Comparing the Long-Term Health Impacts of Increasing Cigarette Excise Taxes and Raising the Legal Smoking Age to 21 in the United States

Health Policy 80 (2007) 378–391

Limiting youth access to tobacco: Comparing the long-term health impacts of increasing cigarette excise taxes and raising the legal smoking age to 21 in the United States

Sajjad Ahmad a,∗, John Billimek b,1

a Department of Civil, Architectural and Environmental Engineering, University of Miami, 1251 Memorial Drive, Coral Gables, FL 33146-0630, United States b Department of Psychology and Social Behavior, University of California, Irvine, 3340 Social Ecology II, Irvine, CA 92697, United States

Abstract

Although many states in the US have raised cigarette excise taxes in recent years, the size of these increases have been fairly modest (resulting in a 15% increase in the per pack purchase price), and their impact on adult smoking prevalence is likely insufficient to meet Healthy People 2010 objectives. This paper presents the results of a 75-year dynamic simulation model comparing the long-term health benefits to society of various levels of tax increase to a viable alternative: limiting youth access to cigarettes by raising the legal purchase age to 21. If youth smoking initiation is delayed as assumed in the model, increasing the smoking age would have a minimal immediate effect on adult smoking prevalence and population health, but would affect a large drop in youth smoking prevalence from 22% to under 9% for the 15–17-year-old age group in 7 years (by 2010)—better than the result of raising taxes to increase the purchase price of cigarettes by 100%. Reducing youth initiation by enforcing a higher smoking age would reduce adult smoking prevalence in the long-term (75 years in the future) to 13.6% (comparable to a 40% tax-induced price increase), and would produce a cumulative gain of 109 million QALYs (comparable to a 20% price increase). If the political climate continues to favor only moderate cigarette excise tax increases, raising the smoking age should be considered to reduce the health burden of smoking on society. The health benefits of large tax increases, however, would be greater and would accrue faster than raising the minimum legal purchase age for cigarettes. © 2006 Elsevier Ireland Ltd. All rights reserved.

Keywords: Legal smoking age; Tobacco taxes; Youth smoking; System dynamics; Simulation; QALYs; Smoking prevalence; Policy

∗ Corresponding author. Present address: Department of Civil and 1. Introduction Environmental Engineering, University of Nevada, 4505 Maryland Parkway, Las Vegas, NV 89154-4015, United States. Although the United States has seen considerable Tel.: +1 702 895 5456; fax: +1 702 895 3936. E-mail addresses: [email protected] (S. Ahmad), declines in tobacco use in recent years, adult smoking [email protected] (J. Billimek). prevalence at the end of this decade is likely to remain 1 Tel.: +1 949 295 7126; fax: +1 949 824 3002. signiﬁcantly above the target established with Healthy

