarayan and Deolalikar, 2004; Goel, 2007).

Other consumer responses may undermine the potential of tobacco taxes to lessen smoking-related harm. This study considers the extent to which—controlling for smoking characteristics, socio-demographics, and policy variables—a tax increase in Thailand induced smokers to behave in one of three ways: 1) switching to a cheaper cigarette brand, 2) purchasing cigarettes by the stick rather than by the pack, and 3) substituting from manufactured cigarettes to handrolled tobacco. Neglecting these consumer responses may cause researchers and policymakers to overestimate the health impact of the tax and, in the case of product substitution, to distort estimates of the price elasticity. Some past studies, detailed below, have examined these compensating behaviors, but only a handful have tracked their adoption in less-developed countries and none have looked at the breadth of outcomes considered here.

This study makes several contributions. We track the compensatory behaviors resulting from a substantial increase in the cigarette excise tax in Thailand. We find large behavioral responses to the tax. Consumers adopt all three compensatory behaviors: substitution toward cheaper brands, substitution toward buying cigarettes by the stick, and substitution toward handrolled tobacco. To our knowledge, we present the first econometric evidence that price influences cigarette package size, namely the decision to buy cigarettes by the stick. Our results are especially applicable to countries where: the sale of individual sticks is common or multiple pack sizes are available; a large share of smokers use non-cigarette products; or a large price spread exists between discount and high-end cigarette brands. In other words, the findings apply to much of the developing world.² The paper closes with a discussion of the implications for consumer welfare and setting tobacco policy.

1.2. Behavioral responses to taxation

This subsection reviews selected price avoidance strategies that smokers undertake in response to cigarette taxes and that fail to improve the consumers’ health. First, some smokers lower their tobacco consumption, but regulate the intensity with which they smoke each cigarette in order to maintain the dose of nicotine (Benowitz, 1999; Adda and Cornaglia, 2006). A smoker may inhale smoke more deeply or cover the ventilation holes with his or her mouth or fingers (Benowitz et al., 1986; Kozlowski, Pope, and Luz, 1988). Intake of nicotine, tar, and carbon monoxide would not decrease proportionately with cigarette consumption. The difficulty in measuring these behaviors has prevented researchers from tracking them on a population level. An important exception is Adda and Cornaglia (2006) who use a marker of nicotine intake in a large sample of smokers to measure this compensatory behavior, finding that tax increases lead smokers to adjust their intensity of smoking. We lack such granular data.

Second, a smoker may switch to another brand of cigarettes following a price increase. Some evidence suggests that smokers switch to brands with higher levels of nicotine or tar in order to maintain levels of nicotine intake while smoking less (Evans and Farrelly, 1998). In developing countries, where smokers may be highly price-sensitive, increased cigarette prices may compel a smoker to switch to a lower-priced brand (Tsai et al., 2005; White et al., 2013), especially in places where a large price spread exists across different brands. This down-trading limits the effectiveness of tobacco taxes in discouraging participation (López, 2002), and allows existing smokers to maintain daily intake without incurring higher costs. In some cases, down-trading entails switching from licit to illicit cigarettes. The production and distribution of illicit cigarettes as a supply-side response to cigarette taxation increases the availability of illicit cigarettes and influences the purchasing decisions of smokers (Merriman, 2010; Stehr, 2005).³ However, an increase in the specific excise tax on cigarettes may lead to substitution toward more expensive brands, or up-trading. This type of tax increase reduces the relative price of high- versus low-quality cigarettes, in effect shifting demand to the untaxed product attribute, quality (Barzel, 1976; Sobel and Garrett, 1998; Espinosa and Evans, 2013). As we discuss later, specific and ad valorem taxes have different effects on the relative price of premium and discount cigarette brands and, thus, lead to different predictions about the expected direction of brand substitution.

Third, a smoker may substitute toward a cheaper tobacco product. Cross-price effects on tobacco use have been found in a number of contexts (Guindon et al., 2011; Hanewinkel, Radden, and Rosenkranz, 2008; Pekurinen, 1989; Ohsfeldt, Boyle, and Capilouto, 1997; Wangen and Biørn, 2006). In Thailand, nearly half of all smokers roll their own cigarettes from loose tobacco (Young et al., 2008). Roll-your-own (RYO) tobacco is believed to be an inferior good in Thailand (Young 2006),⁴ whereby some smokers switch to RYO tobacco as its price decreases relative to the price of manufactured cigarettes. For example, falling income levels in the aftermath of the Asian financial crisis of the 1990s contributed to a large share of cigarette users in Thailand taking up RYO tobacco (Dalvey Group, 2004).

