Business Law paper

legaltobaccoagelaws.pdf

Home >Law homework help >Business Law paper

DO MINIMUM LEGAL TOBACCO PURCHASE AGE LAWS WORK?

CEREN ERTAN YÖRÜK and BARŞ K. YÖRÜK∗

This paper uses a regression discontinuity design to estimate the impact of the minimum legal tobacco purchase age (MLTPA) laws on smoking behavior among young adults. Using data from the confidential version of National Longitudinal Survey of Youth (1997 Cohort), which contains information on the exact birth date of the respondents, we find that the impact of the MLTPA on several indicators of smoking among youth is moderate but often statistically insignificant. However, for those who reported to have smoked before, we show that granting legal access to cigarettes and tobacco products at the MLTPA leads to an increase in several indicators of smoking participation, including up to a 5 percentage point increase in the probability of smoking. These results imply that policies that are designed to restrict youth access to tobacco may only be effective in reducing smoking behavior among certain groups of young adults. (JEL I10, I18, I19)

I. INTRODUCTION

Smoking is identified as a major cause of a wide variety of health problems such as heart dis- ease, stroke, and several different forms of cancer. The U.S. Surgeon General’s report (2014) states that smoking is the leading cause of preventable and premature death and smoking harms nearly every major organ of the body, often in profound ways, causing many diseases and significantly diminishing the health of smokers in general. Furthermore, compared with nonsmokers, smok- ers are more prone to illness and more likely to reach exhaustion. Empirical literature also docu- ments that smoking has a significant impact on other health related outcomes such as mental ill- ness or obesity.1 Through its depressant effects on health, smoking might also have substantive negative spillover effects on economic outcomes, including several labor market outcomes such as

∗This paper uses confidential data provided by Bureau of Labor Statistics (BLS). The views expressed in this paper are those of the authors and do not necessarily reflect those of the BLS. Ertan Yörük: Assistant Professor, School of Management,

The Sage Colleges, Albany, NY 12208. Phone 1- 518-292-1786, Fax 1-518-292-1964, E-mail yorukc@ sage.edu

Yörük: Associate Professor, Department of Economics, University at Albany, SUNY, Albany, NY 12222. Phone 1-518-442-3175, Fax 1-518-442-4736, E-mail [email protected]

1. For example, Gruber and Frakes (2006), Courte- manche (2009), Baum (2009), and Liu et al. (2010) document the relationship between smoking and obesity.

income, productivity, and wages.2 Given these direct and indirect effects of smoking, evaluating the effectiveness of the policies that are used to regulate smoking behavior is vital.

Several recent studies have documented that policies that increase the cost of smoking such as raising taxes on cigarettes and tobacco products and imposing public smoking bans significantly reduce smoking participation and may have positive spillover effects on smoking-related outcomes. For instance, DeCicca and McLeod (2008) find that higher taxes reduce smoking participation by older adults, especially those who are less educated and live in low-income households. Carpenter and Cook (2008) show that tobacco tax increases are associated with significant reductions in smoking participation

2. Leigh and Berger (1989), Levine, Gustafson, and Velenchik (1997), and Auld (2005) show that smoking is neg- atively associated with labor market outcomes.

ABBREVIATIONS

BLS: Bureau of Labor Statistics CDC: Centers for Disease Control and Prevention DLI: Date of their Last Interview MLDA: Minimum Legal Drinking Age MLTPA: Minimum Legal Tobacco Purchase Age NLSY97: National Longitudinal Survey of Youth,

1997 Cohort NYTS: National Youth Tobacco Survey STATE: State Tobacco Activities Tracking

and Evaluation

415 Contemporary Economic Policy (ISSN 1465-7287) Vol. 34, No. 3, July 2016, 415 – 429 Online Early publication October 29, 2015

416 CONTEMPORARY ECONOMIC POLICY

and frequent smoking by youths. Using rich lon- gitudinal data from the German Socioeconomic Panel Study, Anger, Kvasnicka, and Siedler (2011) investigate the impact of public smok- ing bans on smoking behavior. They find that individuals who go out more often to bars and restaurants adjust their smoking behavior as a response to a public smoking ban and as a result, they become less likely to smoke and also smoke less. On the other hand, Carpenter, Postolek, and Warman (2011) find that public-place smoking laws are effective tools at reducing nonsmokers’ and smokers’ exposure to environmental tobacco smoke in a variety of public places on a broad, population-wide scale.

According to the Substance Abuse and Men- tal Health Services Administration of the U.S. Department of Health and Human Services, around 92% of cigarette smokers started smok- ing at or before age 18. As a natural consequence, sales of tobacco products to minors are heavily regulated in the United States. For instance, recently, several states enacted laws that pro- hibit sales of electronic cigarettes to minors and indoor use of electronic cigarettes (Marynak et al. 2014). On the other hand, probably the most direct form of regulation on smoking behavior among youth in the United States is imposing a minimum legal cigarette and tobacco prod- ucts purchase age (hereafter, MLTPA). Although MLTPA is 18 in most states, there are a few states that impose a higher MLTPA.3 Understanding the effect of the MLTPA is particularly important not only because smoking has been linked to several undesirable health and economic out- comes, but also increasing the MLTPA from 18 to 19 or 21 is a current policy debate in many states. Proponents of a higher MLTPA argue that increasing the legal age for tobacco sales will make it difficult for younger teens, such as 16- year-olds, to get cigarettes since younger teens typically know a lot more 18-year-olds than 21-year-olds who might buy them cigarettes. However, opponents of a higher MLTPA argue that there is no sufficient evidence to support that raising the age for being able to buy tobacco products has any real effect on keeping young adults away from tobacco products.

Although several studies have investigated the effect of the MLTPA laws on smoking habits of young adults, most of them have made use of the changes in the MLTPA laws that occurred in the

3. Recently, New York City passed a new age purchasing restriction law raising the MLTPA to 21.

1980s and early 1990s at the state level. However, states where a MLTPA of 18 was imposed might be different in unobserved ways than those states where a higher MLTPA was enforced. If these unobserved differences at the state level are also associated with smoking habits of young adults, then one cannot estimate a consistent effect of the MLTPA on cigarette consumption and smoking- related outcomes using the simple variation of the MLTPA law at the state level. In order to address this shortcoming, we exploit the discon- tinuity in smoking habits of young adults at the MLTPA and use a regression discontinuity (here- after, RD) design to estimate the causal effect of the MLTPA on several indicators of smok- ing participation. Our main identifying assump- tion is that the observed and unobserved deter- minants of smoking-related outcomes are likely to be distributed smoothly across the MLTPA.4

Hence, any possible change in smoking habits and smoking-related outcomes at the MLTPA can solely be attributed to the MLTPA law itself.5

We use a restricted version of the National Longitudinal Survey of Youth, 1997 Cohort (NLSY97) for the empirical analysis. This restricted version contains unique information on the exact birth date of the respondents as well as their state of residence, which is quite important in the context of a RD design.6 Since the MLTPA laws differ at the state level and are imposed based on a simple age cut-off, this information enables one to clearly identify the treatment and control groups and compare the smoking habits of youths who are slightly younger than MLTPA with those who are slightly older than this cut-off age. To our best knowledge, this is the first study

4. This is a partially testable assumption. Relevant tests are presented in Section V.

5. Since our empirical analysis is based on self-reported survey data, youths who are slightly younger than the MLTPA may be more likely to underreport their smoking behavior since it is illegal to purchase cigarettes for those who are under the MLTPA. This could generate a discrete jump in reported levels of smoking participation at the MLTPA even if there is no true change in actual smoking behavior. If this is the case, then the RD estimates may be biased.

