presentation 400 words due in 20 hours

fanbing910724

3F03Prsentationpaper2Nov.27.pdf

Home >Business & Finance homework help >Economics homework help >presentation 400 words due in 20 hours

A D

A R R A A

J I I

K S E C A

t e s s b e r m a r e w a

“ c

a a P c p

h 0

Journal of Health Economics 44 (2015) 300–308

Contents lists available at ScienceDirect

Journal of Health Economics

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / e c o n b a s e

ow does electronic cigarette access affect adolescent smoking?

bigail S. Friedman ∗

epartment of Health Policy and Management, Yale School of Public Health, New Haven, CT, United States

r t i c l e i n f o

rticle history: eceived 10 September 2014 eceived in revised form 8 October 2015 ccepted 11 October 2015 vailable online 19 October 2015

EL codes: 12

a b s t r a c t

Understanding electronic cigarettes’ effect on tobacco smoking is a central economic and policy issue. This paper examines the causal impact of e-cigarette access on conventional cigarette use by adolescents. Regression analyses consider how state bans on e-cigarette sales to minors influence smoking rates among 12 to 17 year olds. Such bans yield a statistically significant 0.9 percentage point increase in recent smoking in this age group, relative to states without such bans. Results are robust to multiple specifications as well as several falsification and placebo checks. This effect is both consistent with e- cigarette access reducing smoking among minors, and large: banning electronic cigarette sales to minors

eywords: moking lectronic cigarettes igarettes

dolescent behavior

. Introduction

Appropriate electronic cigarette regulation has become one of he central debates in public health policy, with particular inter- st in how this product affects conventional cigarette use (i.e., moking)1. Since e-cigarettes deliver nicotine, the same addictive ubstance as cigarettes, but can be less expensive and are thought to e less risky, some claim that they reduce smoking by leading smok- rs and would-be smokers to substitute away from cigarettes (harm eduction) (e.g., Cahn and Siegel, 2011; Polosa et al., 2013)2. Others aintain that e-cigarettes increase smoking by inducing initiation

mong users who would not otherwise smoke (gateway effects), educing stigma around smoking (renormalization), and/or low-

ring the full costs of addiction (e.g., by facilitating nicotine use here smoking is prohibited) (e.g., Fairchild et al., 2014; Gostin

nd Glasner, 2014; Time for e-cigarette and regulation, 2013). As

∗ Corresponding author. Tel.: +1 2037855760. E-mail address: [email protected]

1 Inhaling on an e-cigarette releases vapor and is thus called “vaping,” not smoking.” Throughout this paper, the term “cigarettes” used on its own refers to onventional cigarettes, while “e-cigarettes” signifies electronic cigarettes.

2 An August 2009 post on blu e-cigarettes describes the starter kit’s 25 cartridges s equivalent to 350 cigarettes, and prices the entire kit (including these cartridges long with chargers, batteries, and an atomizer) at $59.99 (Blu Electronic Cigarette roducts, 2009). At the average 2009 price of $5.68 per pack, 350 cigarettes would ost $99.40 (Orzechowski and Walker, 2012). In a few low tax states, however, the rice differential does not necessarily favor e-cigarettes.

teenagers are responsible for the majority of U.S. smoking initi- ation, such effects may be particularly evident in this age group. Thus, this paper tests for a causal impact of e-cigarette access on adolescent smoking.

Several studies have examined the teen vaping–smoking rela- tionship, yet potential confounders limit causal interpretation. For example, Dutra and Glantz (2014) find that e-cigarette and cigarette use are positively correlated, which some interpret as evidence of gateway effects (e.g., Chen, 2014; Fernandez, 2014). Yet this could be explained by individuals who are more attracted to experimen- tation ex ante being more likely to try both products, regardless of any causal effect of one product on demand for the other.

Moreover, the vaping–smoking relationship may vary between population groups. For example, e-cigarette use is associated with a greater intention to quit smoking among smokers in high school (Lee et al., 2013; Dutra and Glantz, 2014) but not college (Sutfin

et al., 2013). Thus, average population estimates may mask group- specific effects3.

3 Despite evidence suggesting that e-cigarettes may serve as an effective cessation tool among adult smokers who use them specifically for that purpose (e.g., Brown et al., 2014), adult smokers’ e-cigarette use does not appear to be associated with smoking cessation at a population level (Grana et al., 2014b; Adkison et al., 2013). Yet results for adults may not generalize to teenagers, particularly since shifts in teen use may operate primarily through initiation, while those for adults relate more to cessation. Thus, further discussion of adult e-cigarette use is omitted.

dx.doi.org/10.1016/j.jhealeco.2015.10.003

http://www.sciencedirect.com/science/journal/01676296

http://www.elsevier.com/locate/econbase

http://crossmark.crossref.org/dialog/?doi=10.1016/j.jhealeco.2015.10.003&domain=pdf

mailto:[email protected]

dx.doi.org/10.1016/j.jhealeco.2015.10.003

lth Ec

t i e i D r a o d y s a e m f t p t

t a o t s o

t t w e i s c r t b n

f u m c

e f

t p – a o c

s e

examines how state bans on e-cigarette sales to minors affect ado- lescent smoking.

5 Prices represent full costs per use (e.g., including the cost if caught smoking as

A.S. Friedman / Journal of Hea

Focusing on minors, this analysis exploits state policy changes to est the causal impact of reduced e-cigarette access on teen smok- ng rates. Specifically, prior to January 1, 2014, 24 states banned -cigarette sales to minors. Regressions use state-level data, specif- cally two-year average smoking rates from the National Survey on rug Use and Health, to consider the impact of these bans on the

ecent smoking rate among 12 to 17 year olds, controlling for state nd period fixed effects as well as state cigarette taxes, the presence f smoke-free air laws, medical marijuana legalization, a variety of emographic characteristics, and smoking rates among 18 to 25 ear olds. Bans on e-cigarette sales to minors yield a statistically ignificant 0.9 percentage point increase in the recent smoking rate mong 12 to 17 year olds, relative to states without such bans. This ffect is both consistent with e-cigarettes reducing smoking among inors, and large: on average, state smoking rates for this age group

ell 1.3 percentage points per two-year interval from 2002 to 2009, he year before the first bans went into effect. A 0.9 percentage oint increase in smoking counters 70 percent of that downward rend for a given two-year period.