People 2010 [1,2]. To meet this objective target, adult age for tobacco of 18 years, the proportion of under- prevalence (which has fallen from 25% in 1995 to age smokers who usually buy their own cigarettes in 23% in 2001 [3] must drop to 12% [4]. Considering stores has been cut in half (from 38.7% to 18.8%) from the effects of tobacco use on mortality, productivity, 1995 to 2003. This reduction in reliance on commer- birth outcomes and quality of life, delayed attainment cial sources of cigarettes, however, has been offset in of smoking rate targets carries significant public health part by increased use of social sources over the same consequences. time span. The proportion of teen smokers who usually Among the best-supported interventions to reduce obtain cigarettes by giving someone else money to buy smoking in the population are increased cigarette them has nearly doubled (from 16% in 1995 to 30% in excise taxes [5–9]. Popular because they are believed 2003) and 9% of underage smokers are usually simply to simultaneously discourage smoking initiation and given cigarettes by adults [15]. encourage cessation while increasing state revenues Gaps in youth access restrictions are problematic [5,9], new excise tax increases have been passed in 35 because they undermine the potentially large health states and the District of Columbia since the beginning benefits of reduced youth prevalence. By one estimate, of 2002. Because the smoking behavior of teenagers, the long-term population health benefits of a given who are the most vulnerable to initiate smoking, is par- decrease in youth smoking initiation probability are ticularly sensitive to price increases, the incremental seven times greater than those resulting from compara- annual benefit of raising taxes grows with every year ble improvements in adult cessation probability [16]. for decades [9,10]. Ninety percent of current adult smokers took up the In spite of strong evidence for the effectiveness of habit before their 18th birthday [17], and more than half cigarette taxes to reduce smoking, there may be limits to of those who initiate smoking in their teens continue how high a tax rate will be politically viable. Recent tax to smoke for 16 years or longer [18]. If youth access to increases, although frequent, have been modest in most tobacco can be restricted, it will provide direct health states, resulting in an average per pack price increase benefits to those who will not initiate smoking. It will of only about 15%, and in several cases are only tem- also benefit those for whom initiation will be simply porary [11]. Resistance from tobacco companies and delayed because of the increased probability of cessa- smokers [5] coupled with concerns about the possible tion associated with a later age of onset [19]. emergence of black markets [12] and an unfair bur- Effectiveness of youth access tobacco programs, den on poor smokers who may lack the resources to however, has been debated in literature. Rigotti et al. quit [13] may discourage lawmakers from setting tax [20] reporting on an evaluation of the effectiveness rates high enough to derive maximum benefit to the of enforcing laws that ban tobacco sales to minors as population. a strategy to reduce tobacco use by adolescents con- Keeping in mind the political costs of additional tax clude, “we found no meaningful difference in smok- increases, policymakers may want to consider other ing behavior between communities that implemented interventions to improve the population’s progress enforcement programs and those that did not.” Later, toward healthy people smoking prevalence goals. in response to comments by Moskowitz et al. [21] Already, many states have implemented programs the authors acknowledge [22] “nonetheless, we remain including education and advertising campaigns, clean optimistic that vigorous enforcement of the law is indoor air laws and telephone support hotlines, but a possible and can stop the illegal sale of tobacco to weakness in current efforts to reduce smoking is the children.” relative ease of youth access to tobacco, which per- In a meta-analysis study, Fichtenberg and Glantz sists even with stricter enforcement of the current legal [22] argue that youth access tobacco programs do not smoking age [14]. affect teen smoking prevalence because as fewer mer- Teenagers obtain cigarettes from two primary types chants sell tobacco to minors, teens will use social of sources: commercial sources (direct retail purchase), sources to obtain tobacco. In an editorial, based on and social sources (buying or being given cigarettes results from Fichtenberg and Glantz study, Ling et al. from friends, acquaintances and relatives). With more [23] conclude that it is time to abandon youth access rigorous enforcement of the minimum legal purchase tobacco programs. This resulted in additional discus- 380 S. Ahmad, J. Billimek / Health Policy 80 (2007) 378–391 sion [24] and a rebuttal from authors [25]. In the dis- individuals. Even if many 19- and 20-year olds who cussion, DiFranza raised questions about the method resemble 21-year olds succeed in buying cigarettes used in the Fichtenberg and Glantz study, and noted following the enactment of an MLPA of 21, more indi- that the meta-analysis improperly combined studies of viduals under 18 may be clearly recognized as being different cohort, cross section, and controlled and non- underage and, therefore, may have their access bet- controlled interventions from three different countries. ter controlled. Given the high risk for adult smoking In a letter to editor, Jason et al. [26] review their posi- associated with initiation before age 18 [17,18], delay- tion and admit that it is premature to abandon youth ing access in this way may carry a significant impact access to tobacco programs. on future smoking behavior. If this is true, as pre- In their discussion of flaws in current youth access dicted for tax increases, the benefits of improved youth restrictions, DiFranza and Coleman [14] suggest that a access restriction would grow with each passing year key step to improving the system would be to raise the as more adolescents enter adulthood as non-smokers minimum legal purchase age (MPLA) for tobacco to [9,10]. The tobacco industry, restricted from overtly 21. Ahmad [27,28] has presented a cost-effectiveness targeting adolescents, appears to be focusing on young analysis of raising the legal smoking age to 21 in adults [43]. Raising the MPLA could also curb some California and the US. of these tobacco company efforts with this age group The legal smoking age in most of the States in USA and may help offset the recent rise in initiation rates is 18. Alaska, Alabama and Utah have minimum smok- among young adults. ing age of 19; Illinois [29] and Massachusetts [30] are This paper provides a comparison of the health ben- increasing the age to 19. Several states including Maine efits that would accumulate for the U.S. population [31], California [32], Connecticut [33], New Jersey following the enactment of age 21 as the new MLPA, [34], New York [35], North Dakota [36], Oregon [37], and compares them to the predicted short- and long- South Carolina [38], Texas [39], and Vermont [40] have term impacts of various levels of excise tax increases. considered (through legislative bills in State assembly) First, we provide an overview of the development and increasing the legal smoking age. The measure carries calibration of the dynamic simulation model used to a significant popular support across party lines in opin- estimate the short- and long-term outcomes of the indi- ion polls [41,42]. cated policy changes. Second, we describe the key The greatest benefit of a higher MLPA, given outcomes estimated in the simulation, and discuss our the increased importance of non-retail sources of assumptions of how the various policy interventions cigarettes, might be the reduction of youth access will affect initiation and cessation rates in the popula- through social sources. DiFranza and Coleman [14] tion in terms of price elasticity for tax increases, and cite a previously unpublished survey by Radeki indi- youth initiation rates for raising the smoking age. Next, cating that 90% of adults approached by minors to we present estimates from the model of the cumula- buy cigarettes are under 21. Although minors routinely tive health impacts, i.e., quality-adjusted life years of encounter 18-year olds at school and elsewhere, they each intervention in both short term (2010) and long typically have much less contact with 21-year olds in term (2078, i.e., after 75 years), and finally we discuss their social circles. There is no direct evidence to indi- the results and present some caveats to consider when cate raising the smoking age will impact smoking rates interpreting them. at all, but it is plausible such a policy change might significantly reduce minors’ access to individuals who can legally purchase tobacco. 2. Methods Combined with rigorous vendor compliance checks, raising the smoking age has the potential to further Levy et al. [44–46] have done extensive work on limit teens’ commercial access to cigarettes by reduc- estimating the effects of policies directed at youth ing cashiers’ ambiguity in determining if a customer access, using a simulation modeling approach. We is of legal age to buy tobacco. As it is, teenagers who developed a dynamic simulation model using Vensim appear older than 16 have a significantly greater success [47] to estimate the population health outcomes result- rate in purchase attempts compared to younger-looking ing from raising taxes on cigarettes and raising the S. Ahmad, J. Billimek / Health Policy 80 (2007) 378–391 381 legal smoking age to 21. This model has been widely age and gender group) is multiplied by group-specific use to explore several tobacco control policies and is relapse probability to indicate how many former smok- described in more detail elsewhere [9,27,28,48–50]. ers will become current smokers again. Age- and gender-specific initiation, cessation and relapse proba- 2.1. Model overview bilities were derived from multiple sources [52,58,59]. Cessation probability estimates were adjusted based The simulated population is initialized to reflect on age of smoking initiation to reflect the increased the size and smoking status distribution of every age likelihood of cessation that accompanies a later age of cohort by gender for the 2003 United States popu- smoking onset. lation [15,51–53]. Then, using change probabilities derived from external sources, the simulation uses a 2.2. Model calibration state-transition model with a time step of 1 year to predict births, deaths, aging, net migration and changes The model’s accuracy was improved through an iter- in smoking status over the next 75 years. A period of ative calibration process for which the model is initiated 75 years was selected to ensure that the full cumula- with actual 1995 population values and run to a future tive lifetime impact of each simulated policy scenario time point. The output is then compared to external esti- is captured in the output. mates of (i) total population size (by age and gender) Change probabilities vary based on the characteris- [51], (ii) current adult smoker prevalence [2], and (iii) tics of members of the population for any given year. life expectancy (by age, gender and smoking status) The number of live births for a given year in the simula- [60,61]. Between iterations, some change probabilities tion is computed by multiplying the number of women were adjusted until the model produced results very in each age cohort by its respective age-specific fertility similar to external estimates of total population in year rate [54]. Similarly, the number of deaths is computed 2025, 2050 and 2075 (within 1% for all groups). Model by multiplying the number of people of each combina- estimates also closely matched externally computed tion of age, gender and smoking status by that group’s values of adult smoking prevalence for 2003 (22.3% respective mortality probability [54,55], and taking the from the model compared to 22.5% from external esti- sum across all groups for each year. Net migration rates mate) and life expectancies (for example, in the model, reflect age and gender specific changes due to people a 45-year-old female current smoker would live an moving in an out of the population [54,56]. average of 33.94 additional years—very close to the Participation in smoking behavior is captured external life expectancy estimate of 33.89 years from in three smoking status categories: never smokers actuarial data). (smoked less than 100 cigarettes in lifetime), current smokers (smoked 100 or more cigarettes in lifetime 2.3. Key outcomes and has smoked in past 30 days), and former smokers (smoked 100 or more cigarettes in lifetime, but none Smoking prevalence for adults (18 years and older) in the past 30 days). The initial population is divided is calculated for every year of the simulation run as the into these three smoking status categories according ratio of the number of current smokers aged 18 or older to external prevalence estimates [15,53,57]. In each and the total population of that age group. Because subsequent year, the current number of never smok- no estimates of price elasticity for smoking participa- ers within each age by gender group is multiplied tion were available for youths aged 14 and younger, by age- and gender-specific initiation probabilities to our annual estimates of youth smoking prevalence only determine how many people leave the never smoker include the ratio of the number of smokers aged 15–17 category to join the current smoker category. The num- and the total number of people in that age group. ber of current smokers (by age and gender group) are Health outcomes are quantified both in terms of total multiplied by group-specific cessation probabilities to accumulated life years (the cumulative sum across the give the number of individuals subtracted from the cur- simulation period of the total number of living members rent smoker group and added to the former smoker of the population at the end of each year) and, as recom- group. Similarly, the number of former smokers (by mended by the US Task Force on Cost-Effectiveness in 382 S. Ahmad, J. Billimek / Health Policy 80 (2007) 378–391