6. MLTPA differs at the state level. Therefore, the infor- mation on state of residence is crucial to identify the treatment and control groups. The information on the exact birthdate of each respondent is also crucial. Suppose that a respondent resides in a state that enforces a MLTPA of 18. Furthermore, suppose that one has information only on the month and year of the birthdate of this respondent. If this respondent was born on January 30, 1980 and interviewed on January 1, 1998, he will be mistakenly coded as an 18-year-old and included in the treatment group (those who are 18 and older). But, this respondent is actually in the control group since he is 29 days younger than 18 at the time of the interview date.

ERTAN YÖRÜK & YÖRÜK: TOBACCO PURCHASE AGE LAWS 417

to investigate the impact of the MLTPA laws at the national level using a RD design. We also investigate the effects of the MLTPA on young adults who belong to different demographic groups or who are subject to different MLTPAs such as 18 and 19.

In the United States, 18 is the age of major- ity, which is also the MLTPA in most states. This fact may potentially bias our estimates if the rights gained at the age of majority also affect the smoking habits of young adults. In order to address this problem, we test the possibility that there exists other significant changes in observ- able characteristics of young adults occurring at the MLTPA that could confound our analysis. For instance, if young adults leave their parents’ house or start to work at the MLTPA, then their employment status and household income should exhibit a discrete change at the MLTPA. How- ever, our RD estimates imply that this is not the case.

We find that the overall effect of the MLTPA on several indicators of smoking among youth is moderate but often statistically insignificant. However, we also show that granting legal access to cigarettes and tobacco products at the MLTPA leads to a statistically significant and consider- able increase in several indicators of smoking for those who reported to have smoked before. In particular, we find that for those who reported to have smoked before, the MLTPA is associated with up to a 5 percentage point increase in the probability of smoking in the past month and a 25% increase in the number of days that they smoke cigarettes per month. These results are robust under alternative parametric and nonpara- metric model specifications.

The rest of this paper is organized as follows. The next section provides background infor- mation on the MLTPA laws and discusses the relevant research. Section III presents the data and empirical methodology and discusses the relationship between the MLTPA and smoking. Section IV presents the results and discusses the sensitivity of the main findings under alternative models and for youths who belong to differ- ent demographic groups. Section V provides a discussion of policy implications and concludes.

II. BACKGROUND AND LITERATURE REVIEW

Since April 1, 1988, the sale of cigarettes and other tobacco products to people under the state purchase age has been prohibited by law in all states of the United States. Any state may set the

purchasing age limit lower than 18 years but this would result in the loss of grants from the Federal Emergency Management Agency that otherwise would be given for a natural disaster in that state. State Tobacco Activities Tracking and Evaluation (STATE) system of Centers for Disease Control and Prevention (CDC) reports the legal age limits for purchasing tobacco products for each state. In general, almost all states in the United States adopted 18 years as the MLTPA. There are few exceptions, however. Alabama, Alaska, and Utah increased their MLTPA from 18 to 19 in 1997 and have enforced the same age limit since then. New Jersey increased its MLTPA from 18 to 19 in 2006. Pennsylvania prohibited individuals under the age of 21 from buying cigarettes and tobacco products until July 10, 2002 but since then, it has enforced a lower MLTPA of 18.7

Empirical evidence on the effectiveness of the youth access laws on smoking is mixed. Chaloupka and Grossman (1996) and DeCicca, Kenkel, and Mathios (2002) find little impact of access restrictions and higher taxes on youth smoking, while Gruber and Zinman (2001) and Chaloupka and Pacula (1998) argue that such restrictions lower the quantity of cigarettes smoked by younger smokers. Chaloupka and Pacula (1998) focus on youth access restriction enforcement and find that more tightly enforced youth access restrictions lower youth smoking. Their estimates suggest that when the limits on youth access are comprehensively and aggres- sively enforced and highly complied with, they significantly reduce the prevalence of youth smoking. Using data from the 1992 and 1993 waves of the Current Population Survey, Hersch (1998) argues that unlike higher cigarette taxes, age-related smoking restrictions have little effect on teen smoking. Recently, in a dynamic simula- tion model, Ahmad and Billimek (2007) find that comparable to a large tax-induced price increase, raising MLTPA to 21 would have a minimal immediate effect on adult smoking prevalence and population health, but would cause almost a 13% drop in smoking prevalence for 15 – 17- year-olds. Using data from Monitoring the Future Survey, DiFranza, Savageau, and Fletcher (2009) find that improving merchant compliance with the prohibition on sales of tobacco to minors

7. Suffolk, Nassau, and Onondaga Counties of New York introduced a higher MLTPA of 19 in 2005, 2006, and 2009, respectively. Since our empirical analysis uses data from 1998 to 2004, it is solely based on the age limits that are enforced at the state level.

418 CONTEMPORARY ECONOMIC POLICY

TABLE 1 Definition of Outcome Variables and Summary Statistics

Variable Name Definition Full

Sample Age <

MLTPA Age >= MLTPA

Smoked Since the DLI

Smoke =1 if the respondent smoked in the last 30 days

0.354 0.309 0.388 0.818 (0.478) (0.462) (0.487) (0.386)

[26,545] [11,558] [14,987] [10,549]

Smoking days Number of days that the respondent smoked in the last 30 days

7.190 5.887 8.160 16.616 (12.001) (11.111) (12.535) (13.275) [26,545] [11,558] [14,987] [10,549]

No. of cigs. Number of cigarettes that the respondent smoked in the last 30 days on the days she smoked

2.921 2.268 3.406 8.604 (6.484) (5.801) (6.910) (8.660)

[26,549] [11,556] [14,993] [8,202]

Avg. cigs Average number of cigarettes smoked by the respondent per day in the last 30 days

2.629 1.990 3.105 7.763 (6.347) (5.587) (6.820) (8.894)

[26,517] [11,547] [14,970] [8,171]

Notes: Sample weighted means are reported. Standard deviations are reported in parenthesis. Number of observations is reported in brackets.

and increasing the price of cigarettes discourage youth smoking.

Past studies that investigate the effect of the MLTPA laws on smoking have two major lim- itations. First, most of these studies cover data from the late 1980s and early 1990s. However, the youth access laws are widely perceived to have been ineffective during this time period due to lack of enforcement (DeCicca et al. 2008). Sec- ond, most of the earlier studies used the state level variation in the MLTPA laws to identify the effect of these laws on smoking behavior. However, states where a lower MLTPA was imposed might be different in unobserved ways than those states where the MLTPA of 19 or 21 was enforced. If these unobserved differences at the state level are also correlated with the smoking habits of young adults, then one cannot estimate a consis- tent effect of the MLTPA laws on smoking and smoking-related outcomes using the simple vari- ation of the MLTPA law at the state level. The RD approach used in this paper alleviates this short- coming by removing the bias from unobserved policy preferences.

Our empirical approach is similar to that of Yan (2014) who employs a RD design to inves- tigate the effect of the MLTPA of 21 in Pennsyl- vania on smoking behavior of young mothers. He finds that the MLTPA in Pennsylvania is associ- ated with a 16% increase in the average number of cigarettes smoked per day among young mothers. He also shows that for smoking mothers, having

legal access to cigarettes just before the end of the first trimester would lower an infant’s birth weight by 60 grams. However, our study differs from that of Yan (2014) in several ways. First, instead of focusing on a single state, we investi- gate the impact of the MLTPA laws at the national level. Second, we employ a different individual level survey (NLSY97) that contains information on the exact birth date and state of residence of the respondents. This data set has also the advan- tage of containing a more comprehensive range of smoking outcomes than previous research.

