Abstract
Introduction.
Cigarette excise taxes are a well-established policy lever for reducing tobacco use. However, estimating the effect of taxes on smoking behavior can be confounded by endogeneity concerns such as selection. This study leverages a unique natural experiment –compulsory relocation of U.S. military service members to installations – to estimate the relationship between state cigarette taxes and smoking behavior without concerns about selection into environments.
Methods.
The current study uses data from the Department of Defense’s 2011 Health-Related Behaviors Survey and 2011 state cigarette excise taxes from the CDC STATE System. Logistic and Poisson regression analyses estimate the cross-sectional associations between state cigarette excise taxes and the following smoking behaviors: current cigarette smoking, frequency of smoking, heaviness of consumption, and cigarette cessation among individuals who smoked while at the current installation.
Results.
Higher taxes are associated with lower odds of current cigarette smoking (AOR = 0.94; 95% CI: 0.89 – 0.98), fewer smoking days per month among current cigarette smokers (IRR = 0.98, 95% CI 0.97 – 0.996), and higher likelihood of quitting smoking among individuals who had smoked at their current installation (AOR = 1.14, 95% CI 1.05 – 1.25). Taxes are not associated with the number of cigarettes smoked per day among current smokers.
Conclusions.
Exogenous assignment to installations in states with higher cigarette taxes is associated with lower likelihood of smoking and greater likelihood of quitting. Findings provide novel evidence in support of a causal impact of cigarette taxes on lower smoking levels among adults.
1. Introduction
Increasing the price of cigarettes through excise taxes is among the most well-established and effective policy levers for reducing tobacco use (Chaloupka et al., 2012; Hopkins et al., 2001; International Agency for Research on Cancer, 2011; World Health Organization, 2015). Past reviews and meta-analyses indicate that, in high income countries, a 10% increase in cigarette price reduces overall cigarette consumption by approximately 4% (i.e., average elasticity of around −0.4) (Chaloupka et al., 2019; International Agency for Research on Cancer, 2011). However, substantial cross-study variation has been documented, particularly with respect to estimates of the impact of taxes on adult smoking (e.g., estimated elasticities ranging from −0.5 to +0.10; Chaloupka et al., 2012; Wilson et al., 2012). Such heterogeneity is likely attributable in part to considerable methodological differences across studies (e.g., data sets, time periods, analytic approaches) (Gallet and List, 2003).
Despite a robust literature, most studies examining how taxes influence adult smoking behavior also share a common methodological limitation —endogeneity (DeCicca and Kenkel, 2015)— which raises concerns about validity of estimates and the strength of causal inference. Random assignment of individuals into different tax environments is generally infeasible, in part because individuals typically choose where they live. Consequently, taxes in their “chosen” state may systematically correlate with a range of observed and unobservable factors related to cigarette smoking propensity. For example, state cigarette taxes vary extensively within the U.S. and exposure to tax environments correlates with sociodemographic and economic factors (Golden et al., 2018) and residents’ attitudes toward tobacco control (Golden et al., 2014). These issues may result in biased estimates and reduce the ability to make strong causal inferences about how policies such as taxes influence use behavior (Guindon et al., 2018; Ruhm et al., 2012; Wasserman et al., 1991). Even longitudinal studies can be undermined with respect to causal inference if, for example, changes in attitudes toward smoking influence changes in policies and vice versa (Gilpin et al., 2004; Hamilton et al., 2008).
To address the problem of endogeneity, this study leverages a natural experiment that provides plausibly exogenous variation in tobacco tax policies: the compulsory relocation of military personnel termed “Permanent Changes of Station” (PCS). PCS moves assign personnel to installations based on the military’s needs and, consequently, are plausibly exogenous to the individual’s health-related behaviors. Individuals do not choose their installations and hence cannot influence the policies they face at their new location Although not strictly “random,” the exogeneity of PCS moves to individual characteristics is well-documented (Lleras-Muney, 2010). Importantly, this natural experiment has been used to assess the relationship between environments and health behaviors and outcomes, such as asthma and obesity (Datar and Nicosia, 2018; Lleras-Muney, 2010). To our knowledge, this is the first study to examine how cigarette excise taxes influence adult smoking behavior using plausibly exogenous assignment of personnel to state “tax environments” as a result of compulsory PCS moves.