Health and Medicine [62], quality-adjusted life years change in price). Consistent with other work [7,63] and (QALYs). building on our previous work [9], we estimated age The QALY combines improvements in length of life group-specific price elasticity of smoking participation and health-related quality of life into a single measure with a weighted ordinary least squares model using a by adjusting each life year that accrues according to standard model of consumption. a quality of life (QOL) score. For example, 10 years The elasticity estimation model takes smoking of life at 85% quality would result in 8.5 QALYs. In behavior and demographic data for over 1 million our model, each year of life represented was adjusted observations from the Centers for Disease Control and by assigning a QOL score based on the individual’s Prevention Behavioral Risk Factor Surveillance Sys- age, gender and smoking status. These QOL scores tems (BRFSS) [53] from all 50 states from 1993 to were summed for the whole population for every year 2000 (except 1993 when Wyoming did not participate) of the model to determine the number of QALYs that and merges it with state-specific cigarette tax and price accumulated that year, and these annual totals were data [64]. A dichotomous dependent variable indicat- summed across the entire simulation period. We used ing whether each individual is a current smoker was QOL scores derived from Quality of Well Being scale then regressed against state-specific cigarette price, data provided by a 1999 personal communication from various socioeconomic and demographic covariates Kaplan. Because we are not comparing present day (including gender, health status, age, race, education, costs to future benefits, we did not apply a discount marital status, and income), and dummy variables for rate to any outcome values. state and year to determine age-specific price elasticity for adults. Because the BRFSS does not include 2.4. Scenarios simulated minors, we used price elasticity estimates for the 15–17 age group from other studies. Harris and Chan [7] and To determine the public health benefit resulting from Tauras and Chaloupka [65] have estimated overall price each of the proposed policy interventions, we first ran elasticity for youth as −0.831 and −0.79, respectively. the simulation assuming no policy change (that is, no Unlike our price elasticity estimates, these estimates change in smoking prevalence across the model run) capture both participation changes and reduced con- to produce results for a status quo base case. Then, sumption. Because our model only addresses partici- we completed additional simulation runs incorporating pation changes, based on evidence from literature, we assumptions about how each proposed policy change estimated that 50% of the adolescent price elasticity would effect initiation, cessation and relapse probabil- was due to reduced prevalence, which resulted in a par- ities across the population, and compared those results ticipation elasticity estimate of −0.4155 from Harris to the status quo case to determine the incremental and Chan [7]. Our estimate is comparable to participa- health benefit of each intervention. tion price elasticity for young adults (−0.35) estimated by Tauras [66] and (−0.416) by Ross and Chaloupka 2.4.1. Raising cigarette taxes [67]. Modeling the impacts of tax increases on smoking These elasticity estimates, summarized in Table 1, behavior required us to estimate price elasticity (that reflect the change in probability of being a current is, the amount of change in behavior predicted per unit smoker that would accompany a given change in