III. DATA AND EMPIRICAL METHODOLOGY

We use data from the NLSY97 for the empir- ical analysis. The NLSY97 consists of a nation- ally representative sample of 9,022 youths who were 12 – 16 years old as of December 31, 1996. Round 1 of the survey took place in 1997. In that round, both the eligible youth and one of that youth’s parents received hour-long personal interviews. Youths continue to be interviewed on an annual basis. In addition to standard demo- graphic information, the survey respondents were also asked detailed questions about their smoking habits. Our outcome variables are derived from these questions. We present the description of these variables and their summary statistics in Table 1 for the full sample, for those who are younger or older than the MLTPA, and for those

ERTAN YÖRÜK & YÖRÜK: TOBACCO PURCHASE AGE LAWS 419

who smoked at least once since the date of their last interview (DLI).8

A unique feature of our data set is that we have obtained access to a confidential version of the NLSY97 with information on respondents’ state of residence, exact date of birth, and exact interview date for each survey year. We use this information to determine the MLTPA for each state and to calculate the exact age in days for each respondent at the time of the interview. Therefore, for each respondent, we were able to determine the MLTPA that she is subject to and the number of days that she is younger or older than the MLTPA. We restrict our sample to those respondents who were surveyed over the period 1998 – 2004, were up to 2 years younger or older than the MLTPA, and were single as of the interview date.9 As in similar surveys of its kind, the respondents of the NLSY97 who reported to have smoked at least once since the DLI were also asked about their smoking habits over the past month.10 This relatively short ref- erence period is desirable since our empirical strategy compares those who are slightly older than the MLTPA with those who are slightly younger than this cutoff age. In order to inves- tigate the impact of the MLTPA on smoking behavior, we consider four main smoking out- comes. These variables are whether the respon- dent smoked over the past month, number of days that she smoked over the past month, number

8. In Table 1, we report the sample weighted means for outcome variables. However, following Yörük (2014), we do not use sample weights in our regressions. In tech- nical sampling report of the NLSY97, Moore et al. (2000) argue that using NLSY97 weights to perform weighted least squares when doing regression analysis may lead to incor- rect estimates. Even in correctly specified models, using sample weights may also increase the variance of esti- mates and lead to loss of efficiency (DuMouchel and Dun- can 1983). Although not reported, sample weighted regres- sions yield comparable estimates. These results are available upon request.

9. Our choice of age bandwidth follows Carpenter and Dobkin (2009) and Yörük and Ertan Yörük (2011, 2013). However, we also present results for alternative age band- widths. The majority of states in the United States requires that a couple be 18 in order to marry without parental permis- sion. Marital status may significantly affect smoking behavior of youths. Given that the MLTPA in most states is 18 as well, including married youths in our sample may cause an identi- fication problem. Therefore, in this paper, we focus on single youths only. Accordingly, we dropped 532 observations, who reported being married as of the interview date.

10. For instance, in the National Youth Tobacco Survey (NYTS), questions on smoking habits typically refer to the prior month. However, NYTS does not provide information on the exact birth date of the respondent and therefore, is not suitable for our empirical strategy.

of cigarettes that she smoked on the days that she actually smoked, and the average number of cigarettes that she smoked per day over the past month.11

We use a RD design to estimate effect of the MLTPA laws on smoking and smoking-related outcomes among young adults.12 This approach exploits the sudden change in smoking habits that may occur at the MLTPA. Since purchase of tobacco products are legally allowed accord- ing to a simple age cutoff, we are able to com- pare outcomes across youths with similar income and other observable individual characteristics, but with very different smoking habits. In par- ticular, this approach compares the individuals who are slightly younger than the MLTPA with those who are slightly older than this age cut- off. Since only a few days separates these groups, it is unlikely that observable and unobservable characteristics of these groups are significantly different. The only difference between these two groups is likely to be the status of their legal access to tobacco products. Therefore, the change in smoking habits at the MLTPA can solely be attributed to the MLTPA law itself. The main RD model used in our empirical analysis can be expressed as follows:

(1) Yit = β ′Xit + γTit + g

( Ait

) + ηt + μi + εit

where Y it represents a particular smoking out- come for individual i at time t. The vector of time dependent observable characteristics for individ- ual i is denoted by Xit and includes dummy variables controlling for household income, stu- dent and employment status of the respondent, and a dummy variable which controls for the birthday celebration effect and equals to one if

11. We do not directly observe the binary variable that measures whether the respondent smoked over the past month. For those respondents who never smoked or did not smoke since the DLI, this binary variable is equal to zero. The respondents who reported smoking at least once since the DLI were asked the following question: "During the past 30 days, on how many days did you smoke a cigarette?" The smoking participation variable for the corresponding question is coded unity if the respondent reported smoking on at least one day during the past month and zero otherwise. The respondents were also asked the following question: "When you smoked a cigarette during the past 30 days, how many cigarettes did you usually smoke each day?" In order to calculate the aver- age number of cigarettes that the respondent smoked per day over the past month, we multiply the number of days that the respondent smoked over the past month with the number of cigarettes that she smoked on those days and divide the result by 30.

12. Imbens and Lemieux (2008), Porter (2003), and Lee and Lemieux (2009) present a detailed discussion of the RD design.

420 CONTEMPORARY ECONOMIC POLICY

the respondent was interviewed during the first month after turning the MLTPA.13 In general, these control variables vary smoothly around the MLTPA. Hence, they have little effect on our estimates. Since our data contain information on some youths at more than one time period, in our empirical analysis, we control for time- invariant individual characteristics (μi) and fixed year effects (ηt) and report the standard errors that are clustered at the individual level.14 The treatment variable is denoted by T it and takes the value of unity if the respondent’s age is greater than or equal to the MLTPA or alternatively, if she is legally allowed to purchase tobacco prod- ucts as of the interview date and zero other- wise. Since we consider 1998 – 2004 period, the MLTPA is 19 for those who reside in Alabama, Alaska, and Utah, 21 for those who reside in Pennsylvania and were interviewed before July 10, 2002, and 18 for the rest of the respondents. The coefficient γ, our main coefficient of inter- est, indicates the impact of the MLTPA law on the relevant outcome variable. Finally, g(Ait) is a smooth function of age profile, which is also known as the forcing variable in the context of a RD design.15 Since we observe the state of resi- dence, and exact birth and interview date for each respondent, we were able to calculate the differ- ence between the interview date and the MLTPA. Therefore, for each respondent, the variable Ait represents the number of days before or after the MLTPA.

Modeling the smooth function of age pro- file correctly is one of the main problems in

13. Since several observations are missing for house- hold income and employment status, we use dummy vari- ables controlling for the "missing" observations for these set of covariates. Following Carpenter and Dobkin (2009), the birthday celebration effect dummy controls for the pos- sibility that young adults may simply change their smok- ing behavior during the first few days following their birth- day. Furthermore, outcome variables refer to a 1-month period. Therefore, those who were interviewed during the first month after turning the MLTPA may be mistakenly placed in the treatment group. The birthday celebration effect dummy controls for this possibility as well. We also esti- mate models using state fixed effects as additional con- trols. Although not reported here, compared with our results, these models produce a similar effect of the MLTPA on outcome variables.

14. Some of the time-invariant individual characteristics such as gender and race are observed. In our empirical model, these variables are captured by the individual fixed effects. Figure S1 in the online Supporting Information shows that these variables are distributed smoothly around the MLTPA.

15. In order to implement the RD design, we assume that the respondents do not have any control over the forcing variable. Since our forcing variable is age, this condition is naturally satisfied.

implementing the RD design. Our most general parametric model with a quadratic polynomial of age that is fully interacted with the treatment vari- able can be written as:

Yit = β ′Xit + γTi +

k=2∑ j=1

αjA j it +

k=2∑ j=1

λj (

Tit × A j it

) (2)

+ ηt + μi + εit.

We test the sensitivity of our parametric mod- els to a variety of functional form assumptions. In particular, we focus on linear and quadratic models, allowing the slope of these functions to vary on each side of the age cutoff (i.e., linear and quadratic splines). We also estimate models with and without interaction terms and for alternative subsamples of respondents.