2. Material and Methods
2.1. Data
This study uses data from the 2011 Department of Defense’s Health-Related Behaviors Survey (HRBS) (Barlas et al., 2013), a survey of non-deployed active duty personnel from the Army, Navy, Marine Corps, Air Force, and Coast Guard that captures information on tobacco use and other behaviors. In the current study, we focus on respondents assigned to installations in the 50 U.S. states and District of Columbia (including a requirement of shore duty for Coast Guard). To help ensure that behaviors correlate with exposure to the tax environment in the state where individuals are stationed, we define sufficient exposure as having lived at the current installation for at least six months; further, we required that individuals assigned to the installation for one year or less have no more than two months away from the installation and those at the installation for more than a year have no more than four months away.
2.2. Measures
2.2.1. Smoking Outcomes
Smoking and tobacco use-related outcomes include current smoking (i.e., those who responded that they had smoked at least 100 cigarettes in their lifetime and reported smoking “some days” or “every day” in the past month) and any current tobacco use (i.e., any past 30-day use of cigarettes, chewing tobacco, snuff, electronic cigarettes, or other nicotine/tobacco products). Among current smokers, we also examine the frequency and quantity of consumption, defined as number of smoking days in the past month and the typical number of cigarettes smoked per day. In addition, we create a derived variable to indicate smoking cessation after arrival at the current installation (yes/no): this variable is defined as not currently smoking and having last smoked a cigarette more than 30 days ago among individuals who endorsed lifetime cigarette smoking and had previously smoked while at their current installation.
2.2.2. State Cigarette Excise Taxes
State-level cigarette taxes in 2011 (i.e., U.S. tax dollars per pack) are taken from the CDC’s State Tobacco Activities Tracking and Evaluation (Centers for Disease Control and Prevention, 2020) and merged with the HRBS data based on installation location.
2.2.3. Covariates
Respondents report a range of sociodemographic factors including: sex, age category (18–20; 21–25; 26–35; 36–45; 46–65), family status, education, children under age 18 in the household, race/ethnicity (White Non-Hispanic, Black Non-Hispanic, Hispanic, Other), military paygrade (i.e., early career enlisted [E1-E4], mid-career enlisted [E5-E9], and warrant and officer [W1-W5, O1-O10]), service branch (Army, Navy, Marine Corps, Air Force, Coast Guard), and time at installation. Participants also reported on severity of past combat exposure (none, low, moderate, high), and non-combat deployment during the past year.
To help isolate effects of taxes from other tobacco policies and socioeconomic factors, we incorporate state-level data on presence of smoke-free air laws in restaurants and bars for 2011 from the CDC’s State Tobacco Activities Tracking and Evaluation(Centers for Disease Control and Prevention, 2020) and the county-level unemployment rate to account for different socioeconomic conditions across counties. To further isolate the effect of taxation from the installation smoking environment, we also construct county-level measures of the installation smoking policy environment as reported by military personnel. We utilize the average responses among peers assigned to the same county for the following items: 1) “In your off-duty hours, how many of your friends smoke cigarettes when you are around them?” (responses: none (0); some (1); most (2)); and 2) “Thinking about your immediate supervisor(s) at the installation where you are currently stationed, how strongly does he/she discourage the use of cigarettes?” (responses: not at all (0), somewhat discourages (1), strongly discourages (2)). All data are merged based on the zip code of the installation.