Fourth, smokers may buy cigarettes by the stick rather than by the pack or carton. Wertenbroch (1998, 2001 advances one possible motivation, namely as a form of self-rationing among “sophisticated” smokers who are aware of a self-control problem. Self-rationing is likely a second-order consideration in our context. Rather, credit-constrained smokers often substitute individual sticks for packs in order to delay or avoid trying to quit smoking, especially in countries where this practice is legal or common.⁵ One illegal vendor of ‘loosies’ in New York City puts it succinctly, “The tax went up, and we started selling 10 times as much. [The city’s mayor] thinks he’s stopping people from smoking. He’s just turning them onto loosies” (Goldstein, 2011). The academic literature has tended to overlook this phenomenon. In one notable exception, a small study in the Philippines finds two countervailing trends among people switching to singletons: it allowed some smokers to continue to consume the same brand even if they deemed a pack too expensive, and it prompted others to switch to imported brands as the absolute price difference between a single domestic and a single imported cigarette was smaller than the price difference for whole packs (Chen, 1997). Many countries ban the sale of individual sticks, but Thailand continues to allow the practice.

1.3. Tobacco Pricing and Tax Structure in Thailand

The Thai government heavily regulates the tobacco industry, exercising price controls over manufactured cigarettes. The Excise Department administers an ad valorem tax on cigarettes and sets a ceiling on retail prices for each price segment. The excise tax rate has steadily increased from 60% of wholesale price in 1994 to 85% in 2010 (Visaruthvong, 2010). In contrast, roll-your-own tobacco is taxed at negligible levels, with no excise tax for native-grown loose tobacco and a specific excise tax of 1 baht (USD 0.029) per kilogram for most non-native loose tobacco.

The Thai government estimates that smoking prevalence among men declined from 44% in 2003 to 39% in 2006 (Thailand NSO, 2007). Researchers have yet to give a full accounting of the reasons for this dip, although simulations attribute 61% of the decline in smoking from 1991 to 2006 to taxation (Levy et al., 2008). Our study isolates the effect of a substantial tax increase during this period on changes in smoking behavior in Thailand.

From 2001 to 2005, nominal cigarette prices and the corresponding excise tax rate did not budge—real prices fell with rising inflation. Not surprisingly, tobacco sales and consumption grew each year during this period, according to industry data (ERC, 2007). In December 2005, the government increased the cigarette excise tax from 75% to 79% of wholesale price.⁶ Subsequently, the mean cigarette price increased 9.0% and the most popular brand rose an average of 13 percent, both in real terms. The difference in these two figures point to variation in firms’ pricing response to the tax increase. All brands passed on the tax increase to consumers, but some used the tax as an opportunity to pass through additional costs. This paper tries to identify the effect of the tax using producer-level and geographic variation. Overall, the policy translates to a rise in the cigarette tax as a percentage of retail price from 63.7% to 68.8%.⁷

2. STUDY DESIGN AND METHODS

2.1. Empirical Model

We use two panel estimation approaches to identify consumers’ responses to Thailand’s tax increase in January 2006: models with community-specific random intercepts and with individual fixed effects. Breusch-Pagan tests of independence indicated that, for all outcomes, these panel data approaches are preferred to pooled ordinary least squares (OLS). In both approaches, we capture the characteristics that influence quitting and then the characteristics that influence consumption in submodels conditional on continuing to smoke at wave 2. Unlike some studies (e.g., Laxminarayan and Deolalikar, 2004; Wangen and Biørn, 2006), we model these processes separately to avoid the assumption that variables affect participation and consumption in the same way. All analyses are restricted to adult smokers and are clustered by each of the 16 markets, defined as province by urbanicity.⁸ Standard errors are corrected for a general, unknown form of heteroskedasticity using a Huber-White sandwich estimator (White, 1980).

The first approach using community-level random effects assumes that unobserved variation is normally distributed and uncorrelated with observed predictors. The random effects model has several advantages. It is more efficient than a fixed effects model and allows for the inclusion of time-invariant predictors of interest. A series of heteroskedasticity-robust Hausman tests of random versus fixed effects at the community and individual levels for each submodel reveal that the random effects model is inconsistent for models of cigarette consumption and brand choice and consistent for the model of stick-buying. For comparability with the fixed effects models, we estimate the random effects models as linear probability models. Logit models yielded similar estimates. OLS estimation of Equation (1) determines the probability of each smoking behavior. Step 1 of the first approach derives the probability of quitting from the following latent variable formulation:

$${\mathit{QUIT}}_{\mathit{ijt}}^{\ast }={\beta }_1+{\beta }_2\mathrm{\Delta }{\mathit{PRICE}}_{\mathit{ijt}}+{\beta }_3{\mathit{PRICE}}_{ij(t-1)}+X_{ij}\gamma +{\varsigma }_j+{\epsilon }_{\mathit{ijt}}$$ (1)

where the propensity for quitting smoking at time t for individual i living in community j is a function of the change in cigarette price, the baseline price, a set of baseline characteristics, a community-specific random intercept, and random error ε_ijt (distributed normal). The socio-demographic characteristics and policy constraints captured by the vector X are detailed below. The random intercept ς_j ~ N (0, ψ) accounts for unobserved heterogeneity between communities, where communities are defined as the survey’s primary sampling unit.