In order to test whether our results are robust to model specification, we also estimate non- parametric RD models that do not impose any functional form assumptions. In these models, following Hahn, Todd, and van der Klaauw (2001) and Porter (2003), we use local linear regressions to estimate the left and right limits of discontinuity at the MLTPA. The difference between the two limits at the MLTPA is the local treatment effect of the MLTPA law on outcome variables. Following Malamud and Pop-Eleches (2011), we estimate this in one step using a triangular kernel which has been shown to be boundary optimal by putting more weight on observations closer to the cutoff point (Cheng, Fan, and Marron 1997).16 The remain- ing estimation issue for nonparametric models is the selection of appropriate bandwidth. Since the RD is identified only at the discontinuity, one has to balance the goals of staying as local to the cutoff point at the MLTPA as possible while ensuring that there exists enough data to yield informative estimates. Although there is currently no widely agreed-upon method for selection of optimal bandwidths in the nonpara- metric RD context, we follow a new method proposed by Imbens and Kalyanaraman (2012). For nonparametric models, we calculate the standard errors using the bootstrap procedure with 1,000 replications. This approach may offer more accurate asymptotic inference than the analytic standard errors (Cameron and Trivedi 2005).

16. We use the "rd" command in STATA to estimate our non-parametric models.

ERTAN YÖRÜK & YÖRÜK: TOBACCO PURCHASE AGE LAWS 421

TABLE 2 Test of the Smoothness of Observable Characteristics Around the MLTPA

Outcome

Student Income Black Hispanic Female Employed

T −0.017 −0.107 0.017 0.016 0.017 0.021 (0.013) (0.118) (0.013) (0.013) (0.014) (0.017)

A × 100 −0.029 0.078 0.003 0.003 0.002 0.017 (0.005)*** (0.076) (0.005) (0.005)** (0.006) (0.007)**

A2 × 10,000 −0.002 −0.006 0.001 0.001 0.001 −0.003 (0.001)*** (0.012) (0.001) (0.001)*** (0.001) (0.001)***

T × A × 100 −0.082 −0.059 −0.007 −0.007 −0.007 −0.004 (0.007)*** (0.083) (0.006) (0.006) (0.007) (0.010)

T × A2 × 10,000 0.011 0.003 −0.000 −0.000 −0.000 0.004 (0.001)*** (0.013) (0.001) (0.001)** (0.001) (0.001)***

Constant 0.814 10.165 0.258 0.258 0.482 0.504 (0.009)*** (0.096)*** (0.009)*** (0.009)*** (0.011)*** (0.012)***

Mean 0.699 10.037 0.265 0.209 0.485 0.508 [0.459] [2.474] [0.441] [0.407] [0.500] [0.500]

No. of obs. 26,738 15,276 26,738 26,738 26,738 26,734 R2 0.1565 0.0027 0.0002 0.0003 0.0002 0.0459

Notes: T is a binary treatment variable which is equal to 1 if the respondent’s age is greater than or equal to the MLTPA as of the interview date. For each respondent, A denotes the difference in days between the interview date and the MLTPA that she is subject to. The signs ** and *** denote the statistical significance at the 5% and 1% significance levels, respectively. Robust standard errors clustered at the individual level are reported in parenthesis. Standard deviations are reported in brackets.

IV. RESULTS

We start our analysis by testing whether the RD design yields credible estimates of the impact of the MLTPA law on smoking behavior. One possible concern is whether the interview date for each respondent is random. For instance, if interviews always take place on the first day of the month and birthdays are heaped in nonrandom ways, this could potentially bias our estimates since the running variable is no longer evenly distributed. To address this concern, in Figure S1 (Supporting Information), we plot histograms of day of birth, day of interview, and running variable for a bandwidth of 30 days. Panels A and B clearly show that the distributions of day of birth and day of interview are random. Similarly, Panel C of this figure shows that the distribution of the running variable around the cutoff MLTPA is random.

In the United States, 18 is the age of major- ity, which is also the MLTPA in most states. In addition to a right to purchase cigarettes and tobacco products, young adults gain several other rights at this age. If these rights are correlated with smoking behavior among young adults, then failure to control for them may bias the esti- mated impact of the MLTPA laws on smoking and smoking-related outcomes. Probably, most of the rights gained at age 18 such as rights to vote, to make a will, to sign a contract, and to

apply for a credit are uncorrelated with smok- ing habits. However, at age 18, young adults also gain rights that will allow them to be indepen- dent from parental control and therefore, they may move out from their parents’ house or may start to work instead of pursuing college edu- cation. These changes may significantly affect smoking habits. In order to address this problem, we test the possibility that there exist other sig- nificant changes in observable characteristics of young adults occurring at the MLTPA that could confound our analysis. For instance, if young adults leave their parents’ house or start to work at the MLTPA, then their observable character- istics such as employment status and household income should exhibit a discrete change at the MLTPA. We estimate Equation (2) separately for each observable covariate using a quadratic spline.17 The results reported in Table 2 suggest that for each covariate, the coefficient of the treat- ment variable is insignificant and hence, observ- able characteristics of young adults are orthogo- nal to the age variable and there is no evidence of significant discontinuous change in any of these variables at the MLTPA.18 We also graph

17. Following the previous literature, our selection of a quadratic polynomial is a result of a visual inspection of data for the best fit. Estimating this model separately for all control variables using linear or cubic splines yields similar results.

18. In column 2, we test whether ln(income) is distributed smoothly around the MLTPA. In empirical models, rather

422 CONTEMPORARY ECONOMIC POLICY

the corresponding age profiles of selected covari- ates in Figure S2 (Supporting Information). The quadratic prediction of each variable appears to fit the actual data well and exhibits either no or an insignificant small jump at the MLTPA. Although we cannot directly test whether the unobservable characteristics of youths vary smoothly across the MLTPA, our finding that observable char- acteristics are smoothly distributed around the MLTPA reduces the concerns about omitted vari- ables bias and suggests that parametric models estimated with or without controls should yield similar results.

A. The Effect of the MLTPA on Smoking

In Table 3, we report the estimates from para- metric regressions of the effect of the MLTPA on alternative measures of smoking participa- tion.19 These regressions contain a quadratic polynomial of age which is fully interacted with a dummy variable indicating an age greater than or equal to the MLTPA. Standard errors are clustered at the individual level to correct for the nonindependence of individual observations over time. The first three columns report the impact of the MLTPA on the probability of smoking in the past month. The results from the model that is estimated without control variables and individual fixed effects are quite similar compared with the model that contains control variables but not individual fixed effects and the model that contains both control variables and individual fixed effects. Under these alter- native specifications, the MLTPA is associated with around 1.9 – 2.9 percentage point increase in the probability of smoking. The summary statistics provided in Table 1 shows that slightly more than 35% of the respondents reported to have smoked in the last 30 days. Therefore, a 1.9 – 2.9 percentage point increase in the prob- ability of smoking at the MLTPA corresponds to a 5.4% – 8.3% increase from the mean of this variable. However, this effect is not statis- tically significant at conventional significance levels. In panel A of Figure S3 (Supporting Information), we superimpose the quadratic fitted lines from the parametric model estimated

than using a continuous measure of income, we use dummy variables controlling for different income ranges and missing information on income. Models estimated using ln(income) yield comparable results and are available upon request.

19. In addition to time dependent observable character- istics, specifications 2, 5, 8, and 11 of Table 3 also con- tain controls for time-invariant characteristics such as dummy variables for being female, black, and Hispanic.

without any controls (the first specification in Table 3) over the mean value of the percent of smokers calculated for each 30-day age block. The figure confirms the estimation results and shows a moderate but statistically insignifi- cant increase in the probability of smoking at the MLTPA.20

The next three columns in Table 3 document the relationship between the number of days that young adults smoke per month and the MLTPA. In our regressions, we use the log transforma- tion of this variable as the dependent variable, therefore the coefficient on the treatment shows the percentage change in smoking days due to the effect of the MLTPA law.21 The estimated coefficients on the treatment variable suggest that youths tend to increase the number of days that they smoke cigarettes per month by 8.8% – 12% at the MLTPA cutoff. Given that on average, the respondents smoke 7.2 days per month, this effect corresponds to a 0.67 – 0.86 day increase in smoking days per month. However, this effect is not statistically significant. Panel B of Figure S3 (Supporting Information) also shows a rel- atively small but statistically insignificant jump in the number of days that young adults smoke cigarettes per month at the MLTPA.