2.3. Statistical Methods
We estimate both unadjusted and adjusted models to assess the impact of state taxes on smoking outcomes. First, we estimate unadjusted models to examine associations between tax rate and smoking outcomes in logistic and Poisson regression models for dichotomous and count outcomes, respectively. Then, in adjusted models, we control for state smoke-free air laws in restaurants and bars, county unemployment rate, individual-level demographic and service characteristics (age, sex, race/ethnicity, education, children in household, family status, service branch, pay grade, combat exposure, and past-year deployment), and time at current installation. Assuming installation assignment is truly exogenous, unadjusted and adjusted models should yield similar results, although the precision of estimates may differ due to the inclusion of covariates.
We also conducted sensitivity analyses which built upon adjusted models and additionally controlled for the average of military peers’ perceptions of smoking culture among friends and installation supervisor efforts to deter smoking, which may reflect local sentiment toward smoking and/or tobacco control policy climate. We do not report these models as the main findings because local sentiment may be influenced, in part, by selection. Finally, in falsification analyses, we estimate models with height as the outcome, which should be unrelated to cigarette taxes. All analyses are conducted using Stata 15.1 (College Station, Texas).
3. Results
Sample characteristics are shown in Table 1. The full analytic sample (N = 16,276) was predominantly male (66.2%) and non-Hispanic white (71.1%) with a modal age of 26–35 (37.5%). Approximately one-in-five (18.8%) respondents were current smokers, 12.4% were daily smokers, and 26.7% reported any current tobacco use.
Table 1.
Sample characteristics
| Percent/Mean (SD) (Full Sample N = 16,322) |
|
|---|---|
| Sociodemographic characteristics | |
| Sex (Male) | 66.3% |
| Race/ethnicity | |
| Non-Hispanic White | 71.1% |
| Non-Hispanic Black | 12.2% |
| Hispanic | 12.2% |
| Other | 4.5% |
| Age group | |
| 18–20 | 4.6% |
| 21–25 | 22.6% |
| 26–35 | 37.5% |
| 36–45 | 27.9% |
| 46+ | 7.5% |
| Parent of children < age 18 | 47.9% |
| Family status | |
| Married Spouse Not Present | 7.5% |
| Married Spouse Present | 59.0% |
| Education | |
| Some College | 50.8% |
| College Degree or Higher | 32.7% |
| Military service characteristics | |
| Military service branch | |
| Army | 16.8% |
| Navy | 19.3% |
| Marine Corps | 20.4% |
| Air Force | 30.8% |
| Coast Guard | 12.7% |
| Pay grade | |
| Early career (E1-E4) | 29.0% |
| Mid-career (E5-E9) | 45.6% |
| Warrant/Officer | 25.4% |
| Combat exposure (any lifetime combat deployment) | |
| Low | 13.0% |
| Moderate | 20.0% |
| High | 16.4% |
| Past year non-combat deployment (yes) | 9.5% |
| Months at current installation | |
| ≤12 months | 26.9% |
| 13–24 months | 33.6% |
| 25–36 months | 21.4% |
| >36 months | 18.2% |
| Smoking characteristics | |
| Currently smoke cigarettes some days or every day (yes) | 18.8% |
| Daily smoking status (yes) | 12.4% |
| Non-daily smoking status (yes) | 6.4% |
| Smoking days per month among current smokers (Mean(SD)) | 23.7 (9.4) |
| Cigarettes per day among current smokers (Mean(SD)) | 8.9 (7.3) |
| Any current tobacco use (yes) | 25.7% |
| Lifetime smokers (any current or former daily/non-daily smoking while at the current installation) (yes) | 30.6% |
| Percent lifetime smokers who quit smoking for at least 30 days while at current installation | 23.5% |
Table 2 shows results from unadjusted, adjusted, and sensitivity models examining associations between state taxes and smoking outcomes. Fully adjusted analyses show that higher state taxes are associated with lower likelihood of current cigarette smoking, daily smoking, any tobacco use, fewer smoking days per month among current smokers, and higher likelihood of quitting among individuals who smoked while at the installation. For example, with respect to current smoking, adjusted analyses show that a $1.00 increase in state cigarette tax correlates with a roughly 6% reduction in odds of current smoking (AOR = 0.937, 95% CI 0.893 – 0.984). Taxes are not associated with cigarettes smoked per day among current smokers. Consistent with exogenous exposure to state tax environments, we observe similar patterns of findings with respect to magnitude and direction of effects for unadjusted and adjusted analyses, albeit with some differences in precision. As expected, falsification tests show no association between state taxes and height (b = 0.027, 95% CI −0.0222 – 0.0761).