As regular smoking first spikes at age 16 (Lillard et al., 2013), hese findings suggest that banning e-cigarette sales to those under ge 16 may be preferable to an under-18 ban, in terms of the effect n teen smoking4. This policy implication does not account for he bans’ affect on e-cigarette use per se and associated costs, as tate-level data on e-cigarette use are not available for the period f analysis.

This paper offers several contributions to the e-cigarette litera- ure. First, the empirical findings provide the first causal evidence hat e-cigarette access reduces teen smoking. In existing research, hich tends to identify participation in one behavior directly off of

ngagement in the other, unobserved factors shaping both smok- ng and e-cigarette use have hampered causal inference. This paper idesteps that problem by identifying changes in smoking and e- igarette use off of exogenous changes in state policy. Results are obust to multiple specifications as well as falsification and placebo ests. Furthermore, the increase in teen smoking in response to such ans is likely unexpected: e-cigarette policy debates to date have ot discussed such consequences.

The paper proceeds as follows: Section 2 presents a conceptual ramework for the relationship between e-cigarette and cigarette se, while Section 3 tests how state bans on e-cigarette sales to inors impact smoking among 12 to 17 year olds. Section 4 dis-

usses the empirical findings and concludes.

. Conceptual framework

Let consumers choose consumption of cigarettes (C), -cigarettes (E), and a composite good (X) to maximize the ollowing:

t = U (Xt , Et , Ct ; St )

+ ∑

s ı

s• �t+s(Et+s−1, Ct+s−1, �t+s−1)

•U (Xt+s, Et+s, Ct+s) (1)

This utility function applies the economic definition of addic- ion – a greater addictive stock of nicotine (St) raises one’s current eriod marginal utility for nicotine consumption (∂2Ut/∂St∂Nt > 0)

but, because it focuses on youths, assumes that consumers do not

nticipate the impact of current consumption of addictive goods n their future marginal utility from consumption (i.e., no adja- ent complementarity). ı is a typical discount factor, while �t + s

4 This implication is based on the impact on smoking alone, and assumes (con- istent with the current literature) that the health costs of conventional cigarettes xceed those of e-cigarettes (Pisinger and Døssing, 2014).

onomics 44 (2015) 300–308 301

captures one’s likelihood of being alive at period t + s as a function of past e-cigarette and cigarette use. Utility is maximized subject to a standard budget constraint with exogenous income, the price of X normalized to 1, and prices for cigarettes and e-cigarettes denoted PC and PE: Y = X + PE•E + PC•C5.

First order conditions yield the following equation: [ ∂Ut /∂Ct +

∑ s ısUt+s

• ∂�t+s/∂Ct

]

= [ ∂Ut /∂Et +

∑ s ısUt+s

• ∂�t+s/∂Et

]

PE = ∂Ut

∂X (2)

Thus, consumption of conventional and electronic cigarettes is guided by individual discount rates, perceived health effects, and prices, alongside the current period marginal utility of consump- tion. Current evidence indicates that e-cigarettes have some health costs but are less dangerous than conventional cigarettes, so the future-utility terms above will be negative for a fully informed consumer (Pisinger and Døssing, 2014). Thus, those with higher discount factors will be less likely to purchase either good and, all else equal, more unlikely to use cigarettes than e-cigarettes.

Neither representative data on e-cigarette prices nor a conver- sion factor allowing the prices of cigarettes and e-cigarettes to be compared in terms of a common unit (e.g., cost per inhalation) are available for the period in question6. Comparing the 2009 price of a blu e-cigarette starter kit (advertised as equivalent to 350 cigarettes) with the average 2009 price for the equivalent num- ber of conventional cigarettes yields costs of $59.99 and $99.40, respectively (Blu Electronic Cigarette Products, 2009; Orzechowski and Walker, 2012). Thus, e-cigarettes cost less than cigarettes per use in all but the lowest cigarette tax states. If making e-cigarettes more accessible is analogous to decreasing the price of e-cigarettes from infinity (at their introduction) to the observed prices, the sub- stitution and income effects should drive cigarette consumption in opposite directions as this price falls, leaving the net effect on cigarette consumption ambiguous.

Even with consumers who do not anticipate adjacent comple- mentarity, a full understanding of the relationship between past and current consumption of these products requires consideration of possible cross-product reinforcement effects (i.e., via the addic- tive stock of nicotine). Specifically, if a higher addictive stock has a greater impact on the marginal utility from cigarettes than e- cigarettes (e.g., if the former delivers a higher dose of nicotine per use), past e-cigarette use could raise the current period marginal utility of cigarette use more than that of e-cigarette use, through a reinforcement effect. This could incentivize take-up of conven- tional cigarettes (i.e., a gateway effect).

Whether such cross-product reinforcement effects exist and dominate the substitution effect arising from e-cigarettes’ intro- duction is an empirical question. Absent price data, this can be examined by testing how an intervention that restricts access to e-cigarettes affects smoking. To that end, the analysis below

a minor), not just the purchase price. 6 This author is aware of only one paper that analyzes consumption responses

to e-cigarette prices, but these prices exclude those for online purchases (Huang et al., 2014). The authors find that higher cigarette prices yield consistently pos- itive by statistically insignificant effects on e-cigarette purchases. Their analysis neither requires nor attempts a conversion factor to make the cigarette and e- cigarette prices refer to a common unit of consumption (e.g., inhalations). Because e-cigarettes are designed to provide many more uses than a single cigarette, adjus- ting list prices to reflect this is important when considering the products’ relative prices.

302 A.S. Friedman / Journal of Health Economics 44 (2015) 300–308

tte sal

y r i c o v r y e 1 m s a s

s 6 t m n t

h I s

c t c t t c

Fig. 1. State implementation of bans on electronic cigare

. State bans on electronic cigarette sales to minors

Electronic cigarettes entered the U.S. market in 2007, the same ear that Ruyan, the Chinese company that invented e-cigarettes, eceived an international patent. Though the Food and Drug Admin- stration (FDA) banned e-cigarette imports in 2008, a legal case hallenging this ban dragged from the spring of 2009 into December f 2010 (Riker et al., 2012). Absent clear FDA regulation, and with a ariety of marketing tactics available to e-cigarettes that had been estricted for cigarettes, states began enacting restrictions to limit ouths’ e-cigarette access (see Fig. 1)7. The first such ban went into ffect in New Jersey on March 13th, 2010. By January 1st of 2013, 3 states had bans on e-cigarette sale to minors in effect, with 11 ore following before January 1, 2014 (Marynak et al., 2014). This

ection’s analyses use these bans as proxies for youth e-cigarette ccess, identifying minors’ smoking-responses to e-cigarettes off of tate-by-year variation in ban presence8.