Table 1 Price effects on smoking in the US Dependent variable 15–17 18–23 24–29 30–39 40–65 65 and over Status as a “current smoker” – (–) −0.0320 (0.010) −0.0265 (0.008) −0.0175 (0.005) −0.0171 (0.004) −0.0124 (0.004) Elasticity for “current smoker” −0.4155 −0.3565 −0.2957 −0.1809 −0.1979 −0.3286 status This table presents weighted ordinary least squares estimates from the BRFSS data, 1993–2000. There are 1,000,013 observations. All regressions estimate controls for gender, age, race, education, income, martial status, health status and region effects although not reported. Standard Errors are in parentheses. S. Ahmad, J. Billimek / Health Policy 80 (2007) 378–391 383 cigarette purchase price. For example, if the price elas- legal smoking age to 21. We varied this assumption in ticity for smoking participation were −0.3 for a given sensitivity analysis and estimated outcomes assuming population, raising the purchase price of cigarettes by 2-year and 1-year shifts in youth initiation rates. 50% would result in a 15% (0.3 times 50%) decline Over time, this reduction in youth initiation would in smoking prevalence. The simulation model incorpo- reduce adult smoking prevalence because fewer teens rates these elasticities to predict the impact of increased would enter adulthood as current smokers and delay- cigarette taxes on smoking participation in the popula- ing the age of initiation for those who smoke as adults tion. We ran scenarios assuming tax increases that raise would increase the subsequent probability of cessation cigarette prices from as little as 10% above status quo [19]. to as much as 100% (a doubling in price), and for all levels of tax-induced price increase at 10% intervals in 2.5. Sensitivity analysis between. To evaluate the sensitivity of outcomes to changes in 2.4.2. Raising the smoking age to 21 important parameter values we carried out a sensitivity Although the true impacts of raising the smoking analysis. For cigarette excise tax increase scenario we age are unknown, it is plausible that an increase in varied price elasticity estimates and for increasing the the minimum purchase age for cigarettes coupled with legal smoking age scenario we varied the youth cohort more rigorous enforcement may reduce youth access to that will be influence by the legislation. cigarettes. To estimate the potential effect of increasing the MLPA for cigarettes to 21, we made the assump- 2.5.1. Tobacco taxes tion that its primary impact on smoking behavior would We varied the price elasticity of tobacco excise taxes be to reduce or delay adolescent smoking initiation for sensitivity analysis between −50% and +50%. We [19]. Specifically, we assume that age-specific initia- carried out five simulation by changing elasticity by tion rates would shift by 3 years so that an 18-year old in −50%, −25%, 0%, +25%, and +50%, where −50% this scenario would have the same likelihood of becom- change means reducing the elasticity by 50%, for exam- ing a current smoker as a 15-year old in the status quo ple, the price elasticity for 24–29-year-old smokers will (as depicted in Fig. 1). A 17-year old in this scenario reduce from −0.2957 to (−0.2957 × 0.5) −0.1478. would have the initiation rate of a 14-year old, and so Similarly, a change of +50% will result in increase on. Adults aged 21 or older would maintain their cur- in price elasticity to (−0.2957 × 1.5) −0.4435 for the rent rate for smoking initiation. We acknowledge that same age group. A 0% change means price elasticity there are no empirical data to support this assumption, values reported in Table 1 are used without any change largely because no state in the US has ever increased the (base case).

2.5.2. Age 21 For base case we assumed that age speciﬁc initiation rates for 18-, 19- and 20-year olds will change following the enactment of law raising the legal purchase age to 21. Through sensitivity analysis we explored two more conservative alternatives assuming that legislation will result in: (i) change in initiation rates for 18- and 19-year olds, i.e., initiation rates for 20-year olds will not be affected) and (ii) change in initiation rates for 18-year olds only, i.e., initiation rates for 19- and 20-year olds will not be affected. For a scenario where raising the legal purchase age will not reduce smoking Fig. 1. Smoking initiation rate in the US assuming a 3-year shift after at all, results of status quo run (model run without any increasing the legal smoking age from 18 to 21. policy change) will hold. 384 S. Ahmad, J. Billimek / Health Policy 80 (2007) 378–391

Table 2 75-Year health impacts of selected anti-smoking interventions in the US Intervention Smoking prevalence in 2078 Cumulative life years Cumulative QALYs 15–17-year olds >18-year olds Total accrued Cumulative Total accrued Cumulative (millions) gain (millions) (millions) gain (millions) Status quo 22.1 22.3 28853 – 24905 – Tax increase (%) 10 21.0 18.9 28905 52 24966 61 20 20.1 16.6 28943 90 25012 106 30 19.2 14.8 28971 119 25046 141 40 18.3 13.5 28994 141 25074 169 50 17.3 12.3 29012 159 25096 191 60 16.4 11.4 29026 174 25115 209 70 15.5 10.6 29038 186 25130 225 80 14.6 10.0 29049 196 25144 238 90 13.7 9.4 29057 205 25155 250 100 12.8 8.9 29065 212 25165 260 Age 21 7.5 13.6 28919 67 25014 109

3. Results lion life years (for net gain of 90 million above the status quo) and 25,012 million QALYs (an additional Table 2 summarizes smoking prevalence and cumu- 106 million above status quo) would accumulate. lative health outcomes for the U.S. population at the The final row of the table presents outcomes after end of the 75-year simulation. The first row presents the 75 years the MLPA is raised to 21, and this policy status quo scenario, assuming that there is no change change, in fact, reduces youth access to tobacco and in smoking behavior throughout the simulation period. delays smoking initiation as assumed. In this scenario, Without intervention, the model predicts that smoking smoking prevalence drops from status quo levels to prevalence for 15–17-year olds will be 22.1% in the 7.5% for 15–17-year olds and to 13.6% for individuals year 2078, and adult prevalence (for ages 18 and older) 18 and older. A total of 28,919 million life years (67 will be 22.3%. A total of 28,853 million life years and million above status quo) and 25,014 million QALYs 24,905 QALYs would accumulate for the population (109 million above status quo) would accrue over the across the 75 years. 75-year period. The next 10 rows in the column present results The intermediate-term effects of increased taxes assuming various rates of excise tax increases. The versus raising the smoking age on 15–17-year-olds’ first of these rows lists the outcomes assuming that smoking prevalence are compared in Fig. 2. The hor- taxes increase nationwide to produce an average 10% izontal line marked “Status quo” indicates the esti- price increase for a pack of cigarettes. In this scenario, mated smoking prevalence in the year 2010 for this the estimate smoking prevalence for 15–17-year olds age group assuming no changes in initiation, cessation in the year 2078 falls to 21% and adult prevalence and relapse rates. As a reference point, a line repre- to 18.9%. The population accrues 28,905 million life senting the Healthy People 2010 objective of 16% for years over the course of the simulation period—52 mil- all high school students (not just 15–17-year olds) is lion more than was estimated for the status quo—and also presented. The 10 bars in solid black represent 24,966 million QALYs for a cumulative gain of 61 2010 prevalence estimates assuming tax-induced prices million. Subsequent rows can be interpreted similarly. increases of 10–100%. For example, if taxes are raised For example, if taxes are increased to produce a 20% to the point that the average per pack purchase price of increase in average per pack cigarette prices, preva- cigarettes increases by 10%, the smoking prevalence lence for 15–17-year olds would fall to 20.1%, adult for this age group in 2010 would be approximately prevalence would be 16.6%, and a total of 28,943 mil- 21%. For a 20% price increase, prevalence would be S. Ahmad, J. Billimek / Health Policy 80 (2007) 378–391 385