Step 2 of the first approach specifies separate models of the probability of each of four behaviors, conditional on smoking at wave 2: cigarette consumption, brand choice (brands in low-price versus high-price segments), the package size of cigarettes last purchased (packs or cartons versus individual sticks), and tobacco product type (manufactured cigarettes versus roll-your-own tobacco). The propensity for each behavior is represented as follows:

$$({\mathit{BEHAVIOR}}_{\mathit{ijt}}^{\ast }\mid {\mathit{QUIT}}_{\mathit{ijt}}=0)={\beta }_1+{\beta }_2{\mathit{PRICE}}_{\mathit{ijt}}+X_{ij}\delta +{\phi }_j+u_{\mathit{ijt}}$$ (2)

which includes the same baseline covariates, a community-specific random intercept (φ_j), and an error term (u_ijt). The equations for three of four outcomes are implemented using a linear probability model. The fourth behavior—tobacco product type—is modeled as a multinomial logit function to account for three tobacco product types: cigarette use only, RYO tobacco use only, and use of both cigarettes and RYO tobacco. Likelihood ratio tests indicated that none of these outcome categories should be combined (χ² > 200, p < 0.001 for all pairs of alternatives). Maximum likelihood estimation of the multinomial random effects model is performed using adaptive quadrature (with 30 integration points), which leads to stable results relative to most other integration methods (Rabe-Hesketh, Skrondral, Pickles, 2005).

The second estimation approach captures how changes in price affect behavior using a fixed effects estimator. Unlike the random effects estimator that uses within- and between-community (and thus, within- and between-market) variation in cigarette prices across waves, the fixed effects estimator relies only on within-community (and thus, within-market) variation to identify the quasi-experimental effects of the tax increase. As such, fixed effects estimators are less susceptible to omitted variable bias than are random effects estimators. Given two time periods, the fixed effects estimator is equivalent to including a first differences model. Individual fixed effects remove the effect of all influences that vary between individuals and between time-specific events that affect the outcome, and identifies behavioral changes using within-person price variation. We implement the fixed effects specifications as linear probability models.⁹ As before, submodels are estimated conditional on quit status. The multinomial model of tobacco product type is not compatible with the fixed effects approach, and is only estimated using random effects.

A drawback of the person-level fixed effects estimator in this context is that it may lead to noisy results because of the limited sample size and the large number of degrees of freedom used in estimation. For comparison, we also present a community fixed effects model that isolates the effects of within-community price variation. Communities are nested within markets, and so this model only uses within-market price variation. If the community fixed effects model has much smaller standard errors than the person-level model, it would imply that the latter lacks sufficient power for inference.

We report price coefficients as marginal price effects and price elasticities for ease of interpretation.¹⁰

2.2. Data and Variables

This study employs the first two waves of the International Tobacco Control Southeast Asia Survey. The survey draws respondents using a stratified, multi-stage sampling design, in which the primary strata consist of Bangkok and the four regions of Thailand (Hammond et al., 2008). Two provinces were selected within each region. A secondary stratification consisted of urban and rural districts within each province.¹¹ Survey questions were identical across the two rounds, although socio-demographic information was only collected at baseline. Baseline information was collected from January to February 2005 during in-person interviews with Thai adults age 18 and older. The follow-up survey was conducted August to September 2006. Of those interviewed in wave 1, interviewers re-contacted 78% in wave 2. This sample attrition is comparable to many other household surveys fielded in developing countries (Alderman et al., 2000). Compared to those interviewed in both waves, non-responders tended to be older, less educated, and more rural, and as a result more likely to use RYO tobacco that costs less on average. The magnitude of these differences is small; selective attrition should not greatly bias the results.

The survey was limited to regular smokers, defined as individuals who smoked at least weekly and smoked more than 100 cigarettes in their lifetime. Our analysis excludes female respondents, who comprise a small, possibly non-representative portion of the sample, only 5.3% of continuing smokers. In addition, cultural norms in Thailand may lead some women to systematically under-report their tobacco use. Overall, 1,436 men were interviewed in both waves. After implementing complete case analysis, the balanced panel includes 1,422 men.