Table 2.
Unadjusted and adjusted associations between cigarette tax rate and smoking outcomes
| Current Smoking | Current Daily Smoking | Any Tobacco Use | Number of smoking days per month among current smokers | Cigarettes per day among current smokers | Quit at base | |
|---|---|---|---|---|---|---|
| Unadjusted analyses | OR (95% CI) | OR (95% CI) | OR (95% CI) | IRR (95% CI) | IRR (95% CI) | OR (95% CI) |
| 0.919 † (0.835 – 1.011) | 0.912 † (0.829 – 1.003) | 0.941 (0.861 – 1.028) | 0.979** (0.964 – 0.994) | 0.985 (0.943 – 1.029) | 1.137** (1.044 – 1.239) | |
| Observations | 16,322 | 16,319 | 16,326 | 3,055 | 3,056 | 4,107 |
| Adjusted analyses | AOR (95% CI) | AOR (95% CI) | AOR (95% CI) | IRR (95% CI) | IRR (95% CI) | AOR (95% CI) |
| 0.937** (0.893 – 0.984) | 0.935** (0.890 – 0.982) | 0.956* (0.917 – 0.997) | 0.983* (0.969 – 0.996) | 0.989 (0.954 – 1.024) | 1.143** (1.046 – 1.249) | |
| Observations | 16,322 | 16,319 | 16,326 | 3,055 | 3,056 | 4,107 |
| Sensitivity analyses (adjusting for installation tobacco climate) | AOR (95% CI) | AOR (95% CI) | AOR (95% CI) | IRR (95% CI) | IRR (95% CI) | AOR (95% CI) |
| 0.941* (0.893 – 0.992) | 0.940* (0.893 – 0.990) | 0.958* (0.918 – 0.999) | 0.985* (0.972 – 0.999) | 0.983 (0.945 – 1.021) | 1.113* (1.011 – 1.227) | |
| Observations | 14,950 | 14,947 | 14,952 | 2,775 | 2,777 | 3,736 |
Note. Each cell represents a separate logistic or Poisson regression model. AOR = adjusted odds ratio for state cigarette tax rate (in U.S. dollars) from logistic regression models. IRR = incident rate ratio for state cigarette tax rate (in U.S. dollars) from Poisson regression models. 95% CI = Lower and upper limits 95% confidence interval. Adjusted models include covariates for presence of smoke-free air laws in restaurants or bars, county unemployment rate, and individual-level sociodemographic and service characteristics (age, sex, race/ethnicity, education, children in household, family status, service branch, pay grade, combat exposure, and past-year deployment) and time at installation; sensitivity models additionally control for installation smoking culture. Bolded values indicate statistically significant effects at p < .05.
p < .10
p < .05
p < .01
4. Discussion
This is among the first studies to exploit plausibly exogenous assignment in policy environments to estimate the impact of taxes on adult smoking. Consistent with a wealth of evidence supporting the utility of tax increases in tobacco control efforts (Chaloupka et al., 2019; Chaloupka et al., 2012; Gallet and List, 2003; International Agency for Research on Cancer, 2011; Wilson et al., 2012; World Health Organization, 2015), assignment to states with higher cigarette excise taxes was associated with lower likelihood of smoking and any tobacco use, fewer smoking days per month, and greater odds of quitting. However, we find no association between higher taxes and cigarettes smoked per day among current smokers. This pattern is consistent with findings from other work (Callinson and Kaestner, 2014), and suggests that higher taxes may help to reduce smoking participation but not necessarily level of consumption among established smokers (e.g., those who continue to smoke following tax increases).