Fig. 2. Youth (15–17-year olds) smoking prevalence in the US in 2010. just over 20%. The ﬁnal hashed bar represents the tax-induced price increases ranging from 10% (result- 2010 smoking prevalence for 15–17-year olds assuming in a smoking prevalence of approximately 21%) ing the smoking age is raised to 21, but taxes do not to 100% (producing over 16% prevalence). Under the increase—just under 9%. Comparing the results of tax model’s assumptions, raising the smoking age would increases and raising the MLPA, even doubling the produce only a very small decline in adult prevalence to price of cigarettes (the 100% increase scenario) would 21.7% (shown in the hashed bar) by 2010. Adult preva- fall well short of the reduction in 15–17-year-old smok- lence does not reach the Healthy People 2010 target in ing prevalence by 2010 predicted for raising the MLPA any scenario. to 21. Fig. 3 makes a similar comparison for 2010 adult Figs. 4–6 compare the long-term effectiveness of smoking prevalence. Again, reference lines marking setting the MLPA to 21 and raising taxes to various the status quo adult prevalence of just over 22% and levels. Base case estimates and results from sensitiv- the Healthy People 2010 objective of 12% are included. ity analyses of price elasticity and initiation rate shifts The solid bars represent prevalence estimates assuming are also included. In each ﬁgure, horizontal reference

Fig. 3. Adult smoking prevalence in the US in 2010. 386 S. Ahmad, J. Billimek / Health Policy 80 (2007) 378–391

Fig. 4. Youth (15- to 17-year olds) smoking prevalence in the US in 2078. lines indicate the outcome of raising the smoking age scenario that produces an effect equivalent to that of to 21 assuming initiation rate shifts of 3 years (the base raising the smoking age to 21 assuming the indicated case), 2 years and 1 year, but assuming no tax increase. shift in initiation rate. The curves plotted in each ﬁgure represent outcomes For example, the base case plot in Fig. 4 indicates for levels of taxation causing per pack purchase price to that smoking prevalence for 15–17-year olds in the year increase from 0% (the status quo condition) to 100%. In 2078 would range from 22% assuming the status quo addition to the base case plots, the results of sensitivity (0% tax increase) to under 13% assuming a doubling of analyses assuming price elasticities 50% lower, 25% cigarette prices (100% tax-induced price increase). No lower, 25% higher and 50% higher are plotted for each tax increase scenario would produce a youth prevalence ﬁgure. Any point at which a plotted curve intersects reduction equivalent to that predicted for MLPA of 21 with a horizontal reference line indicates a taxation scenario (with no tax increase) in the base case (3-year

Fig. 5. Adult smoking prevalence in the US in 2078. S. Ahmad, J. Billimek / Health Policy 80 (2007) 378–391 387

Fig. 6. Cumulative QALYs gained in the US in 2078. shift) or assuming a 2-year shift in initiation rate. If 4. Discussion initiation rate only shifts by 1 year, however, raising the MLPA to 21 would produce a decline in prevalence Adult cigarette smoking prevalence in US is declin- comparable to a approximately 68% price increase. ing as expected but not as desired [68]. In this work, Fig. 5 shows adult smoking prevalence after 75 we compare two policy options for their potential years. Comparing base case scenarios for both interven- to reduce smoking prevalence and estimate resulting tions it can be noted that a tax-induced price increase health impacts. If youth smoking initiation is delayed of approximately 40% will result in comparable reduc- as assumed in the model, raising the smoking age to tion in adult smoking prevalence offered by MLPA, 21 would have little immediate effect on adult smok- i.e., smoking prevalence will be just over 13% at the ing, but would bring about a rapid reduction in youth end of the 75-year period for both interventions. If, prevalence by reducing initiation. Over time, as more however, the resulting shift in initiation rate follow- adolescents enter adulthood as never smokers, lower ing MLPA is 2 years or 1 year (contrary to 3 years lifetime initiation rates and higher cessation probabil- assumed in base case) the level of tax required to ities will produce an overall decline in adult preva- achieve comparable reduction in smoking prevalence lence equivalent to an aggressive excise tax increase will be approximately 25% or 12%, respectively. Sim- (producing a 40% rise in cigarette price). Across a ilarly, an increase or decrease of 25% in price elasticity 75-year period, the higher smoking age would yield will shift the breakeven taxation point to 32% or 52%, an increased accumulation of QALYs comparable to respectively. what would be observed following a 20% tax-induced Fig. 6 shows a cumulative gain in QALYs ranging price increase—comparable to the average size of from zero (in the status quo) to 260 million (with 100% all state tax increases enacted since the beginning of tax-induced price increase), and approximately 110 2002. million cumulative QALYs saved assuming a higher Although slower to affect youth initiation rates, smoking age—equivalent to the result of increasing moderate to high excise tax increases produce an imme- taxes to raise prices by 20%. The equivalent taxation diate decline in adult prevalence and rapid accumula- levels are approximately 13% and 7% if shift in age tion of population health beneﬁts by impacting all age is assumed to be 2 years or 1 year, respectively. An groups simultaneously. Plus, even a relatively modest increase or decrease of 25% in price elasticity will impact on youth smoking translates into lower adult shift the breakeven taxation point to 17% or 28%, prevalence and adds to the cumulative long-term ben- respectively. eﬁt of the policy. 388 S. Ahmad, J. Billimek / Health Policy 80 (2007) 378–391