The analysis includes five dependent variables: 1) quitting, 2) daily cigarette consumption, 3) decision to buy cigarettes by the pack rather than by the pack, 4) price segment of the cigarette brand most frequently purchased, and 5) type of tobacco product. Quitting is based on self-reported smoking status at wave 2. Other dependent variables are limited to continuing smokers. Cigarette consumption is defined as average self-reported daily intake, including factory-made and handrolled cigarettes. The variable for buying packs or cartons versus individual sticks—is generated from self-reports. Brand choice is based on self-reports of the most commonly smoked brand at each wave.¹² We divided brands into three price segments, or price tiers, based on a review of industry sources (ERC, 2007). Due to the small percentage of smokers who selected the most expensive tier (about 1% in each wave), we combined the mid- and high-price tiers. Respondents reported use of cigarettes, RYO tobacco, or both. The models for brand choice and package size include only the subset of smokers who used factory-manufactured cigarettes at both waves, excluding RYO tobacco users. Thus, these analyses have smaller starting populations. Similarly, a sub-analysis of RYO cigarettes includes only those individuals who smoked cigarettes at wave 1, excluding all others.

Our models include two measures of market cigarette prices and one measure of market RYO tobacco prices. The tobacco price variables are constructed from self-reported prices, scaled to the equivalent of a standard 20-cigarette pack, and reported as real prices in 2005 baht.¹³ Substantial price variation exists across waves, brands, and manufactured versus RYO products. We assume that these market prices are exogenous to individual consumers. The market is defined as the mean province-byurbanicity price of cigarettes in a given wave. Our first measure of cigarettes prices, used in our econometric models of quit status, cigarette consumption, and choice of tobacco product is the overall mean market price of cigarettes. Those models also include a measure of the mean market price of RYO tobacco. Missing at baseline, the price of RYO tobacco was generated using the average change in the price of pipe tobacco in Thailand between the two time points, based on data from the Economist Intelligence Unit (EIU, 2007). Many users of RYO tobacco grow their own product or smoke unbranded, locally grown leaf, yielding more stable prices over time, which are less sensitive to the factors shifting cigarette prices. The small price change for RYO tobacco made loose tobacco a relatively better bargain than cigarettes following the tax increase.

The second measure of cigarette prices is the average market price of a consumer’s brand in a given wave.¹⁴ One issue that arises in calculating the brand-specific price is that the self-reported price at wave 2 is endogenous if a person’s brand choice at that time reflects the brand-specific changes in price. Some studies using micro-level data have used an endogenous price measure (e.g., Tsai et al., 2005).¹⁵ To circumvent this endogeneity problem, we use the brand-specific mean market price at wave 2 of the brand smoked at wave 1, which is plausibly exogenous to the smoker.¹⁶ One could argue that smoking patterns exhibit a high inter-temporal correlation, implying that tobacco prices at a given point in time are endogenous. While this logic may hold for baseline prices, the price change of the baseline brand is the result of an exogenous shock that each consumer must take as given. However, our brand-specific price measure may still be endogenous to the extent that tobacco producers passed through more of the excise tax to consumers who were expected to be less price-sensitive. This reverse causality would lead us to overstate the effect of the tax on price avoidance behavior. We cannot observe the ex ante price sensitivity of consumers, but manufacturers do not appear to have differentially set the pass-through rate for smokers in different price segments of the market. For example, the dominant low-price brand (Wonder) increased its price by 12.0% following the tax. The other five most popular brands, all belonging to the mid-price segment and listed in descending order of popularity, raised their prices by 12.7% (Krongthip), 6.9% (Saifon), 11.2% (L&M), and 13.1% (Krungtong). Producers’ pricing strategy likely reflects competitive considerations more than the expected price response of consumers.

Table 1 provides basic descriptive statistics. We selected background characteristics for inclusion in the analysis based on their significance in previous studies and their hypothesized association with the compensating behaviors under consideration. Smoking characteristics include average daily cigarette consumption at baseline and the type of tobacco used at baseline (mainly RYO tobacco or not). The theory of health capital suggests that an individual’s demand for health—and by extension, behavioral—decisions about smoking—is related to age, education, and income (Grossman, 1972). Age also captures the duration of time a person has smoked (Pearson’s ρ of 0.93), which is a predictor of a person’s level of addiction. A large literature shows that a socio-economic gradient exists for health behaviors, including smoking decisions, where socio-economic status (SES) is typically proxied by educational attainment and family income (Cawley and Ruhm, 2011; Cutler and Lleras-Muney, 2010; Grossman and Kaestner, 1997). We control for categorical educational attainment (primary or less, secondary, beyond secondary), annual household income at baseline in 100,000 baht, and baseline residence in an urban area.

Table 1.