Specifications 7 – 9 in Table 3 report the esti- mated effect of the MLTPA law on the num- ber of cigarettes that young adults smoke on the days they actually smoke.22 On average, young adults smoke 2.9 cigarettes on the days they actually smoke. Under alternative model specifications, although the number of cigarettes

20. Since the MLTPA is a pre-determined cutoff age known to public, individuals may inter-temporally substitute smoking before the MLTPA to smoking after the MLTPA. If this is the case, the key identifying assumption of the RD design that the errors are independent of the running variable will be violated. If individuals substitute smoking just before the MLTPA, one should observe a significant decrease in smoking trends just before the MLTPA cutoff. However, Figure 3 shows that this is not the case. There is no significant drop in any of the smoking variables just before the MLTPA cutoff.

21. In this model, the dependent variable is ln(Smoking days + 0.1). We add 0.1 to the outcome variable since some youths do not smoke. Alternative transformations such as adding 0.5 to this variable before taking its log yield similar results.

22. If the respondent is a nonsmoker, this variable is coded as zero. In our regressions, we use the log transforma- tion of this variable as the dependent variable. In particular, in specifications 7 – 9 of Table 3, the dependent variable is ln(No. of cigs. + 0.1). Therefore, the coefficient on the treat- ment variable represents the percentage change in the number of cigarettes smoked due to the MLTPA law. Similarly, in the last three columns of Table 3, the dependent variable is ln(Avg. cigs + 0.1).

ERTAN YÖRÜK & YÖRÜK: TOBACCO PURCHASE AGE LAWS 423

T A

B L

E 3

T he

E ff

ec t

of th

e M

L T

PA on

S m

ok in

O u

tc om

S m

ok e

S m

ok in

g D

ay s

N o.

of C

ig s.

A vg

.C ig

(1 )

(2 )

(3 )

(4 )

(5 )

(6 )

(7 )

(8 )

(9 )

(1 0)

(1 1)

(1 2)

T 0.

01 9

0. 02

9 0.

02 0

0. 08

8 0.

12 0

0. 08

9 0.

04 6

0. 07

7 0.

06 7

0. 03

7 0.

05 5

0. 06

5 (0

.0 15

) (0

.0 16

) (0

.0 13

) (0

.0 74

) (0

.0 77

) (0

.0 59

) (0

.0 59

) (0

.0 62

) (0

.0 47

) (0

.0 55

) (0

.0 57

) (0

.0 41

) A ×

10 0

0. 00

1 −

0. 00

7 −

0. 00

9 0.

02 5

− 0.

01 7

− 0.

03 3

0. 02

9 −

0. 00

4 −

0. 02

5 0.

03 6

0. 00

1 −

0. 01

8 (0

.0 06

) (0

.0 06

) (0

.0 07

) (0

.0 31

) (0

.0 31

) (0

.0 34

) (0

.0 24

) (0

.0 24

) (0

.0 26

) (0

.0 22

) (0

.0 22

) (0

.0 24

) A

2 ×

10 ,0

00 −

0. 00

1 −

0. 00

2 −

0. 00

1 −

0. 00

6 −

0. 00

9 −

0. 00

5 −

0. 00

4 −

0. 00

6 −

0. 00

4 −

0. 00

3 −

0. 00

5 −

0. 00

4 (0

.0 01

) (0

.0 01

)* *

(0 .0

01 )

(0 .0

04 )

(0 .0

04 )*

* (0

.0 03

) (0

.0 03

) (0

.0 03

)* *

(0 .0

03 )

(0 .0

03 )

(0 .0

03 )

(0 .0

02 )

T ×

A ×

10 0

0. 01

6 −

0. 00

4 0.

01 0

0. 06

4 −

0. 03

7 0.

03 4

0. 05

1 −

0. 03

7 0.

02 2

0. 04

1 −

0. 04

0 0.

01 3

(0 .0

08 )

(0 .0

09 )

(0 .0

08 )

(0 .0

39 )

(0 .0

42 )

(0 .0

35 )

(0 .0

31 )

(0 .0

33 )

(0 .0

27 )

(0 .0

28 )

(0 .0

30 )

(0 .0

24 )

T ×

A 2 ×

10 ,0

00 0.

00 0

0. 00

3 0.

00 0

− 0.

00 3

0. 01

2 0.

00 1

− 0.

00 4

0. 00

9 0.

00 1

− 0.

00 5

0. 00

8 0.

00 1

(0 .0

01 )

(0 .0

01 )*

* (0

.0 01

) (0

.0 06

) (0

.0 06

)* *

(0 .0

05 )

(0 .0

05 )

(0 .0

05 )*

* (0

.0 04

) (0

.0 04

) (0

.0 04

) (0

.0 03

) C

on st

an t

0. 30

5 0.

55 0

0. 26

4 −

0. 80

9 0.

61 3

− 1.

03 9

− 1.

14 5

0. 16

3 −

1. 37

8 −

1. 32

4 −

0. 13

1 −

1. 53

0 (0

.0 11

)* **

(0 .0

22 )*

** (0

.0 39

)* **

(0 .0

54 )*

** (0

.1 14

)* **

(0 .1

82 )*

** (0

.0 43

)* **

(0 .0

98 )

(0 .1

40 )*

** (0

.0 40

)* **

(0 .0

88 )

(0 .1

32 )*

** C

on tr

ol s

N o

Y es

N o

Y es

N o

Y es

N o

Y es

In d.

fi xe

d ef

fe ct

s N

o N

o Y

es N

o N

o Y

es N

o N

o Y

es N

o N

o Y

es N

o. of

ob s.

26 ,5

45 26

,5 45

26 ,5

45 26

,5 45

26 ,5

45 26

,5 45

26 ,5

49 26

,5 49

26 ,5

49 26

,5 17

26 ,5

17 26

,5 17

R 2

0. 00

8 0.

07 4

0. 02

4 0.

01 0

0. 09

5 0.

03 6

0. 01

2 0.

10 7

0. 04

1 0.

01 2

0. 11

2 0.

04 7

N o

te s:

L og

tr an

sf or

m ed

ou tc

om es

ar e

us ed

as de

pe nd

en t

va ri

ab le

s ex

ce pt

fo r

th e

fi rs

t ou

tc om

e. T

is a

bi na

ry tr

ea tm

en t

va ri

ab le

w hi

ch is

eq ua

l to

1 if

th e

re sp

on de

nt ’s

ag e

is gr

ea te

r th

an or

eq ua

l to

th e

M L

T PA

as of

th e

in te

rv ie

w da

te .

F or

ea ch

re sp

on de

nt ,

A de

no te

s th

e di

ff er

en ce

in da

ys be

tw ee

n th

e in

te rv

ie w

da te

an d

th e

M L

T PA

th at

sh e

is su

bj ec

t to

. R

eg re

ss io

ns th

at in

cl ud

e co

nt ro

l va

ri ab

le s

al so

in cl

ud e

bi rt

hd ay

ce le

br at

io n

du m

m y

an d

su rv

ey ye

ar fi

xe d

ef fe

ct s.

T he

co nt

ro l

va ri

ab le

s ar

e di

sc us

se d

in th

e te

xt .

T he

si gn

s **

an d

** *

de no

te th

e st

at is

ti ca

l si

gn ifi

ca nc

e at

th e

5% an

d 1%

si gn

ifi ca

nc e

le ve

ls ,r

es pe

ct iv

el y.