Modeling outcomes over a 75-year time span allows educating law enforcement, retailer and the public us to capture the slowly accumulating benefits of a about the policy and more thorough id checking by policy change, but requires that caveats be considered cashiers—all of which carry costs. Given that each of when interpreting the results. For example, all inter- the hundreds of millions of packs sold nationwide costs vention scenarios are compared against a status quo society US$ 8.61 in medical care and lost productivity, condition in which smoking behaviors do not change, however, some or all of the policy’s costs will be offset even though smoking prevalence has declined in recent by the savings and benefits it generates [27,28,69]. years. Although we could have included a modest Additionally, efforts to control youth access to decline in baseline smoking rates, we presumed that the tobacco are typically met with increased “leakage” same forces that would produce such a decline in the of cigarettes into the teen population through social status quo would also be present in the intervention sce- means [70]. Even with a higher purchase age and rig- narios. Rather than introducing one more assumption orous enforcement, minors will still be able to access into the model about rate of decline in smoking preva- cigarettes through social channels, but not easily in a lence for 75 years into the future without improving the quantity large enough to support a regular smoking results, we opted to keep the status quo prevalence rate habit. Because the policy change would reduce teens’ constant. Because the same assumptions are used in all access to legal buyers in their daily routines and limit scenarios, the results presented can, at a minimum, be their success in buying cigarettes from stores, fewer compared to each other. teens would be likely to obtain enough cigarettes to sus- The impacts of raising the smoking age to 21 require tain a high level of consumption. Other sources such as making assumptions in the model. We assume that internet vendors with weak age verification procedures by increasing the age gap between high school stu- would also have to be more tightly regulated to ensure dents and legal buyers, coupled with increased enforce- maximum benefit of the policy change [71]. ment, teen access to cigarettes by commercial and Similarly, tax increases may motivate leakage in the social means will drop, and initiation rates will shift form of smuggling. Black market activity can dilute the accordingly. We varied this assuming during sensi- effects of tax increases by making cheaper cigarettes tivity analysis and also considered more conservative available to motivated smokers. Vendors on Native changes in initiation rates. Estimates of the potential American reservations open additional opportunities to impacts of the policy change, however, could change purchase cigarettes in person or over the internet at low using a different assumption, and certainly there is no prices with little or no tax [72]. Because the proposed direct evidence that youth initiation would be impacted policy change is nationwide, it does not create any new at all. incentive for interstate smuggling, but may result in With respect to the tax increase, numerous estimates increased smuggling from other countries. Although of price elasticity for smoking participation exist in the the exact magnitude of this effect is unknown, research literature. For example, price elasticity for adolescents’ suggests that the impact of smuggling is not large, and overall demand for cigarettes has been estimated any- smuggling may reduce but will not eliminate the ben- where from −0.8 to −1.4 [10]. The overall demand efits of excise tax increases [73,74]. elasticity figure from which we computed adolescent Another concern surrounding cigarette excise taxes participation elasticity was 0.831 [7], which is on the is their potential recessivity [13]. In general, as a pro- low end of this range. If the actual elasticity is, in fact, portion of disposable income, excise taxes represent greater than this, smoking prevalence would be lower, a larger burden to poor people than to more affluent and health benefits would be greater than the values individuals. Because price elasticities for smoking are reported in this paper. Inflation is not included in the higher for people with lower incomes, however, raising model because we assume that even though cigarette excise taxes may actually reduce the relative tax burden prices will increase in absolute terms over time, they for poorer individuals as their total cigarette consump- will not change relative to the price of other goods. tion (and expenditure) drops by a greater amount than Both of the proposed interventions carry with them for more wealthy smokers [75]. Even if this is true in the practical considerations. Raising the MLPA to 21 must aggregate, however, policy-makers should be sensitive be accompanied by strict vendor compliance checks, to the burden high excise taxes may place on individu- S. Ahmad, J. Billimek / Health Policy 80 (2007) 378–391 389 als who do not quit—particularly if a lack of resources [8] Chaloupka F, Wechsler H. Price, tobacco control policies and is an impediment to cessation. smoking among young adults. Journal of Health Economics Both types of anti-smoking interventions discussed 1997;16(3):359–73. [9] Ahmad S. Increasing excise taxes on cigarettes in California: a here have their own advantages and obstacles which dynamic simulation of health and economic impacts. Preventive policymakers must balance in pursuing a health pro- Medicine 2005;41(1):276–83. motion agenda. Estimates from our simulation model [10] Kaplan RM, Ake CF, Emery SL, Navarro AM. Simulated suggest that impacts of raising the smoking age to effect of tobacco tax variation on population health in Cal- 21 may be comparable to the effects predicted for ifornia. American Journal of Public Health 2001;91(2):239– 44. the typical moderate level of tax increases passed [11] Campaign for tobacco free kids. State cigarette tax increases in recent years throughout the U.S. Although large since January 1, 2002. Available at: http://tobaccofreekids.org/ excise tax increases would have the largest and most research/factsheets/pdf/0239.pdf. Accessed August 20, 2004. immediate effects on smoking prevalence and popu- [12] Fleenor P. How excise tax differentials affect interstate smug- lation health, if the political climate does not support gling and cross-border sales of cigarettes in the United States. Washington, DC: Tax Foundation; 1998 [Background Paper No. such a policy, alternatives such as strengthening youth 26.]. access restrictions by raising the MLPA should be [13] Remler DK. Poor smokers, poor quitters, and cigarette tax considered. regressivity. American Journal of Public Health 2004;94(2): 225–9. [14] DiFranza JR, Coleman M. Sources of tobacco for youths in communities with strong enforcement of youth access laws. Acknowledgement Tobacco Control 2001;10:323–8. [15] Youth Risk Behavior Surveillance System (YRBSS), The National Institute of Drug Abuse supported this 1995–2003. Atlanta, GA: National Center for Chronic Disease work through a grant (PHS GrantDA13332) to the Uni- Prevention and Health Promotion; 2003 [database online]. versity of California Irvine Transdisciplinary Tobacco Available at: http://www.cdc.gov/HealthyYouth/yrbs/data/ index.htm. Accessed August 9, 2004. Use Research Center (http://www.tturc.uci.edu/). [16] Tengs TO, Osgood ND, Lin TH. The public health impact of changes in smoking behavior: results from the tobacco policy model. Medical Care 2001;39:1131–41. References [17] U.S. Department of Health and Human Services. Preventing tobacco use among young people: a report of the Surgeon [1] Mendez D, Warner KE. Smoking prevalence in 2010: why the General. Atlanta, GA: U.S. Department of Health and Human healthy people goal is unattainable. American Journal of Public Services, Public Health Service, Centers for Disease Control Health 2000;90(3):401–3. and Prevention, National Center for Chronic Disease Preven- [2] Centers for Disease Control and Prevention. Cigarette smoking tion and Health Promotion, Office on Smoking and Health; among adults—United States, 2002. MMWR 2004;53(20): 1994. 428–31. Available at: http://www.cdc.gov/mmwr/preview/ [18] Pierce JP, Gilpin E. How long will today’s new adolescent mmwrhtml/mm5320a2.htm,. Accessed May 20, 2004. smoker be addicted to cigarettes? American Journal of Public [3] U.S. Centers for Disease Control and Prevention, TobaccoInfor- Health 1996;86(2):253–6. mation and Prevention Source (TIPS). Smoking prevalence [19] Chen J, Millar WJ. Age of smoking initiation: implications for among U.S. adults. Available at: http://www.cdc.gov/tobacco/ quitting. Health Reports 1998;9(4):39–46. research data/adults prev/prevali.htm. Accessed September 9, [20] Rigotti NA, DiFranza JR, Chang Y, Tisdale T, Kemp B, Singer 2004. DE. The effect of enforcing tobacco-sales laws on adolescents’ [4] U.S. Department of Health and Human Services. Healthy people access to tobacco and smoking behavior. New England Journal 2010 (conference ed., 2 vols.). Washington, DC: U.S. Depart- of Medicine 1997;337(15):1044–51. ment of Health and Human Services; 2000. [21] Moskowitz JM, Malvin J, Rigotti NA, Singer DE, DiFranza [5] Levy DT, Chaloupka F, Gitchell J. The effects of tobacco control JR. Smoking in the young. New England Journal of Medicine policies on smoking rates: a tobacco control scorecard. Journal 1998;338:472–3 [correspondence (letter to editor)]. of Public Health Management and Practice 2004;10(4):338– [22] Fichtenberg CM, Glantz SA. Youth access interventions do not 53. affect youth smoking. Pediatrics 2002;109:1088–92. [6] Lewit EM, Coate D. The potential for using excise taxes to [23] Ling PM, Landman A, Glantz SA. It is time to abandon youth reduce smoking. Journal of Health Economics 1982;1:121–45. access tobacco programmes. Tobacco Control 2002;11:3– [7] Harris JE, Chan SW. The continuum-of-addiction: cigarette 6. smoking in relation to price among Americans aged 15–29. [24] DiFranza JR. It is time to abandon bad science. In: Tobacco Health Economics 1999;8:81–6. control; 13 May 2002 [electronic letter to the editor]. 390 S. Ahmad, J. Billimek / Health Policy 80 (2007) 378–391