Descriptive statistics

	Wave 1 Mean (SD)	Wave 2 Mean (SD)	Difference in means	t-test of (1) vs. (2) (p-value)	N
	(1)	(2)	(3)	(4)	(5)
Dependent Variables
Quit (vs. not)	--	0.127	--	--	1,422
Cigarettes per day	14.27 (8.50)	11.94 (8.43)	−2.33	<0.001	2,172
Price tier
Tier 1 (low-price brands)	0.261	0.305	0.044	0.142	912
Tier 2 (mid price brands)	0.726	0.689	−0.037	0.216	912
Tier 3 (high-price brands)	0.013	0.007	−0.007	0.315	912
Package size
Stick	0.073	0.259	0.186	<0.001	904
Pack	0.914	0.721	−0.192	<0.001	904
Carton	0.013	0.020	0.007	0.435	904
Product type
Manufactured cigarettes only	0.408	0.376	−0.032	0.099	2,470
RYO tobacco only	0.338	0.500	0.162	<0.001	2,470
Both	0.254	0.125	−0.130	<0.001	2,470
Independent Variables
Price of manufactured cigarettes per pack, in 2005 baht	33.20 (1.80)	36.22 (2.54)	3.03	<0.001	2,844
Brand-specific price of manufactured cigarettes per pack, in 2005 baht	33.93 (4.90)	36.17 (4.58)	2.25	<0.001	2,470
Price of RYO tobacco per pack- equivalent, in 2005 baht	3.62 (1.46)	3.28 (1.48)	-0.34	<0.001	2,844
Annual household income, in 100,000 baht	1.15 (2.08)	--	--	--	2,844
Age	48.0 (13.6)	--	--	--	2,844
Education
Primary or less	0.764	--	--	--	2,844
Secondary	0.159	--	--	--	2,844
Beyond secondary	0.077	--	--	--	2,844
Urban	0.255	--	--	--	2,844
Exposure to tobacco advertising	0.139	--	--	--	2,840
Exposure to anti-tobacco messages	0.431	--	--	--	2,840
Smoking restrictions at home or work	0.530	--	--	--	2,844

Note: Continuous variables include a standard deviation; categorical variables do not. N represents the number of person-year observations available for each variable. Column (4) presents two-tailed t-tests for a linear combination of coefficients.

Finally, environmental factors, often in the form of anti-tobacco policies, contribute to consumption patterns. These factors include a person’s exposure to tobacco industry advertising, exposure to anti-tobacco media campaigns, and policies restricting smoking at home and at work. Smoking regulations may increase the time costs of smoking, alter smoking norms, and increase awareness of smoking-related harms.

3. RESULTS

Many men in the sample altered their smoking behavior across waves (Table 1; transition probabilities not shown): 12.7% of all smokers quit; 50.1% of all smokers decreased consumption and 19.9% increased consumption by more than one cigarette; 8.3% of cigarette users switched to a cheaper brand and 3.5% switched to a more expensive brand; and 21.7% of cigarette users switched from buying cigarettes by the pack to buying cigarettes by the stick and 3.1% switched from sticks to packs. Moreover, 19.1% of cigarette-only smokers at baseline transitioned to some use of RYO tobacco at wave 2, compared to 9.4% of RYO users who transitioned to some use of manufactured cigarettes, and nearly two-thirds (63.7%) of mixed users at wave 1 used RYO tobacco only at follow-up. Table 1 also indicates that the market price for cigarettes increased 3.0 baht (9.1%) across waves; brand-specific market prices increased 2.3 baht (6.6%) across waves; and RYO tobacco prices remained relatively flat (decline of 0.3 baht) across waves.

Table 2 shows the output from the multivariate estimation of the correlates of quitting (Models 1–2) and daily cigarette consumption (Models 3–6). An increase in price is significantly associated with an increased probability of quitting, according to the community random effects models with and without controls (Models 1–2). A 10-baht price increase—about one-quarter of the price of a pack of the most popular brand—translates into a 1–2% point (roughly 10–20%) increase in the probability of quitting. The elasticity of quitting with respect to tobacco prices appears to fall between 0.2 and 0.4. Prior estimates of the price elasticity of smoking prevalence in low- and middle-income countries cover a fairly uniform distribution from 0 to 1 (IARC, 2011).

Table 2.