Our findings show a clear association between higher state cigarette taxes and lower likelihood of smoking and greater likelihood of quitting. For example, in adjusted models, PCS assignment to a state with $1.00 higher state excise tax is associated with a roughly 6% lower odds of current smoking and approximately 14% greater odds of quitting smoking among those who smoked while at their current installation. These estimates suggest a slightly more robust impact of taxes on smoking compared to a recent study, which used longitudinal data from the Current Population Survey Tobacco Use Supplements (1995–2007) and a paired difference-in-difference approach and reported small effects of tax increases on adult smoking (e.g., a $1 increase in taxes correlated with a 2% decrease in smoking participation) (Callinson and Kaestner, 2014).
The use of this natural experiment imposes some limitations. First, it requires a focus on military personnel between ages 18 and 65. As such, findings may not generalize to all populations. Analyses are limited to individuals who had been at the installation for at least six months. Thus, findings do not capture near-term effects of tax exposure on behavior. Data are also cross-sectional and do not include information on individuals’ previous installation or history of tax exposure, which precludes examination of “changes” in taxes pre-post PCS moves. The 2011 data also precede more recent changes in the tobacco policy and product landscape. In addition, this study focuses only on state excise taxes and does not include local-level (e.g., city) taxes; in 2011, additional local cigarette taxes were in place in approximately 460 communities (Centers for Disease Control and Prevention, 2012). We were also unable to assess actual price per pack, which can vary widely due to multiple factors such as retailer pricing, promotional activities and discounting practices.
5. Conclusions
This study takes an important step by using a natural experiment to overcome the endogeneity problem that is common in the policy literature. Findings represent a novel contribution to the large evidence base supporting a role for cigarette excise taxes in reducing adult tobacco use.
Highlights.
Endogeneity can bias estimates of effects of cigarette excise taxes on tobacco use.
This study assessed tax effects using a natural experiment to overcome endogeneity.
Higher state taxes correlated with lower odds of smoking and higher odds of quitting.
Findings provide novel support for a role of state taxes in reducing adult smoking.
Acknowledgments
The study was funded by the National Institute on Alcoholism and Alcohol Abuse (R01AA025622; PI: Nicosia).
Role of Funding Source
The funding agency had no role in in study design, data collection, data analysis or interpretation, manuscript preparation, or the decision to submit for publication.
Footnotes
Conflict of Interest Statement
No conflict declared.
Declarations of interest:
none.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
References
- Barlas FM, Higgins WB, Pflieger JC, Diecker K, 2013. 2011 Health related behaviors survey of active duty military personnel. DTIC Document. [Google Scholar]
- Callinson K, Kaestner R, 2014. Do higher tobacco taxes reduce adult smoking? New evidence of the effect of recent cigarette tax increases on adult smoking. Economic Inquiry 52, 155–172. [Google Scholar]
- Centers for Disease Control and Prevention, 2012. State Cigarette Excise Taxes — United States, 2010–2011. MMWR Morbidity and Mortality Weekly Report 61, 201–204. [PubMed] [Google Scholar]
- Centers for Disease Control and Prevention, 2020. State Tobacco Activities Tracking and Evaluation (STATE) System. https://www.cdc.gov/statesystem/index.html.accessed.