R ob

us t

st an

da rd

er ro

rs cl

us te

re d

at th

e in

di vi

du al

le ve

l ar

e re

po rt

ed in

pa re

nt he

si s.

424 CONTEMPORARY ECONOMIC POLICY

smoked increases by 4.6% – 7.7% due to the MLTPA (0.13 – 0.22 cigarette increase from the mean), this effect is not statistically significant at conventional significance levels. Panel C of Figure S3 (Supporting Information) also illus- trates this result and shows a small and insignif- icant change in the number of cigarettes that young adults smoke on the days they actually smoke at the MLTPA.

The last three columns of Table 3 show that the average number of cigarettes smoked per day increases by 3.7% – 6.5% at the MLTPA. The mean of this variable is 2.6. Therefore, this effect corresponds to a 0.1 – 0.17 cigarette increase from the mean. However, similar to other smoking- related outcome variables, this effect is not sig- nificant at conventional significance levels. Panel D of Figure S3 (Supporting Information) illus- trates this finding and shows that smoking behav- ior among young adults exhibits similar trends around the MLTPA cutoff.

B. Alternative Samples

The effect of the MLTPA on smoking behavior might be different in states that enforce alterna- tive MLTPAs. In the first two specifications of Table 4, we investigate the effect of the MLTPA on smoking behavior among young adults who are subject to different MLTPAs such as 18 and 19.23 In order to create a subsample of youths who are subject to a MLTPA of 18, we drop the respondents who reside in Alabama, Alaska, and Utah and those who reside in Pennsylvania and were interviewed before July 10, 2002 from the full sample. Similarly, in order to create a sub- sample of youths who are subject to a MLTPA of 19, we only use data from the respondents who reside in Alabama, Alaska, and Utah at the time of the interview. The coefficients from the para- metric models that contain a full set of controls and are estimated using a quadratic spline and interaction terms imply that the MLTPA of 18 is associated with a 1.5 percentage point increase in the probability of smoking, a 6.7% increase in the number of days that young adults smoke cigarettes per month, a 5.5% increase in the

23. We do not report the effect of the MLTPA of 21 on smoking outcomes for two main reasons. First, at age 21, young adults also gain legal access to alcohol, which may affect smoking participation and confound our analysis. Second, in our sample, the number of young adults who are subject to a MLTPA of 21, i.e., those who reside in Pennsylvania and were interviewed before July 10, 2002, is 204. Due to the small sample size, the results from the RD regressions may not yield credible estimates.

number of cigarettes smoked on the days that they actually smoked cigarettes, and a 5.7% increase in the average number of cigarettes smoked per day. These estimates are slightly smaller than but quite similar to the results from the full sample and remain statistically insignificant. On the other hand, we find that the MLTPA of 19 is associ- ated with a 4.9 percentage point increase in the probability of smoking and a 21.8% increase in the number of smoking days per month. How- ever, neither of these estimates are statistically significant at conventional significance levels. We also acknowledge that these estimates may not be informative and should be interpreted with cau- tion due to the large standard errors.

Young adults may be placed in different school cohorts and hence, subject to differ- ent peer effects based on their date of birth. Although school entry age rules differ across states and even school districts can set their own rules, almost all states require students to be at a certain age before August, September, or October 1st of a given year. Therefore, two students who were born on the same year but at different months may be one grade apart from each other. In order to test whether this possi- bility affects our results, we estimate separate regressions for those who were born in May, June, or July (Young for grade) and those who were born in November, December, or January (Old for grade). Suppose that student A was born on July 30, 1980 while student B was born on November 1st of the same year. Although student A is only 3 months older than student B, she will start school a full year early due to school entry age laws. However, student A will be among the youngest in her cohort and will reach the MLTPA later than her peers while student B will be among the oldest in her cohort and will reach the MLTPA earlier than her peers. The results reported in Table 4 show that the impact of the MLTPA on smoking habits of those who gain access to tobacco products earlier than their peers is relatively smaller compared with those who gain access to tobacco prod- ucts later than their peers. However, for both groups, the effect of the MLTPA law on alter- native indicators of smoking behavior remains statistically insignificant.

We also test whether the response of those who were born in summer to the MLTPA law is signif- icantly different than the rest of the sample. This is possible since those who were born in sum- mer will also reach the MLTPA during summer when students are out of school. For this analysis,

ERTAN YÖRÜK & YÖRÜK: TOBACCO PURCHASE AGE LAWS 425

TABLE 4 The Effect of the MLTPA on Smoking: Alternative Samples

Outcome

Smoke Smoking Days No. of Cigs. Avg. Cigs.

Alternative MLTPAs MLTPA of 18 0.015 0.067 0.055 0.057

(0.016) (0.061) (0.048) (0.042) [25,274] [25,274] [25,276] [25,247]

MLTPA of 19 0.049 0.218 0.017 −0.041 (0.067) (0.295) (0.224) (0.192) [844] [844] [845] [843]

Alternative samples Young for grade 0.063 0.137 0.071 −0.012

(0.049) (0.226) (0.180) (0.165) [6,471] [6,471] [6,463] [6,461]

Old for grade 0.007 0.076 0.082 0.109 (0.024) (0.106) (0.085) (0.076) [6,824] [6,824] [6,830] [6,818]

Summer birthdays 0.108 0.421 0.228 0.176 (0.052)** (0.241) (0.189) (0.172)

[6,733] [6,733] [6,726] [6,721] Smoked since the DLI 0.051 0.247 0.095 0.162

(0.025)** (0.118)** (0.057) (0.101) [10,549] [10,549] [8,202] [8,171]

Smoked at least once since age 16 0.034 0.150 0.111 0.106 (0.024) (0.105) (0.083) (0.074) [14,829] [14,829] [14,833] [14,801]

Male 0.031 0.137 0.119 0.104 (0.019) (0.085) (0.068) (0.061) [13,670] [13,670] [13,662] [13,654]

Female 0.011 0.053 0.029 0.034 (0.018) (0.081) (0.063) (0.055) [12,875] [12,875] [12,887] [12,863]

Non-Hispanic and non-black 0.012 0.059 0.051 0.062 (0.018) (0.083) (0.063) (0.062) [13,979] [13,979] [13,983] [13,968]

Hispanic or black 0.029 0.132 0.096 0.079 (0.019) (0.083) (0.063) (0.052) [12,566] [12,566] [12,566] [12,549]

Income less than $20,000 0.029 −0.020 −0.031 −0.110 (0.047) (0.226) (0.179) (0.164) [4,516] [4,516] [4,517] [4,510]

Notes: Log transformed outcomes are used as dependent variables in the last three columns. All regressions are estimated using a parametric model that contains a quadratic polynomial of age, which is fully interacted with the treatment variable. All regressions also include birthday celebration dummy, individual and year fixed effects, and time variant control variables as discussed in the text. The sign ** denotes the statistical significance at the 5% significance level. Robust standard errors clustered at the individual level are reported in parenthesis. Number of observations is reported in brackets.

we restrict our sample to those who were born in June, July, or August. The estimation results presented in Table 4 show that for this group of young adults, the MLTPA is associated with a 10.8 percentage point increase in the probability of smoking.

Young adults who have never smoked until the MLTPA are unlikely to change their smoking behavior once they are granted legal access to tobacco products. However, MLTPA laws may significantly affect the smoking behavior of those who have smoked before. We test this pos- sibility in Table 4. When the sample is restricted to those who smoked at least once since the

DLI, the impact of the MLTPA on smoking behavior considerably increases and becomes significant. In particular, for those who have smoked before, the MLTPA is associated with a 5.1 percentage point increase in the probabil- ity of smoking in the past month and a 24.7% increase in the number of days that they smoke cigarettes per month. Table 1 shows that on aver- age, those who smoked at least once since the DLI smoke 16.6 days a month. Therefore, those who smoked at least once since the DLI tend to increase their smoking days approximately 4.1 days once they gain legal access to tobacco products.