Price effects on quitting and cigarette consumption

	Quit (1)	Quit (2)	Cigs/day (3)	Cigs/day (4)	Cigs/day (5)	Cigs/day (6)
Manufactured cigarette price change	0.0089** (0.0037)	0.0147*** (0.0036)
Manufactured cigarette price at baseline	−0.0011 (0.0045)	−0.0045 (0.0045)
RYO tobacco price change	0.0211* (0.0112)	0.0150 (0.0103)
RYO tobacco price at baseline	0.0054 (0.0037)	0.0010 (0.0040)
Manufactured cigarette price			−0.0217 (0.1437)	−0.0402 (0.1429)	0.0572 (0.1642)	0.0389 (0.1671)
RYO tobacco price			−1.0168*** (0.2814)	−1.0232*** (0.2637)	−0.1051 (0.6707)	−0.1422 (0.6793)
Manufactured cigarette price elasticity	0.2133** (0.0886)	0.3527*** (0.0828)	−0.0590 (0.3911)	−0.1108 (0.3928)	0.1551 (0.4442)	0.1040 (0.4469)
RYO tobacco price elasticity	−0.0572* (0.0311)	−0.0408 (0.0281)	−0.2855*** (0.0794)	−0.2917*** (0.0773)	−0.0281 (0.1802)	−0.0374 (0.1801)
Baseline covariates		X		X	X
Community random effects	X	X	X	X
Community fixed effects					X
Person fixed effects						X
Mean of dependent variable	0.1266	0.1266	13.11	13.11	13.11	13.11
Number of observations	1,422	1,422	2,172	2,172	2,172	2,172
Number of random/fixed effects	36	36	36	36	36	1,089

Note: OLS regression coefficients are reported as marginal effects and elasticities. Robust standard errors, in parentheses, are clustered at the market level. Baseline covariates are listed in Table 1. The quit models use cross-sectional data. The consumption models use the full panel data.

^*p<0.10,

^**p<0.05,

^***p<0.01

Thailand’s tax increase appears to have led some smokers to quit smoking, but it did not alter the smoking intensity of continuing smokers (Table 2, Models 3–6). The marginal effect of cigarette prices on consumption is small and of the wrong sign for two of four models. When we restrict the sample to cigarette users at both waves (i.e., those who we would expect to be most responsive to the cigarette tax), the results remain similar (data not shown). Nicotine dependence may impede the ability of long-time smokers to compensate for the price increase by decreasing cigarette intake. We do not have data on the price elasticity of smoking initiation, but the combined price elasticity of quitting and intensity are in line with prior work on the unconditional demand for smoking in developing countries (IARC, 2011).¹⁷

Table 3 provides the main results for the price effect on smokers’ package size last purchased (Models 1–4) and brand choice (Models 5–8). We find that smokers are significantly more likely to purchase cigarettes by the stick as the market price for cigarettes increases. The community random effects models (Models 1 and 2) and community fixed effects model (Model 3) suggest a strong relationship, with a 4.7 to 6.5% point increase in purchasing sticks following a 10-baht increase. The individual fixed effects model yields a marginal price effect of similar magnitude, but the coefficient is no longer significant at the 5% level. The standard error more than doubles between the community and individual fixed effects models, suggesting that the imprecision may result from limited power to detect the effect. The price elasticity estimates range from roughly 1.0 to 1.3, implying that a 1% increase in cigarette prices corresponds to a 1.0–1.3% increase in the probability of purchasing individual sticks.

Table 3.

Price effects on package size and brand choice

	Sticks (1)	Sticks (2)	Sticks (3)	Sticks (4)	Price Tier (5)	Price Tier (6)	Price Tier (7)	Price Tier (8)
Brand-specific cig. price	0.0065*** (0.0023)	0.0049*** (0.0019)	0.0047** (0.0020)	0.0075 (0.0058)	−0.0812*** (0.0033)	−0.0815*** (0.0034)	−0.0815*** (0.0034)	−0.0757*** (0.0050)
Brand-specific price elasticity	1.2844** (0.5056)	0.9896*** (0.3836)	0.9962** (0.4324)	1.5944 (1.2163)	−9.8803*** (0.5415)	−10.0820*** (0.4365)	−10.0557*** (0.4161)	−9.3420*** (0.6193)
Baseline covariates		X	X			X	X
Community random effects	X	X			X	X
Community fixed effects			X				X
Person fixed effects				X				X
Mean of dependent variable	0.1659	0.1659	0.1659	0.1659	0.2829	0.2829	0.2829	0.2829
Number of observations	904	904	904	904	912	912	912	912
Number of random/fixed effects	36	36	36	452	36	36	36	456

Note: Regression coefficients for the price variable are reported as marginal effects and elasticities. Robust standard errors, in parentheses, are clustered at the market level. The sample is restricted to smokers of manufactured cigarettes in both waves. Models 1–4 show the results for the price effects on purchasing cigarettes by the stick, as opposed to by the pack, and Models 5–8 shows the results for selecting a brand from a high-price tier, as opposed to the low-price tier. Baseline covariates are listed in Table 1.