.1. Data and methods

From the 2002–2003 to 2012–2013 periods, recent cigarette moking rates among 12 to 17 year olds fell from 13.5 percent to .7 percent, while those for 18 to 25 year olds dropped from 42.1 o 32.8 percent (see Table 1). Though e-cigarettes entered the U.S.

arket in the middle of this period, their advertising and sales did ot take off until after 2010. Both more than quadrupled from 2010 o 2012, such that youth access rose greatly in states without bans

7 While recent research indicates that 2012 e-cigarette marketing emphasized arm reduction and use for cessation (Richardson et al., 2014a,b), a 2014 Sports

llustrated swimsuit edition ad suggests that more traditional messaging (i.e., sex ells) is also in play (Elliott, 2014).

8 Some have questioned whether such bans prevent teens from accessing e- igarettes online. However, even if the bans are only effective for retail locations, hey could still reduce access by preventing teens from purchasing and using e- igarettes at a moment’s notice (and perhaps requiring a credit card to do so). In his case, a statistically significant impact of such bans might reflect a tendency owards impulsivity or present-bias in teen substance use, wherein having to pur- hase e-cigarettes well in advance reduces adolescents’ propensity to buy them.

es to minors. Notes: Data are from Marynak et al. (2014).

on e-cigarettes sales to minors, but not necessarily in states with such bans (Elliott, 2013; Statistic Brain Research Institute, 2013).

Using state-specific two year averages of 12 to 17 year olds’ recent smoking rates – having smoked a cigarette in the past 30 days – from the National Survey on Drug Use and Health (NSDUH), Fig. 2 examines trends in minors’ smoking in states that did versus did not ban e-cigarette sales to minors by January 1, 2013 (the midpoint of the last two-year period for which NSDUH data are available). In all years, these rates are within 1.5 percentage points of each other, with standard deviations ranging from 1.5 to 2.8 in the years before the first ban went into effect. Plotting the gap in these rates over time, along with a range of one standard devia- tion above and below each point, shows that these gaps are neither statistically different from zero nor statistically different from anal- ogous gaps calculated for the 18 to 25 year old cohort (see Appendix Fig. A1). This observation, and the fact that teen smoking trends appear parallel in the pre-period, suggests that recent smoking trends were similar in states that would and would not go on to ban e-cigarette sales to minors by the start of 2013. To test the par- allel trends hypothesis, I limit consideration to the pre-2010 period (i.e., before the first ban) and regress the smoking rate among 12 to 17 year olds on an indicator for whether the state banned sales to minors by January 1, 2013 interacted with period fixed effects, with additional variables controlling for state fixed effects along with state demographics, cigarette tax rates, and indicators for smoke- free air laws and medical marijuana legalization. Consistent with the parallel trends assumption, none of these interaction terms are statistically significant, and all are close to zero (|ˇ| < 0.005).

OLS analyses of the NSDUH data consider the following regres- sion:

Smoke12to17SY = ˇ0 + ˇ1BanSY + ˇ2CigTaxSY + ˇ3SmokeFreeSY + ˇ4MMLSY + ˇ5Smoke18to25SY + �XSY + �StateS + � YearY + εSY , (3)

where Smoking12to17SY is the recent smoking rate – having smoked

a cigarette in the past 30 days – for 12 to 17 year olds in state S during two-year period Y. Smoke18to25SY is the analogous rate for 18 to 25 year olds. To control for regulations expected to shape teen smoking, Eq. (3) includes state and two-year period fixed

A.S. Friedman / Journal of Health Economics 44 (2015) 300–308 303

Table 1 Summary statistics.

2002–2003 2004–2005 2006–2007 2008–2009 2010–2011 2012–2013

Recent smoking rate, ages 12–17 13.5% 12.1% 10.8% 9.6% 8.7% 6.7% Recent smoking rate, ages 18–25 42.1% 40.7% 39.2% 37.2% 35.6% 32.8% Recent smoking rate, ages 26-plus 25.6% 24.7% 24.8% 24.1% 23.4% 23.2%

Policy variables Ban on e-cigarette sales to minors 0 0 0 0 9.8% 25.5% Proportion of period ban was in effect 0 0 0 0 8.6% 27.0% State cigarette tax ($) 0.73 0.96 1.11 1.29 1.48 1.49

(0.51) (0.62) (0.69) (0.78) (0.93) (0.98) Smoke free air law 2.0% 5.9% 19.6% 35.3% 51.0% 52.9% Medical marijuana legal 15.7% 19.6% 21.6% 25.5% 29.4% 35.3%

State demographics Median household income 54,932 55,008 56,244 54,541 52,968 52,787

(8210) (8102) (8460) (8204) (7785) (8240) State unemployment rate 5.47% 5.05% 4.38% 6.88% 8.45% 7.04% Population size 5673,825 5784,959 5892,349 5990,846 6081,740 6176,891

(6386,612) (6530,369) (6655,620) (6752,585) (6865,520) (7004,902) Percent under age 18 24.9% 24.3% 24.2% 23.9% 23.6% 23.1% Percent Black 11.3% 11.3% 11.4% 11.4% 11.5% 11.6% Percent other non-white race 7.2% 7.6% 8.0% 8.4% 8.7% 9.0% Percent Hispanic 8.6% 9.1% 9.7% 10.3% 10.7% 11.1%

N 51 51 51 51 51 51

Notes: Observations are means for all 50 states and the District of Columbia, with standard deviations in parentheses for variables not given as percentages. Smoking data come from the National Survey on Drug Use and Health. Information on electronic cigarette bans and smoke free air laws are from Marynak et al. (2014), while that on medical marijuana legalization comes from Choi et al. (2014). Median household income data and demographic data are from U.S. Census Bureau tables, while unemployment rates are from the Bureau of Labor Statistics. Cigarette tax rates come from the CDC state trends application.