426 CONTEMPORARY ECONOMIC POLICY

We further extend our analysis of the effect of the MLTPA laws on those who have smoked before and estimate the effect of this policy on those who smoked at least once since their 16th birthday. Our results show that for this group of young adults, the MLTPA is associated with a 3.4 percentage point increase in the probability of smoking in the past month, a 15% increase in the number of days that they smoke cigarettes per month, and a 11.1% increase in the number of cigarettes that they smoke on the days they actually smoke. However, these estimates are not significant at conventional significance levels.

The gender differences in smoking behav- ior are well documented (Bauer, Göhlmann, and Sinning 2007). We also investigate whether the impact of the MLTPA on smoking behavior dif- fers by gender. Table 4 shows that the MLTPA is not a significant determinant of smoking behav- ior for females. For males, the MLTPA is asso- ciated with a 3.1 percentage point increase in the probability of smoking, a 11.9% increase in the number of cigarettes that they smoke on the days they actually smoke, and a 10.4% increase in the average number of cigarettes smoked per day. However, these estimates are not significant at conventional significance levels.

In the last three specifications of Table 4, we test whether the effect of the MLTPA law on smoking behavior differs by race and income. Estimating separate models for different groups of young adults, that is, non-Hispanics and non-blacks, Hispanic or blacks, and those whose household income is less than $20,000, we find that the effect of the MLTPA law on smoking behavior among young adults remains insignificant.24

Table S1 (Supporting Information) reports the estimated impact of the MLTPA on smoking outcomes under alternative parametric and non- parametric specifications. We first test whether our results robust to the exclusion of the birth- day celebration effect dummy from the set of the control variables. Compared with the main results reported in Table 3, the results from this alternative specification are very similar. Next, we change the age of the MLTPA to be one year earlier than it actually is in each state and call this new variable “placebo MLTPA.” Compared with the main results, the coefficient of the placebo MLTPA dummy changes sign and becomes neg- ative. Furthermore, except for average cigarettes

24. Household income is calculated in 2005 dollars. U.S. Census poverty threshold for a family of four in 2005 was $19,971.

smoked per day, the effect of the placebo MLTPA on other outcomes is statistically insignificant. When we replicate the same analysis by restrict- ing the sample to those states where the MLTPA is 18 (the placebo MLTPA is 17), we find that the effect of the placebo dummy on all outcomes is statistically insignificant. In the remaining specifications of Table S1, we test whether our results are sensitive to selection of different functional forms. The effect of the MLTPA on smoking among young adults is significant under parametric models that are estimated using linear polynomial of age and under the model that is estimated using a quadratic polynomial of age but without any interaction terms. In particular, the results from these models imply that the prob- ability of smoking among young adults increases up to 1.7 percentage points at the MLTPA. Similarly, young adults tend to increase the number of days that they smoke cigarettes by up to 9%, the number of cigarettes that they smoke on the days they actually smoke by up to 6.2%, and the average number of cigarettes that they smoke per day by up to 6.2% when they gain legal access to tobacco products. However, results from the nonparametric RD models show that the MLTPA is not a significant determinant of smoking behavior among young adults. This finding is in line with the results from the most flexible parametric models that contain control variables and fixed individual and year effects and are estimated using a quadratic polynomial of age which is fully interacted with a dummy variable indicating an age greater than or equal to the MLTPA. Finally, we also test whether our results from the most flexible parametric model that contain a quadratic polynomial of age and its interaction with the treatment term are robust to the selection of alternative age bandwidths. We estimate this model for each outcome variable using alternative bandwidths that range from 1 month to 1 year and report our findings in Figure S4 (Supporting Information) along with the main results that are generated using an age bandwidth of 2 years. Regardless of the selection of bandwidth, the estimates from these alternative specifications are quite similar to the main results and the impact of the MLTPA law on alternative indicators of smoking behavior remains statistically insignif- icant. Furthermore, 95% confidence intervals of estimates from models that are estimated using smaller bandwidths are relatively large since for these specifications, the sample size is relatively small.

ERTAN YÖRÜK & YÖRÜK: TOBACCO PURCHASE AGE LAWS 427

V. CONCLUSION

In this paper, we investigate the effect of the MLTPA on smoking behavior among young adults using a confidential version of the NLSY97, which contains information on respondents’ state of residence and exact date of birth. This information is unique and enabled us to clearly identify the treatment and control groups. Although there has been a considerable amount of research on the effect of the MLTPA and youth access laws on smoking and smoking- related outcomes, existing studies have two major limitations. First, although the decision to adopt a higher MLTPA might be endogenous, most of the existing studies have made use of the changes in the MLTPA laws that occurred in the 1980s and early 1990s at the state level. Second, none of the existing studies investigate the effect of the MLTPA laws on smoking participation and smoking-related outcomes at the national level using a RD design.

Using a RD approach, we find that the MLDA laws have a moderate and economically meaning- ful impact on smoking habits of young adults. In particular, we find that the MLTPA is associated with around 1.9 – 2.9 percentage point increase in the probability of smoking, and a 3.7% – 6.5% increase in the average number of cigarettes smoked per day. Although these estimates are statistically significant under certain specifica- tions, they are statistically insignificant under the nonparametric and relatively flexible para- metric specifications. Furthermore, we show that the effects of alternative MLTPAs such as 18 and 19 on smoking behavior are also statistically insignificant. However, our findings suggest that granting legal access to cigarettes and tobacco products at the MLTPA leads to an increase in several indicators of smoking participation for those who reported to have smoked before. We show that the probability of smoking among this group tends to increase up to 5 percentage points at the MLTPA. We also find that those who have smoked at least once since the DLI tend to increase the number of days that they smoke per month by 24.7% at the MLTPA cutoff.

In general, our results are robust under several alternative model specifications and imply that policies that are designed to restrict youth access to tobacco may be effective in reducing smok- ing participation among certain groups of young adults. Although our results have important pol- icy implications, they should also be interpreted with caution. By definition, the RD approach

used in this paper has a very good internal but limited external validity. Hence, our results hold for those who are around the MLTPA cutoff, but cannot be generalized to the whole population of young adults. They also do not allow us to mea- sure the long-term effects of the MLTPA laws.

MLTPA laws are also quite similar to the mini- mum legal drinking age (MLDA) laws since both policies restrict access based on a predetermined age cutoff. In contrast to the relatively limited literature on the effectiveness of MLTPA laws, several recent studies employed RD type models to investigate the effect of the MLDA of 21 on alcohol consumption and alcohol consumption- related outcomes.25 The findings from these stud- ies show that once young adults gain legal access to alcohol at age 21, they tend to significantly increase their alcohol consumption. The effects of the MLDA laws on other alcohol-related out- comes are relatively mixed. The existing stud- ies find that gaining legal access to alcohol at the MLDA is associated with deterioration in academic performance and increased mortality rate due to alcohol-related accidents. However, the effect of the MLDA law on risky behav- iors, smoking, and marijuana use is insignificant. Although our paper is the first to test the rela- tionship between the MLTPA and smoking at the national level using a RD design, similar to the MLDA laws, MLTPA laws may also have neg- ative spillover effects on other smoking-related outcomes such as alcohol consumption, obe- sity, psychological well-being, and health status. Therefore, further research is needed to inves- tigate the effects of the MLTPA law on other smoking-related outcomes. This calls for detailed survey data on smoking and smoking-related out- comes for young adults.

REFERENCES

Ahmad, S., and J. Billimek. “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.” Health Policy, 80(3), 2007, 378 – 91.

Anger, S., M. Kvasnicka, and T. Siedler. “One Last Puff? Public Smoking Bans and Smoking Behavior.” Journal of Health Economics, 30(3), 2011, 591 – 601.

Auld, M. C. “Smoking, Drinking, and Income.” Journal of Human Resources, 40(2), 2005, 505 – 18.