[25] DiFranza JR, Fichtenberg CM, Glantz SA, Ling P, Land- [42] Preventing Tobacco Addiction Foundation. The case for man A, Glantz SA. Is it time to abandon youth access pro- Tobacco 21. Available at: http://www.tobacco21.org/california/ grammes? Authors’ replies. Tobacco Control 2002;11(3):282– thecase.asp. Accessed August 9, 2004. 4. [43] Gilpin EA, White VM, Pierce JP. What fraction of young adults [26] Jason LA, Pokorny SB, Schoeny ME, Fichtenberg CM, Glantz are at risk for future smoking, and who are they? Nicotine SA. It is premature to abandon youth access to tobacco pro- Tobacco Research 2005;7(5):747–59. grams. Pediatrics 2003;111(4):920–1 [letter to editor]. [44] Levy DT, Cummings KM, Hyland A. A simulation of the effects [27] Ahmad S. The cost-effectiveness of raising the legal smok- of youth initiation policies on overall cigarette use. American ing age in California. Medical Decision Making 2005; Journal of Public Health 2000;90:1311–4. 25(3):330–40. [45] Levy DT. A simulation model of tobacco youth access policies. [28] Ahmad S. Closing the youth access gap: the projected health Journal of Health Politics, Policy and Law 2000;25(6):1023–50. beneﬁts and cost savings of a national policy to raise the [46] Levy DT, Friend K, Holder H, Carmona M. Effect of policies legal smoking age to 21 in the United States. Health Policy directed at youth access to smoking: results from the SimSmoke 2005;75(1):74–84. computer simulation model. Tobacco Control 2001;10:108–16. [29] Illinois Assembly Bill 1034. Available at: http://www.ilga.gov/ [47] Vensim [object-oriented modeling environment]. Release 5.1. legislation/legisnet92/summary/920HB1034.html. Accessed Harvard, MA: Ventana Systems Inc; 2001. November 10, 2005. [48] Tengs T, Ahmad S, Savage J, Moore R, Gage E. The AMA [30] Massachusetts Assembly Bill #1824. Available at: http://www. proposal to mandate nicotine reduction in cigarettes: a simu- mass.gov/legis/184history/h01824.htm. Accessed November lation of the population health impacts. Preventive Medicine 10, 2005. 2005;40(2):170–80. [31] Maine Assembly Bill #LD982. Available at: http://www. [49] Tengs T, Ahmad S, Moore R, Gage E. Federal policy mandat- mainelegislature.org/legis/bills 121st/billtexts/ld098201- ing safer cigarettes: a hypothetical simulation of the anticipated 1.asp. Accessed November 10, 2005. population health gains or losses. Journal of Policy Analysis [32] California Assembly Bill #AB221. Available at: http:// and Management 2004;23(4):857–72. info.sen.ca.gov/cgi-bin/postquery?bill number=ab 221&sess= [50] Ahmad S, Billimek J. Estimating the health impacts of tobacco PREV&house=B&site=sen. Accessed November 10, 2005. harm reduction: a simulation modeling approach. Risk Analysis [33] Connecticut Assembly Bill #769. Available at http://search. 2005;25(4):801–12. cga.state.ct.us/2003/tob/s/2003SB-00769-R00-SB.html. [51] US Census Bureau, Population Division, (NP-D1-A). Annual Accessed November 10, 2005. projections of the resident population by age, sex, race, [34] New Jersey Assembly Bill #A279 (2006–07 session). Available and Hispanic origin: middle series 1999 to 2100; 2001. at http://www.njleg.state.nj.us/bills/bills0001.asp. Accessed http://www.census.gov/population/www/projections/natdet- March 22, 2006. D1A.html. Accessed March 12, 2003. [35] New York Assembly Bill #S05301. Available at: http:// [52] U.S. Department of Health and Human Services. National Cen- assembly.state.ny.us/leg/?bn=S05301&sh=t. Accessed Novem- ter for Health Statistics. Teenage attitudes and practices survey, ber 10, 2005. vol. II. Ann Arbor, MI: Inter-University Consortium for Politi- [36] North Dakota Assembly Bill #1183. Available at http://www. cal and Social Research; 1994 [(ICPSR) 6375]. state.nd.us/lr/assembly/59-2005/bill-text/FATR0200.pdf. Last [53] Behavioral Risk Factor Surveillance System (BRFSS). Atlanta, accessed on November 10, 2005. GA: National Center for Chronic Disease Prevention and [37] Oregon Assembly Bill #2753. Available at http://www. Health Promotion; 1995 [database online]. Available at: http:// leg.state.or.us/03reg/measures/hb2700.dir/hb2753.intro.html. www.cdc.gov/BRFSS/technical infodata/surveydata/1995.htm. Accessed November 10, 2005. Accessed March 28, 2003. [38] South Carolina Assembly Bill #H3404. Available at http:// [54] US Bureau of the Census, Population Division. Component www.scstatehouse.net/cgi-bin/query2003.exe?ﬁrst=DOC& assumptions of the resident population by age, sex, race, and querytext=tobacco&category=Legislation&session=115& Hispanic origin: middle series, 1999 to 2100. Washington, DC: conid=1571057&result pos=0&keyval=1153404&printornot US Bureau of the Census. Publication NP-D5. Available at: =N. Accessed November 10, 2005. http://www.census.gov/population/www/projections/natdet- [39] Texas Assembly Bill #HB 345. Available at http://www. D5.html. Accessed on October 22, 2002. capitol.state.tx.us/cgi-bin/db2www/tlo/billhist/billhist.d2w/ [55] U.S. Department of Health and Human Services, National Cen- report?LEG=78&SESS=R&CHAMBER=H&BILLTYPE= ter for Health Statistics. National mortality followback survey. B&BILLSUFFIX=00342. Accessed November 10, 2005. Ann Arbor, MI: Inter-University Consortium for Political and [40] Vermont Assembly Bill #H0105. Available at http://www.leg. Social Research; 2000. state.vt.us/database/status/summary.cfm?Bill=H%2E0105& [56] US Bureau of the Census. Methodology and assumptions for Session=2006. Accessed November 10, 2005. the population projections of the United States: 1999 to 2100. [41] Merkle D. Too young to smoke? Most in new poll would Washington, DC: US Bureau of the Census; Population Division stub out cigarette sales to teens. ABCNEWS.com. Available Working Paper No. 38. at: http://more.abcnews.go.com/sections/business/dailynews/ [57] U.S. Department of Health and Human Services, National Cen- smokingage poll020613.html. Accessed September 2, 2003. ter for Health Statistics. Teenage attitudes and practices survey, S. Ahmad, J. Billimek / Health Policy 80 (2007) 378–391 391