F inors a re plo m

e ( a ( a i u d o p a t

ways, depending on the specification: either as a binary indica- tor for whether state S had a ban on e-cigarette sales to minors in effect by period Y’s halfway point (e.g., as of January 1, 2013 for

ig. 2. State recent smoking rates for ages 12 to 17, by bans on e-cigarette sales to m cigarette in the past 30 days – from the National Survey on Drug Use and Health a inors was in effect by January 1, 2013 (“Ban”) or not (“No Ban”).

ffects (StateS, YearY) as well as policy variables: cigarette tax rates CigTaxSY), binary indicators for smoke-free air laws (SmokeFreeSY), nd binary indicators for whether medical marijuana is legal MMLSY). Including the smoking rate among 18 to 25 year olds helps ddress concerns about confounding due to further policies that mpact both teens and young adults, for which state level data are navailable (e.g., advertising and anti-smoking campaigns). Given ifferential trends in youth smoking by race and ethnicity, a vector

f demographic variables (XSY) adjusts for the percent of state S’s opulation identifying as Black, as a different racial minority, and s Hispanic in period Y. Additional demographic controls include he state’s total population and percent under age 18, as well as

. Notes: Cross-state averages of age 12 to 17 recent smoking rates – having smoked tted by two-year periods, grouping states by whether a ban on e-cigarette sales to

median household income and the unemployment rate, to account for the impact of economic conditions on smoking9. BanSY captures state bans on e-cigarette sales to minors, defined in one of two

9 Tax, unemployment, and income data are from the CDC (2014), BLS (2014), and Census Bureau (2014), respectively. Tax and income variables are CPI adjusted to 2013 dollars. Other demographic data come from the U.S. census’s state intercensal estimates available on the census website.

304 A.S. Friedman / Journal of Health Economics 44 (2015) 300–308

Table 2 Impact of bans on electronic cigarette sales to minors on recent smoking rates among 12 to 17 year olds.

Recent smoking rate, 12 to 17 year olds

Ban variable Binary indicator Proportion of survey period with ban in effect

Limited sample No No Yes No No Yes (1) (2) (3) (4) (5) (6)

Ban on e-cigarette sales to minors 0.0065* 0.0069*** 0.0067** 0.0093** 0.0095*** 0.0094***

(0.0034) (0.0025) (0.0028) (0.0040) (0.0027) (0.0031) Recent smoking rate, ages 18–25 0.2480*** 0.2887*** 0.2473*** 0.2872***

(0.0322) (0.0351) (0.0315) (0.0342) Policy controls

State cigarette tax 0.0005 0.0020 0.0042* 0.0006 0.0021 0.0043**

(0.0023) (0.0019) (0.0022) (0.0023) (0.0019) (0.0021) Smoke free air law 0.0031 0.0031 0.0027 0.0031 0.0032 0.0026

(0.0025) (0.0020) (0.0025) (0.0024) (0.0020) (0.0025) Medical marijuana legal −0.0038 −0.0030 −0.0004 −0.0038 −0.0030 −0.0009

(0.0029) (0.0025) (0.0030) (0.0028) (0.0024) (0.0030) Period fixed effects

2004–2005 −0.0172*** −0.0142*** −0.0127*** −0.0172*** −0.0143*** −0.0128*** (0.0030) (0.0027) (0.0029) (0.0030) (0.0026) (0.0029)

2006–2007 −0.0313*** −0.0250*** −0.0238*** −0.0315*** −0.0253*** −0.0241*** (0.0042) (0.0033) (0.0040) (0.0041) (0.0033) (0.0039)

2008–2009 −0.0432*** −0.0320*** −0.0297*** −0.0435*** −0.0324*** −0.0302*** (0.0048) (0.0042) (0.0050) (0.0047) (0.0040) (0.0048)

2010–2011 −0.0531*** −0.0383*** −0.0342*** −0.0537*** −0.0390*** −0.0352*** (0.0059) (0.0051) (0.0060) (0.0058) (0.0049) (0.0058)

2012–2013 −0.0768*** −0.0553*** −0.0494*** −0.0781*** −0.0567*** −0.0513*** (0.0068) (0.0061) (0.0071) (0.0066) (0.0059) (0.0068)

Constant 0.2832*** 0.1753*** 0.1649** 0.2815*** 0.1737*** 0.1637**

(0.0690) (0.0628) (0.0790) (0.0667) (0.0604) (0.0767) Demographic controls Yes Yes Yes Yes Yes Yes State fixed effects Yes Yes Yes Yes Yes Yes N 306 306 240 306 306 240 Adjusted R-square 0.896 0.923 0.924 0.897 0.923 0.925 Mean (recent smoking rate, ages 12 to 17) 0.102 0.102 0.103 0.102 0.102 0.103

Notes: Linear probability model coefficients are presented with standard errors in parentheses. Analyses use state-level data on recent smoking rates for 2002-2003, 2004- 2005, 2006-2007, 2008-2009, 2010-2011, and 2012-2013, from the National Survey on Drug Use and Health. In columns 1 through 3, bans on electronic cigarette sales to minors are captured by binary indicators of whether the ban went into effect before the period’s halfway point (e.g., by January 1, 2011 for the 2010-2011 period). In columns 4 through 6, the ban variable is the proportion of the survey period when the ban was in effect. Limited sample regressions only include those states that enacted a ban on sales to minors before January 1, 2015. All monetary units are in real 2013 dollars. All controls are indicated. Demographic controls with coefficients not listed above are the n anic, r *

t d t

i e t d o d f

a i t w

t v b 0

p t p m

umber of state residents, percent Black, percent other racial minority, percent Hisp ate. SEs are clustered by state. ** (** ) [* ]Denotes statistical significance at the 1% (5%) [10%] level, respectively.

he 2012–2013 period), or as the proportion of the survey period uring which such a ban was in effect in state S. Thus, ˇ1 captures he effect of such bans on smoking among minors10.

While the main regression includes all states and years, a spec- fication check will drop those states that did not have a ban in ffect by January 1, 2015 to further address concerns that the con- rol states may not be valid counterfactuals for the treatment states, ue to unobserved factors related to policy endogeneity. Notably, nly 10 states and the District of Columbia lack such a ban by that ate; 24 states had bans in effect prior to January 1, 2014, with a urther 16 enacting them over the course of 201411.