- Chaloupka FJ, Powell LM, Warner KE, 2019. The use of excise taxes to reduce tobacco, alcohol, and sugary beverage consumption. Annual Review of Public Health 40, 187–201. [DOI] [PubMed] [Google Scholar]
- Chaloupka FJ, Yurekli A, Fong GT, 2012. Tobacco taxes as a tobacco control strategy. Tobacco Control 12, 172–180. [DOI] [PubMed] [Google Scholar]
- Dall TM, Zhang Y, Chen YJ, 2007. Cost associated with overweight and obesity, high alcohol consumption, and tobacco use within the military health system’s TRICARE Prime enrolled population American Journal of Health Promotion 22, 120–139. [DOI] [PubMed] [Google Scholar]
- Datar A, Nicosia N, 2018. Assessing social contagion in body mass index, overweight, and obesity using a natural experiment. JAMA Pediatrics 172, 239–246. [DOI] [PMC free article] [PubMed] [Google Scholar]
- DeCicca P, Kenkel D, 2015. Synthesizing econometric evidence: the case of demand elasticity estimates. Risk Analysis 35, 1073–1085. [DOI] [PubMed] [Google Scholar]
- Gallet CA, List JA, 2003. Cigarette demand: a meta-analysis of elasticities. Health Economics 12, 821–835. [DOI] [PubMed] [Google Scholar]
- Gilpin EA, Lee L, Pierce JP, 2004. Changes in population attitudes about where smoking should not be allowed: California versus the rest of the USA. Tobacco Control 13, 38–44. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Golden SD, Kong AY, Lee JGL, Ribisl KM, 2018. Disparities in cigarette tax exposure by race, ethnicity, poverty status and sexual orientation, 2006–2014, USA. Preventive Medicine 108, 137–144. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Golden SD, Ribisl KM, Perreira KM, 2014. Economic and political influence on tobacco tax rates: a nationwide analysis of 31 years of state data. American Journal of Public Health 104, 350–357. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Guindon GE, Paraje GR, Chaloupka FJ, 2018. The impact of prices and taxes on the use of tobacco products in Latin America and the Caribbean. American Journal of Public Health 108, S492–S502. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hamilton WL, Biener L, Brennan RT, 2008. Do local tobacco regulations influence perceived smoking norms? Evidence from adult and youth surveys in Massachusetts. Health Education Research 23, 709–722. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Hopkins DP, Briss PA, Ricard CJ, Husten CG, Carande-Kulis VG, Fielding JE, Alao MO, McKenna JW, Sharp DJ, Harris JR, Woollery TA, Harris KW, 2001. Reviews of evidence regarding interventions to reduce tobacco use and exposure to environmental tobacco smoke. American Journal of Preventive Medicine 20, 16–66. [DOI] [PubMed] [Google Scholar]
- International Agency for Research on Cancer, 2011. Effectiveness of Tax and Price Policies for Tobacco Control In: Cancer I.A.f.R.o. (Ed.), IARC Handbooks of Cancer Prevention, Tobacco Control. World Health Organization, Lyon, France. [Google Scholar]
- Lleras-Muney A, 2010. The Needs of the Army: Using Compulsory Relocation in the Military to Estimate the Effect of Air Pollutants on Children’s Health. The Journal of Human Resources 45, 549–590. [Google Scholar]
- Ruhm CJ, Jones AS, McGeary KA, Kerr WC, Terza JV, Greenfield TK, Pandian RS, 2012. What U.S. data should be used to measure the price elasticity of demand for alcohol? Journal of Health Economics 31, 851–862. [DOI] [PMC free article] [PubMed] [Google Scholar]
- Wasserman J, Manning WG, Newhouse JP, Winkler JD, 1991. The effects of excise taxes and regulations on cigarette smoking. Journal of Health Economics 10, 43–64. [DOI] [PubMed] [Google Scholar]
- Wilson LM, Avila Tang E, Chander G, Hutton HE, Odelola OA, Elf JL, Heckman-Stoddard BM, Bass EB, Little EA, Haberl EB, Apelberg BJ, 2012. Impact of tobacco control interventions on smoking initiation, cessation, and prevalence: a systematic review. Journal of Environmental and Public Health 2012, 961724. [DOI] [PMC free article] [PubMed] [Google Scholar]
- World Health Organization and the Secretariat of the WHO Framework Convention on Tobacco Control, 2015. The economic and health benefits of tobacco taxation. World Health Organization; Geneva, Switzerland. Document number: WHO/NMH/PND/15.16. [Google Scholar]