^*p<0.10,

^**p<0.05,

^***p<0.01

We find that smokers are also significantly more likely to purchase a low-priced brand as price increases (Models 5–8). A 10-baht increase in prices leads to roughly a 7–8% point increase in the likelihood of trading down to a cheaper brand, and this relationship holds across all models, including the individual fixed effects model. Brand choice is highly price-elastic, such that a 1% price increase corresponds with a 9–10% increase in the probability of down-trading.

Table 4 shows the effect of tobacco prices on choice of tobacco product using a multinomial logit regression with community random effects. In Model 1, higher cigarette prices are associated with reduced use of cigarettes and greater use of RYO tobacco among the full sample of smokers. Likewise, higher RYO tobacco prices decrease the chances of consuming RYO tobacco and increase the chances of using cigarettes and of using a mixture of RYO tobacco and cigarettes, although none of those estimates are significant. Restricting the sample to cigarette smokers at baseline (Model 2), cigarette prices strongly relate to a reduction in cigarette use and an increase in sole and mixed use of RYO tobacco. A 10-baht increase in cigarette prices reduce cigarette use by 24% points and increase sole use of RYO tobacco by 12% points and mixed use by 12% points. The cross-cigarette price elasticities are particularly large: 9.2 for sole RYO use and 3.7 for mixed use. The positive cross-price elasticities suggest that both mixed use and RYO tobacco are substitute goods for cigarettes.

Table 4.

Price effects on choice of tobacco product type

	(1)			(2)
	Cig only	RYO only	Both	Cig only	RYO only	Both
Manufactured cigarette price	−0.0103 (0.0077)	0.0176** (0.0075)	−0.0073 (0.0099)	−0.0236*** (0.0079)	0.0119** (0.0055)	0.0118 (0.0075)
RYO tobacco price	−0.0148 (0.0328)	0.0054 (0.0253)	0.0093 (0.0124)	−0.0032 (0.0193)	−0.0012 (0.0079)	0.0044 (0.0129)
Manufactured price elasticity	−1.3927*** (0.4167)	2.0829*** (0.3851)	−1.7433*** (0.5709)	−1.1334*** (0.3855)	9.2493** (4.3585)	3.6604 (2.5502)
RYO tob. price elasticity	0.1746** (0.0748)	−0.1515** (0.0738)	−0.0262 (0.1085)	−0.0159 (0.0948)	−0.0802 (0.6087)	0.1242 (0.3808)
Cigarette users at baseline				X	X	X
Baseline covariates	X	X	X	X	X	X
Community random effects	X	X	X	X	X	X
Mean of dependent variable	0.3919	0.4186	0.1895	0.8110	0.0507	0.1383
Number of observations	2,470	2,470	2,470	1,164	1,164	1,164
Number of clusters	36	36	36	36	36	36

Note: The price coefficients from multinomial logit models with community random effects are reported as marginal effects and elasticities. Robust standard errors, in parentheses, are clustered at the market level. Model 1 includes all smokers. Model 2 includes the subset of smokers who used factory-manufactured cigarettes at baseline and thus captures those who switched from cigarettes to RYO tobacco use or mixed use between waves. Baseline covariates are listed in Table 1.

^*p<0.10,

^**p<0.05,

^***p<0.01

Uptake of sole use of RYO tobacco is not related to its own-price effect. This finding may stem from the relatively stable mean price of RYO tobacco over time and the small number of smokers who jumped directly from only buying cigarettes to only buying RYO tobacco. Mixed use typically serves as a conduit during this transition. Thus, it is not surprising that smokers are more likely to choose mixed use when RYO prices increase (RYO price elasticity of 0.12).

4. DISCUSSION

This paper describes the price-minimizing consumption decisions of smokers in Thailand. A substantial increase in Thailand’s cigarette excise tax rate made smokers more likely to quit smoking, exhibiting the potential of tobacco taxation to improve public health. Yet, our analyses also indicate that many Thai smokers muted the health impact of the cigarette price increase by altering their consumption patterns: switching to cheaper brands, purchasing loose cigarettes instead of packs, and smoking cheaper substitutes, namely RYO tobacco. The price increase due to Thailand’s tobacco excise tax increase significantly influenced all of these behaviors.¹⁸

The welfare consequences of smokers’ strategic responses depend on the economic model being applied. The standard rational addiction model (Becker and Murphy, 1988) assumes that smokers fully account for the personal costs of smoking, such that any compensating behavior is consistent with utility maximization. Evidence drawn from more recent models that capture individuals’ behavioral biases—such as present bias derived from time-inconsistent preferences and projection bias derived from underestimating the degree to which preferences will change in the future—indicate that smoking may be welfare-reducing (Gruber and Köszegi, 2001; Levy, 2010), implying that compensating behavior is harmful to consumers who are naïve about future smoking behavior. Given sufficiently large behavioral biases or externalities, the government has a stake in setting public policy that takes into account compensating behavior.