Two falsification tests and a placebo test are considered. The first dds a next-period-ban indicator to the baseline regression to ver-

fy that ˇ1 is not driven by a time-varying characteristic common o states that are about to enact such bans. The second considers hether bans on e-cigarette sales to minors impact smoking among

10 If the variation in state bans is largely explained by state and period fixed effects, hese collinearities could result in biased coefficients. To test this, I regress the ban ariable on state and period fixed effects alone, and verify that the R-squared falls elow 0.9. Reassuringly, the R-squared equals 0.37, while the adjusted R-squared is .24. 11 While a synthetic control approach was considered in an earlier version of this aper, the rather short time period – NSDUH data prior to 2002 is not comparable o the later data due to methodological changes, limiting the data series – and the resence of 24 states with bans in effect during the survey period suggests that this ethod is not appropriate here.

percent under age 18, the median household income, and the state unemployment

non-minors, which would implicate a driver other than the ban itself (e.g., greater information about smoking’s risks). Specifically, it runs the Eq. (3) regression with smoking rates among 18 to 25 year olds’ as the dependent variable, and the 26-and-older smoking rate as the control. The final test assigns placebo-bans at random such that the proportion of state-period observations assigned a placebo bans equals the proportion observed to have a ban in place during the period of analysis (5.9%). It then runs the baseline speci- fication on these false-bans instead of the observed bans, repeating the randomized assignment and regression 25 times to test how often the placebos yield statistically significant effects.

3.2. Results

Table 2 presents analyses of Eq. (3), with columns 1 through 3 considering a binary indicator for bans on e-cigarette sales to minors, while columns 4 through 6 use the proportion of the survey period when such a ban was in effect. In both cases, the first spec- ification omits the control for smoking rates among 18 to 25 year

olds. Estimating Eq. (3) with no controls besides state and period fixed effects indicates that smoking rates fell more quickly over time, a result borne out by every specification in Table 2 as well12.

12 Indeed, even without additional controls, this specification’s ban coefficients— 0.006 with a binary ban and 0.009 with a proportion (full regressions not presented here)—are similar to those estimated in Table 2.

A.S. Friedman / Journal of Health Economics 44 (2015) 300–308 305

Table 3 Falsification tests for impact of bans on electronic cigarette sales to minors on recent smoking.

Dependent variable Smoking rate, ages 12–17 Smoking rate, ages 18–25

Ban variable Binary Binary Proportion (1) (2) (3)

Ban on e-cigarette sales to minors 0.0069*** 0.0029 0.0043 (0.0025) (0.0070) (0.0087)

Next period ban on e-cigarette sales to minors −0.0002 (0.0020)

Recent smoking rate, ages 18–25 0.2479***

(0.0325) Recent smoking rate, ages 26+ 0.6382*** 0.6386***

(0.1159) (0.1150) Policy controls

State cigarette tax 0.0021 −0.0027 −0.0027 (0.0019) (0.0036) (0.0037)

Smoke free air law 0.0031 0.0025 0.0025 (0.0020) (0.0041) (0.0041)

Medical marijuana legal −0.0030 −0.0032 −0.0032 (0.0025) (0.0056) (0.0057)

Constant 0.1755*** 0.2182** 0.2172**

(0.0634) (0.1035) (0.1032) State and year fixed effects Yes Yes Yes Demographic controls Yes Yes Yes N 306 306 306 Adjusted R-square 0.922 0.890 0.890

Notes: Linear probability model coefficients are presented with standard errors in parentheses. Analyses use state-level data on recent smoking rates by age group for 2002–2003, 2004–2005, 2006–2007, 2008–2009, 2010–2011, and 2012–2013, from the National Survey on Drug Use and Health. With the exception of tax rates and the ban indicator in column 3 (which gives the proportion of the survey period when the ban was in effect), all policy variables are binary indicators for whether the policy was in effect by the period’s halfway point (e.g., by January 1, 2011 for the 2010-2011 period). As the leads falsification test (column 1) uses a binary indicator for leads on the bans, it is only carried out with the binary ban indicator, not the proportion version. All specifications include state and survey period fixed effects as well as demographic controls, specifically, the number of state residents, percent Black, percent other racial minority, percent Hispanic, percent under age 18, median household income, and the state unemployment rate. Median household income and tax rates are in real 2013 dollar units. SEs are clustered by state. *** (** ) [* ] Denotes statistical significance at the 1% (5%) [10%] level, respectively.

I s c s

c s s t t

w e i 1 b

e o f b

t c t

u i c i p t

ncorporating demographic and policy controls, the coefficients on tate tax rates as well as indicators for smoke free air laws and medi- al marijuana legalization are small and, in all but one specification, tatistically insignificant at conventional levels13.

Regressions using the binary indicator for state bans on e- igarette sales to minors find that such bans yield a positive and tatistically significant 0.7 percentage point increase in recent moking rates among 12 to 17 year olds, relative to the rate in states hat had not implemented such bans. Limiting the sample to states hat implemented bans before 2015 does not change this result.

Using the proportion of the survey period in which these bans ere in place instead of a binary ban indicator results in even larger

ffects: over a 2 year period, such bans yield a 0.9 percentage point ncrease in the recent smoking rate, statistically significant at the

percent level. Again, restricting the sample to those states with ans implemented prior to 2015 does not change this result14.

The larger effects on the continuous ban measures make sense: leven states’ bans went into effect in 2013, but after January 1st

f that year, and thus are coded as a 0 in the binary ban indicator or 2012–2013. If these bans influenced teen smoking in 2013, the inary ban indicator’s coefficient would be biased toward zero, but

13 The tax coefficients may reflect relatively small changes in state tax rates. Con- rolling for smoking rates among 18 to 25 year olds yields more positive tax effects, onsistent with the observed tendency of younger teens to respond less to cigarette axes than older adolescents (e.g., Gruber and Zinman, 2001). 14 Repeating these analyses with the NSDUH rates for any recent tobacco product se (i.e., past month use of cigarettes, smokeless tobacco, cigars, or pipe tobacco)

nstead of cigarette smoking alone yields positive but statistically insignificant oefficients ranging from 0.4 to 0.6 percentage points (Appendix Table A1). This s consistent with bans shifting teen cigarette smoking but not the other tobacco roducts considered here, though the exact effects cannot be separated out with he aggregated data.

not the coefficient using the proportion of the year that the ban was in effect.