vol. II. Ann Arbor, MI: Inter-University Consortium for Politi- [67] Ross H, Chaloupka FJ. The effect of cigarette prices on youth cal and Social Research; 1994 [(ICPSR) 6375]. smoking research paper series, no. 7; February 2001. ImpacTeen [58] US Bureau of the Census. Methodology and assumptions for research initiative. Available at: http://www.uic.edu/orgs/ the population projections of the United States: 1999 to 2100. impacteen/generalarea PDFs/pricepaperFebruary2001.pdf. Washington, DC: US Bureau of the Census; Population Division Last accessed November 24, 2004. Working Paper No. 38. [68] Mendez D, Warner KE. Adult cigarette smoking prevalence: [59] U.S. Department of Health and Human Services, National Cen- declining as expected (not as desired). American Journal of ter for Health Statistics. National Health Interview Survey. Public Health 2004;94(2):251–2. Health Promotion and Disease Prevention Supplement. Wash- [69] Centers for Disease Control and Prevention. Sustaining state ington, DC; 1991. Available at: http://www.icpsr.umich.edu. programs for tobacco control: data highlights 2004. Available Accessed March 28, 2003. at: http://www.cdc.gov/tobacco/datahighlights/DataHighlights. [60] American Academy of Actuaries. Final report of the Amer- pdf. Accessed August 9, 2004. ican Academy of Actuaries’ Commissioners Standard Ordi- [70] White MM, Gilpin EA, Emery SL, Pierce JP.Facilitating adoles- nary Task Force. Washington, DC: American Academy of cent smoking: who provides the cigarettes? American Journal Actuaries; 2002 [accessed March 29, 2003]. Available at: of Health Promotion 2005;19(5):355–60. http://www.actuary.org/life/CSO 0702.htm. [71] Ribisl KM, Williams RS, Kim AE. Internet sales of cigarettes [61] Hatton Financial Inc. 2002. Life expectancy table [accessed to minors. JAMA 2003;290(10):1356–9. March 29, 2003]. Available at: http://www.hatton.ca/ [72] Hyland A, Higbee C, Bauer JE, Giovino GA, Cummings KM. lifeexpectancies.htm. Cigarette purchasing behavior when prices are high. Jour- [62] Gold MR, Siegel JE, Russell LB, Weinstein MC. Cost- nal of Public Health Management Practice 2004;10(6):497– effectiveness in health and medicine. New York: Oxford Uni- 500. versity Press; 1996. [73] Farrelly MC, Nimsch CT, James J. State cigarette excise taxes: [63] Lewit EM, Coate D. The potential for using excise taxes to implications for revenue and tax evasion. Research Trian- reduce smoking. Journal of Health Economics 1982;1:121–45. gle International Report. Available online at: http://www.rti. [64] Orzechowski W, Walker R. The tax burden on tobacco, volume org/pubs/8742 Excise Taxes FR 5-03.pdf. Accessed July 20, 35. Arlington, VA: Orzechowski and Walker; 2001. 2004. [65] Tauras JA, Chaloupka FJ. Price, clean indoor air, and cigarette [74] Gruber J, Sen A, Stabile M. Estimating price elasticities when smoking: evidence from longitudinal data for young adults. there is smuggling: the sensitivity of smoking to price in Canada. National Bureau of Economic Research Working Paper No. Journal of Health Economics 2003;22(5):821–42. W6937; 1999. [75] Lindblom E. State cigarette taxes beneﬁt lower-income smok- [66] Tauras JA. Public policy and smoking cessation among young ers and families. Available at: http://tobaccofreekids.org/ adults in the United States. Health Policy 2004;68(3):321–32. research/factsheets/pdf/0147.pdf. Accessed July 20, 2004.