A new finding emerging from this analysis is the strong positive relationship between price and the substitution of packs in favor of individual sticks. More research is needed to understand whether this decision is a detour or a shortcut on the path to cessation. In our sample, 18% of those who last bought sticks at wave 1 had quit by wave 2, compared to 12% of those who last bought packs (data not shown). Thus, at least for some, the decision to quit smoking involves a period during which smokers purchase sticks. On the one hand, stick-buying may facilitate cessation. Khwaja, Silverman, and Sloan (2007) provide evidence that stick-buying can function as a self-control mechanism. They present data that 27% of U.S. smokers in their sample admit to purchasing packs instead of cartons in order to limit consumption. On the other hand, smokers may turn to buying individual sticks after encountering higher prices as a way to delay cessation. In our sample, stick-buying allowed many smokers to continue using the same brand, with three-quarters of those who switched to sticks remaining loyal to the same brand across waves.¹⁹ Consumption patterns related to the purchase of individual sticks are not well researched and are in need of more attention from investigators.

As some observers note, the trend toward higher quit rates in the industrialized world has yet to occur in many developing countries (Abdullah and Husten, 2004). The abundance of very cheap cigarettes, even adjusting for differences in cost of living, may be one promoter of low quit rates (Blecher and Van Walbeek, 2009). Down-trading to cheaper cigarettes is the most price-elastic of the behaviors we examined. The tax structure plays an important role in the desirability of this compensatory behavior. Whereas an ad valorem tax increase makes cheap brands relatively cheaper, an increase in a specific excise tax diminishes the cost advantage of trading down to a cheaper brand. Thailand’s price controls lead its ad valorem tax to function more like a specific tax, in practice; even still, had Thailand increased its specific excise tax rather than its ad valorem rate, we expect that brand-switching would have been less pronounced.

Most RYO tobacco is exempt from Thailand’s excise tax on cigarettes. If the Thai government wants to make further inroads into curbing tobacco use, then one policy direction may involve raising the specific tax for RYO tobacco. Our results indicate that smokers’ choice of tobacco type is highly price-sensitive, and smokers frequently oscillate between RYO tobacco during bad economic times and cigarettes when their luck improves. To counter a similar problem in India where bidis are under-taxed, Sunley (2008) recommends prohibiting the sale of unbranded products, requiring the reporting of sales of processed bidi tobacco to the government, and equalizing the excise tax rate on bidis to that of micro non-filter cigarettes. Similar tactics may be effective in Thailand.

This study had several limitations. Thailand implemented two other tobacco control reforms between the survey waves, the introduction of pictorial warnings on tobacco packages and a ban on point-of-sale advertising, both of which could confound the ability to isolate price effects. The survey questionnaire did not provide a robust control for the impact of warning labels on quitting or switching to RYO tobacco. The best available evidence points toward the new pictorial warnings raising awareness of labels but not demonstrably changing behavior. Although men were more likely to notice the new pictorial labels than the old text labels, the frequency with which respondents noticed the labels did not relate to daily cigarette consumption nor to the decision to purchase cigarettes by the stick versus by the pack and did not alter the magnitude of the price effects in these models.²⁰ The introduction of the warning labels would also not be expected to affect smokers’ choice of price tier. As for the point-of-sale ban, its most likely effect would be to attenuate smoking initiation. Smokers in the sample initiated the habit nearly three decades earlier on average, which makes it less likely that the ban would directly influence such veteran smokers, especially because the ban was selectively followed.²¹ Another concern is that the 22% of respondents lost to follow-up did not attrite randomly. Although significant differences existed in some baseline characteristics between those lost and those interviewed in both waves, the magnitude of these differences was small, and those characteristics are controlled for in our model.

Researchers still have much to learn about the mechanisms underlying strategic responses to sin taxes. One critical question is which types of smokers are most likely to exhibit strategic behavior. Heterogeneous responses would have implications for ongoing debates regarding the potential regressivity and incidence of sin taxes. Future studies of cigarette taxes might also attempt to model simultaneously the decisions of brand choice and tobacco type, in order to tease out how factors differentially affect these two behaviors. Another fruitful approach might use the entire set of available prices. In both regards, the present study was limited by sample size. Longitudinal studies from other settings, in particular those with longer panels, also would help elucidate the dynamics of the trends described here. Although some aspects of compensation are context-specific, such as the type of product substitution—RYO tobacco, in the case of Thailand—the general patterns may hold across contexts. As the evidence base in this area of tobacco research grows, policymakers would do well to design policies that address the nuances of smokers’ consumption decisions.