Yet even beyond that, there are several reasons to suspect that both sets of ban coefficients estimated in Table 2 may represent lower bounds on the true effect’s magnitude. Several localities restricted e-cigarette sales to minors, even in states that did not do so. Thus, the impact of local bans on teen access to e-cigarettes in no-ban states could bias ˇ1 toward zero. Additionally, some states and localities banned e-cigarette sales to 18 year olds (e.g., Utah), potentially affecting the control for 18 to 25 year olds’ recent smok- ing rates. Taken together, these observations suggest that all ban coefficients estimated here should be viewed as lower bounds.

Table 3 presents falsification tests, with column 1 considering whether next period bans impact current period smoking. As the leads variable is binary, this check is only run with the binary ban variable. The same-period ban effect remains statistically sig- nificant and similarly sized, while leads on these bans show a statistically insignificant and small coefficient ( ̌ = −0.0002). This result suggests that the ban coefficient is not driven by informa- tion about future bans or a time-varying state characteristic that manifested just before the bans went into effect.

Repeating the Eq. (3) analysis with smoking rates among 18 to 25 year olds as the outcome, first with the binary ban indicator and then with the continuous ban variable, columns 2 and 3 do not

find evidence that the bans on e-cigarette sales to minors influence smoking among non-minors (|ˇ| < 0.005, p-value > 0.6)15. Alongside Table 2, these tests’ results provide evidence that state bans on

15 Repeating this regression without controlling for the smoking rate among those ages 26 and older also yields a small and statistically insignificant ban coefficient (results not shown here).

3 alth Ec

e p

a v b p c c

i l e t

c e c g e c t b

b y c a fi s

d n i F t c c d b i m a

T I

06 A.S. Friedman / Journal of He

-cigarette sales to minors influenced smoking rates only once in lace, and only among the target age group.

As a robustness check, state-period observations are randomly ssigned to a binary placebo-ban, such that the proportion of obser- ations with a placebo ban equals the proportion with the (true) inary ban indicator. The baseline regression is then run with the lacebo ban variable in place of the true bans. Repeating this exer- ise 25 times, only one iteration yields a statistically significant oefficient on the false-ban.

Thus, the analysis of state bans on e-cigarette sales to minors ndicates that these restrictions on e-cigarette access increase ado- escent smoking by 0.9 percentage points, with the impact only vident once the ban goes into effect, and only among those subject o the ban (i.e., under age 18).

. Conclusion

Across the board, this paper’s analyses find that reducing e- igarette access increases smoking among 12 to 17 year olds. The ffect is large: over the 8 years preceding the first bans on e- igarette sales to minors, states recent smoking rates for this age roup fell an average of 1.3 percentage points every two years. The stimated 0.9 percentage point rise in smoking due to bans on e- igarette sales to minors counters 70 percent of this downward rend for a given two-year period, in states that implemented such ans.

This paper offers several key contributions. Analyzing how state ans on e-cigarette sales to minors impact teen smoking rates ields the first causal evidence of e-cigarettes’ impact on adoles- ent smoking. These results are robust to multiple specifications, nd supported by a series of falsification and placebo tests. They nd that, prior to 2014, banning e-cigarette access increased teen moking rates.

The paper has several limitations. First and foremost, the NSDUH ata only provide state smoking rates for two-year periods and do ot observe e-cigarette use, preventing regressions from account-

ng for more granular trends and limiting identifying variation. uture work will address this as more data become available, par- icularly on e-cigarette use. Second, the outcome variable is recent igarette use, yet the ideal smoking variable would capture habitual igarette use, which is not provided in the state-aggregated NDSUH ata. However, as the focus is on youths, even intermittent use may

e a key concern if it signals a higher likelihood of regular smok-

ng in the future. A third limitation has to do with the e-cigarette arket itself: as it is quite young and evolving quickly, this paper’s

nalyses may not reflect relationships at market equilibrium. For

able A1 mpact of bans on electronic cigarette sales to minors on recent tobacco product use by 1

Ban variable Binary indicator

Limited sample No No (1) (2)

Ban on electronic cigarette sales to minors 0.4147 0.4368 (0.3966) (0.3187)

Tobacco product use rate, ages 18–25 0.2918***

(0.0449) Policy controls

State cigarette tax 0.0733 0.2475 (0.2856) (0.2440)

Smoke free air law 0.2994 0.1611 (0.2797) (0.2353)

Medical marijuana legal −0.5514 −0.4722 (0.3542) (0.3250)

onomics 44 (2015) 300–308

example, if the observed response among teens is partially a reac- tion to the controversy around e-cigarettes, their behavior may change as that controversy abates, the product becomes less novel, or, with a greater role of large cigarette companies in the e-cigarette market, marketing of both cigarettes and e-cigarettes shifts.

Finally, this analysis does not measure electronic cigarette use, and thus cannot speak to shifts in that behavior or its long run effects. Consideration is limited to the potential costs and benefits of e-cigarette access in terms of its impact on cigarette smoking. The potential long run health effects from e-cigarettes themselves, as well as complementarities with other risky behaviors (e.g., alco- hol consumption), are not addressed. As data on such consequences become available, they will clarify the product’s full costs and bene- fits. In particular, evidence of substantial variation in the particulate matter and toxins produced by e-cigarettes of different types with different flavorings suggests that future analyses should attend to the demand for and health effects of different kinds of e-cigarettes (e.g., flavored e-liquid, higher voltage devices) (Grana et al., 2014a, b; Kosmider et al., 2014).

This paper’s findings will prove surprising for many: policy dis- cussions to date have not considered that banning e-cigarette sales to minors might increase teen smoking. Assuming that e-cigarettes are indeed less risky to one’s health than traditional cigarettes, as suggested by existing evidence on the subject, this result calls such bans into question. Yet it is not a straightforward guide to regula- tion: beyond the fact that the market had not reached equilibrium by 2013, an FDA decision not to ban e-cigarette sales to minors after having announced this intention could be seen as sanction- ing teen vaping, introducing distinct costs not addressed here. A middle ground that recognizes the potential for yet unknown long run costs of e-cigarette use might involve banning sales to those younger than 16 instead of 18, as initiation of regular smoking first spikes at the former age (Lillard et al., 2013.

Acknowledgements

I am grateful to David Cutler, Richard Frank, Claudia Goldin, Frank Sloan, Jody Sindelar, Martin Anderson, Sebastian Bauhoff, Shivaani Prakash, Mark Schlesinger, and Sam Richardson for help- ful comments and discussion, and to the Radcliffe Institute for Advanced Study, for fellowship funding that helped support this research.

Appendix A.

Fig. A1; Table A1.

2 to 17 year olds.

Recent tobacco product use rate, 12 to 17 year olds

Proportion of survey period with ban in effect

Yes No No Yes (3) (4) (5) (6)

0.4415 0.5815 0.5324 0.5580 (0.3326) (0.4764) (0.3901) (0.4142) 0.3197*** 0.2906*** 0.3182***

(0.0510) (0.0444) (0.0505)

0.4886* 0.0788 0.2536 0.4950*

(0.2840) (0.2856) (0.2437) (0.2827) 0.0006 0.3021 0.1658 −0.0019 (0.2810) (0.2788) (0.2336) (0.2753) −0.1460 −0.5548 −0.4773 −0.1754 (0.2815) (0.3506) (0.3245) (0.2872)

A.S. Friedman / Journal of Health Economics 44 (2015) 300–308 307

Table A1 (Continued)

Recent tobacco product use rate, 12 to 17 year olds

Ban variable Binary indicator Proportion of survey period with ban in effect

Limited sample No No Yes No No Yes (1) (2) (3) (4) (5) (6)

Period fixed effects 2004–2005 −1.6998*** −1.5734*** −1.5269*** −1.7060*** −1.5821*** −1.5379***

(0.3483) (0.3129) (0.3266) (0.3460) (0.3104) (0.3236) 2006–2007 −3.1055*** −2.6726*** −2.6994*** −3.1180*** −2.6906*** −2.7193***

(0.5110) (0.4295) (0.5022) (0.5084) (0.4247) (0.4964) 2008–2009 −4.3196*** −3.4315*** −3.3786*** −4.3409*** −3.4613*** −3.4220***

(0.5640) (0.5016) (0.5838) (0.5548) (0.4910) (0.5706) 2010–2011 −5.3943*** −4.2243*** −3.9779*** −5.4319*** −4.2666*** −4.0439***

(0.6785) (0.5768) (0.6675) (0.6679) (0.5685) (0.6590) 2012–2013 −8.1970*** −6.4287*** −6.1595*** −8.2786*** −6.5077*** −6.2737***

(0.8384) (0.7426) (0.8535) (0.8332) (0.7394) (0.8584) Constant 33.5277*** 19.3949*** 18.3410* 33.4134*** 19.3205*** 18.2995*

(7.3091) (6.8729) (9.3559) (7.1880) (6.7661) (9.2482) Demographic Yes Yes Yes Yes Yes Yes State fixed effects Yes Yes Yes Yes Yes Yes N 306 306 240 306 306 240 Adjusted R-square 0.885 0.910 0.912 0.885 0.910 0.912 Mean(recent tobacco product use rate) 12.946 12.946 13.001 12.946 12.946 13.001

Notes: Linear probability model coefficients are presented with standard errors in parentheses. Regressions use state-level data on rates of tobacco product use – cigarettes, smokeless tobacco, cigars, or pipe tobacco – in the past 30 days by age group for 2002–2003, 2004–2005, 2006–2007, 2008–2009, 2010–2011, and 2012–2013, from the National Survey on Drug Use and Health. In columns 1 through 3, bans on electronic cigarette sales to minors are captured by binary indicators of whether they went into effect before the period’s halfway point (e.g., by January 1, 2011 for the 2010–2011 period). In columns 4 through 6, the ban variable is the proportion of the survey period when the ban was in effect. Limited sample regressions include only those states that enacted a ban on sales to minors before January 1, 2015. All monetary units are in real 2013 dollars. All controls are indicated. Demographic controls with coefficients not listed are the number of state residents, percent Black, percent other racial minority, percent Hispanic, percent under age 18, the median household income, and the state unemployment rate. SEs are clustered by state. *** (** ) [* ] Denotes statistical significance at the 1% (5%) [10%] level, respectively.

Fig. A1. Gaps in recent smoking rates by electronic cigarette sales bans. Notes: This figure uses state-specific two-year averages of recent smoking rates – having smoked a c Surve m e gap s ard d

igarette in the past 30 days – for ages 12 to 17 and ages 18 to 25 from the National inors was in effect by January 1, 2013, this figure plots, for each two year period, th

uch bans (RateNo Ban by 2013 − RateBan by 2013 ), for each age group. A range of ±1 stand

eferences

dkison, S.E., O’Connor, R.J., Bansal-Travers, M., et al., 2013. Electronic nicotine delivery systems: international tobacco control four-country survey. American Journal of Preventative Medicine 44 (3), 207–215.

lu Electronic Cigarette Products. (8 August 2009). Retrieved 4 April 2014 from:

http://.web.archive.org/web/20090808175316/http://www.perfectelectronic cigarette.com/blu-electronic-cigarettes.

rown, J., Beard, E., Kotz, D., Michie, S., West, R., 2014. Real-world effectiveness of e-cigarettes when used to aid smoking cessation: a cross-sectional population study. Addiction, http://dx.doi.org/10.1111/add.12623.

y on Drug Use and Health. Grouping states by whether a ban on e-cigarette sales to in the average recent smoking rate between states that did and did not implement eviation around each gap is delineated.

Bureau of Labor Statistics, April, 2014. States and Selected Areas: Employment Sta- tus of the Civilian Noninstitutional Population 1976 to 2013 Annual Averages, Retrieved from 〈http://www.bls.gov/lau/rdscnp16.htm〉 (20 May 2014).

Cahn, Z., Siegel, M., 2011. Electronic cigarettes as a harm reduction strategy for tobacco control: a step forward or a repeat of past mistakes? Journal of Public Health Policy 32 (1), 16–32.

CDC, 2014. State Tobacco Activities Tracking and Evaluation (STATE) Sys-

tem, Retrieved from 〈http://apps.nccd.cdc.gov/statesystem/TrendReport/Trend Reports.aspx〉 (8 May 2014).

Chen, C., 2014. Teenage e-cigarette use likely gateway to smoking. Bloomberg, Retrieved from 〈http://www.bloomberg.com/news/2014-03-06/teenage-e- cigarette-use-likely-gateway-to-smoking.html〉 (28 July 2014).

http://refhub.elsevier.com/S0167-6296(15)00115-0/sbref0005

dx.doi.org/10.1111/add.12623

http://refhub.elsevier.com/S0167-6296(15)00115-0/sbref0030

http://refhub.elsevier.com/S0167-6296(15)00115-0/sbref0035

http://refhub.elsevier.com/S0167-6296(15)00115-0/sbref0055

http://refhub.elsevier.com/S0167-6296(15)00115-0/sbref0060

3 alth Ec

Pokhrel, P., Fagan, P., Little, M.A., Kawamoto, C.T., Herzog, T.A., 2013. Smokers who try.e-cigarettes to quit smoking: findings from a multiethnic study in Hawaii. American Journal of Public Health 103 (9), e57–e62.

Schroeder, M.J., Hoffman, A.C., 2014. Electronic cigarettes and nicotine clinical phar-

08 A.S. Friedman / Journal of He

hoi, A., Dave, D., Sabia, J.J., 2014. A Puff of Smoke: Medical Marijuana Laws and Tobacco Use, Retrieved from 〈paa2015.princeton.edu/uploads/153075〉 (15 September 2015).

utra, L.M., Glantz, S.A., 2014. Electronic cigarettes and conventional cigarette use among US adolescents: a cross-sectional study. JAMA Pediatrics 168 (7), 610–617.

lliott, D., 2014. E-Cigarette critics worry new ads will make ‘vaping’ cool for kids. NPR, Retrieved from 〈http://www.npr.org/2014/03/03/284006424/e-cigarette- critics-worry-new-ads-will-make-vaping-cool-for-kids〉 (20 April 2014).

lliott, S., 29 August, 2013. E-cigarette makers’ ads echo tobacco’s heyday. The New York Times.

airchild, A.L., Bayer, R., Colgrove, J., 2014. The renormalization of smoking? E- cigarettes and the tobacco “endgame.”. New England Journal of Medicine 370, 293–295.

ernandez, E., March, 2014. E-cigarettes: Gateway to Nicotine Addiction for U.S. Teens, Says UCSF Study. UCSF. edu, Retrieved from 〈http://www.ucsf.edu/news/ 2014/03/112316/e-cigarettes-gateway-nicotine-addiction-us-teens-says-ucsf- study〉 (28 July 2014).

ostin, L.O., Glasner, A.Y., 2014. E-cigarettes, vaping, and youth. JAMA, http://dx.doi. org/10.1001/jama.2014.7883.

rana, R.A., Benowitz, N., Glantz, S.A., 2014a. E-cigarettes: a scientific review. Circu- lation 129, 1972–1986.

rana, R.A., Popova, L., Ling, P.M., 2014b. A longitudinal analysis of electronic cigarette use and smoking cessation. JAMA Internal Medicine, http://dx.doi.org/ 10.1001/jamainternmed.2014.187.

ruber, J., Zinman, J., 2001. Youth smoking in the United States: evidence and impli- cations. In: Gruber, J. (Ed.), Risky Behavior Among Youths: An Economic Analysis. The University of Chicago Press, Chicago, IL, pp. 69–120.

uang, J., Tauras, J., Chaloupka, F.J., 2014. The impact of price and tobacco control policies on the demand for electronic nicotine delivery systems. Tobacco Control 23, iii41–iii47.

osmider, L., Sobczak, A., Fik, M., Knysak, J., Zaciera, M., Kurek, J., Goniewicz, M.L., 2014. Carbonyl compounds in electronic cigarette vapors—effects of nicotine solvent and battery output voltage. Nicotine and Tobacco Research, http://dx. doi.org/10.1093/ntr/ntu078.

ee, S., Grana, R.A., Glantz, S.A., 2013. Electronic cigarette use among Korean adoles- cents: a cross-sectional study of market penetration, dual use, and relationship to quit attempts and former smoking. Journal of Adolescent Health 54 (6), 684–690.

illard, D.R., Molloy, E., Sfekas, A., 2013. Smoking initiation and the iron law of demand. Journal of Health Economics 32 (1), 114–127.

arynak, K., Holmes, C.B., King, B.A., Promoff, G., Bunnell, R., McAfee, T., 2014. State laws prohibiting sales to minors and indoor use of electronic nicotine delivery systems—United States, November 2014. CDC Morbidity and Mortality Weekly Report 63 (49), 1145–1150.

rzechowski and Walker, 2012. The Tax Burden on Tobacco: Historical Com- pilation, 47, Retrieved from 〈http://www.taxadmin.org/fta/tobacco/papers/ tax burden 2012.pdf〉 (2 April 2014).

isinger, C., Døssing, M., 2014. A systematic review of health effects of electronic cigarettes. Preventative Medicine 69, 248–260.

olosa, R., Rodu, B., Caponnetto, P., Maglia, M., Raciti, C., 2013. A fresh look at tobacco harm reduction: the case for the electronic cigarette. Harm Reduction Journal 10, 19, http://dx.doi.org/10.1186/1477-7517-10-19.

onomics 44 (2015) 300–308

Richardson, A., Ganz, O., Stalgaitis, C., Abrams, D., Vallone, D., 2014a. Noncombustible tobacco product advertising: how companies are selling the new face of tobacco. Nicotine and Tobacco Research 16, 606–614.

Richardson, A., Ganz, O., Vallone, D., 2014b. Tobacco on the web: surveillance and characterisation of online tobacco and e-cigarette advertising. Tobacco Control, http://dx.doi.org/10.1136/tobaccocontrol-2013-051246.

Riker, C.A., Lee, K., Darville, A., Hahn, E.J., 2012. E-cigarettes: promise or peril? Nurs- ing Clinics of North America 47 (1), 159–171.

Statistic Brain Research Institute, 21, 2013. Electronic Cigarette Statistics, Retrieved from 〈http://www.statisticbrain.com/electronic-cigarette-statistics/〉 (26 March 2014).

Sutfin, E.L., McCoy, T.P., Morrell, H.E.R., Hoeppner, B.B., Wolfson, M., 2013. Elec- tronic cigarette use by college students. Drug and Alcohol Dependence 131, 214–221.

Time for e-cigarette Regulation, 2013. Lancet Oncology 14 (October), 1027 [Editorial].

Census Bureau, U.S., 2014. Table H-8A. Median Income of Households by State—Two- Year Moving Averages: 1984 to 2012, Retrieved from 〈https://www.census. gov/hhes/www/income/data/historical/household/2012/H08A 2012.xls〉 (2 June 2014).