Bauer, T., S. Göhlmann, and M. Sinning. “Gender Differences in Smoking Behavior.” Health Economics, 16(9), 2007, 895 – 909.

25. See, for instance, Carpenter and Dobkin (2009), Ertan Yörük and Yörük (2012, 2015), Carrell, Hoekstra, and West (2011), Carpenter and Dobkin (2011), and Yörük and Ertan Yörük (2011, 2013).

428 CONTEMPORARY ECONOMIC POLICY

Baum, C. L. “The Effects of Cigarette Costs on BMI and Obesity.” Health Economics, 18(1), 2009, 3 – 19.

Cameron, A. C., and P. K. Trivedi. Microeconometrics: Meth- ods and Applications. New York: Cambridge University Press, 2005.

Carpenter, C., and P. J. Cook. “Cigarette Taxes and Youth Smoking: New Evidence from National, State, and Local Youth Risk Behavior Surveys.” Journal of Health Economics, 27(2), 2008, 287 – 99.

Carpenter, C., and C. Dobkin. “The Effect of Alcohol Con- sumption on Mortality: Regression Discontinuity Evi- dence from the Minimum Drinking Age.” American Economic Journal: Applied Economics, 1(1), 2009, 164 – 82.

. “The Minimum Legal Drinking Age and Public Health.” Journal of Economic Perspectives, 25(2), 2011, 133 – 56.

Carpenter, C., S. Postolek, and C. Warman. “Public-Place Smoking Laws and Exposure to Environmental Tobacco Smoke (ETS).” American Economic Journal: Economic Policy, 3(3), 2011, 35 – 61.

Carrell, S. E., M. Hoekstra, and J. E. West. “Does Drinking Impair College Performance? Evidence from a Regres- sion Discontinuity Approach.” Journal of Public Eco- nomics, 95(1), 2011, 54 – 62.

Chaloupka, F., and M. Grossman. “Price, Tobacco Control Policies, and Youth Smoking.” NBER Working Paper No. 5740, National Bureau of Economic Research, Cambridge, MA, 1996.

Chaloupka, F., and R. L. Pacula. “An Examination of Gen- der and Race Differences in Youth Smoking Respon- siveness to Price and Tobacco Control Policies.” NBER Working Paper, No. 6541, National Bureau of Economic Research, Cambridge, MA, 1998.

Cheng, M. Y., J. Fan, and J. S. Marron. “On Automatic Boundary Correction.” Annals of Statistics, 25(4), 1997, 1691 – 708.

Courtemanche, C. “Rising Cigarette Prices and Rising Obe- sity: Coincidence or Unintended Consequence?” Jour- nal of Health Economics, 28(4), 2009, 781 – 97.

DeCicca, P., and L. McLeod. “Cigarette Taxes and Older Adult Smoking: Evidence from Recent Large Tax Increases.” Journal of Health Economics, 27(4), 2008, 918 – 29.

DeCicca, P., D. Kenkel, and A. Mathios. “Putting Out the Fires: Will Higher Taxes Reduce the Onset of Youth Smoking?” Journal of Political Economy, 110(1), 2002, 144 – 69.

DeCicca, P., D. Kenkel, A. Mathios, Y. J. Shin, and J. Y. Lim. “Youth Smoking, Cigarette Prices, and Anti-Smoking Sentiment.” Health Economics, 17(6), 2008, 733 – 49.

DiFranza, J. R., J. A. Savageau, and K. E. Fletcher. “En- forcement of Underage Sales Laws as a Predictor of Daily Smoking among Adolescents — A National Study.” BMC Public Health, 9(1), 2009, 107.

DuMouchel, W. H., and G. J. Duncan. “Using Sample Survey Weights in Multiple Regression Analyses of Stratified Samples.” Journal of the American Statistical Associa- tion, 78(383), 1983, 535 – 43.

Ertan Yörük, C., and B. K. Yörük. “The Impact of Drinking on Psychological Well-Being: Evidence from Minimum Drinking Age Laws in the United States.” Social Sci- ence and Medicine, 75(10), 2012, 1844 – 54.

. “Alcohol Consumption and Risky Sexual Behavior among Young Adults: Evidence from Minimum Legal Drinking Age Laws.” Journal of Population Economics, 28(1), 2015, 133 – 57.

Goel, R. K., and M. J. Morey. “The Interdependence of Cigarette and Liquor Demand.” Southern Economic Journal, 62(2), 1995, 441 – 59.

Gruber, J., and M. Frakes. “Does Falling Smoking Lead to Rising Obesity?” Journal of Health Economics, 25(2), 2006, 183 – 87.

Gruber, J., and J. Zinman “Youth Smoking in the United States: Evidence and Implications,” in Risky Behavior Among Youths: An Economic Analysis, edited by J. Gruber. Chicago: University of Chicago Press, 2001, 69 – 120.

Hahn, J., P. Todd, and W. van der Klaauw. “Identification and Estimation of Treatment Effects with a Regression- Discontinuity Design.” Econometrica, 69(1), 2001, 201 – 9.

Hersch, J. “Teen Smoking Behavior and the Regulatory Envi- ronment.” Duke Law Journal, 47(6), 1998, 1143 – 70.

Imbens, G., and K. Kalyanaraman. “Optimal Bandwidth Choice for the Regression Discontinuity Estimator.” Review of Economic Studies, 79(3), 2012, 933 – 59.

Imbens, G., and T. Lemieux. “Regression Discontinuity Designs: A Guide to Practice.” Journal of Economet- rics, 142(2), 2008, 615 – 35.

Lee, D. S., and T. Lemieux. “Regression Discontinuity Designs in Economics.” NBER Working Paper No. 02138, National Bureau of Economic Research, Cambridge, MA, 2009.

Leigh, J. P., and M. C. Berger. “Effects of Smoking and Being Overweight on Current Earnings.” American Journal of Preventive Medicine, 5(1), 1989, 8 – 14.

Levine, P. B., T. A. Gustafson, and A. D. Velenchik. “More Bad News for Smokers? The Effects of Cigarette Smok- ing on Wages.” Industrial and Labor Relations Review, 50(3), 1997, 493 – 509.

Liu, F., N. Zhang, K. W. Cheng, and H. Wang. “Reduced Smoking and Rising Obesity: Does Smoking Ban in the Workplace Matter?” Economics Letters, 108(3), 2010, 249 – 52.

Malamud, O., and C. Pop-Eleches. “Home Computer Use and the Development of Human Capital.” Quarterly Journal of Economics, 126(2), 2011, 987 – 1027.

Marynak, K., C. B. Holmes, B. A. King, G. Promoff, R. Bun- nell, and T. McAfee. “State Laws Prohibiting Sales to Minors and Indoor Use of Electronic Nicotine Delivery Systems — United States, November 2014.” Morbidity and Mortality Weekly Report, 63(49), 2014, 1145 – 50.

Moore, W., S. Pedlow, P. Krishnamurty, and K. Wolter. National Longitudinal Survey of Youth 1997 (NLSY97) Technical Sampling Report. Chicago: NORC, 2000.

Porter, J. “Estimation in the Regression Discontinuity Model.” Unpublished Manuscript, Harvard University, Department of Economics, 2003.

U.S. Department of Health & Human Services (USDHHS). “The Health Consequences of Smoking — 50 Years of Progress: A Report of the Surgeon General.” Rockville, MD: USDHHS, 2014.

Yan, J. “The Effects of a Minimum Cigarette Purchase Age of 21 on Prenatal Smoking and Infant Health.” Eastern Economic Journal, 40(3), 2014, 289 – 308.

Yörük, B. K. “Can Technology Help to Reduce Underage Drinking? Evidence from the False ID Laws with Scan- ner Provision.” Journal of Health Economics, 36(1), 2014, 33 – 46.

Yörük, B. K., and C. Ertan Yörük. “The Impact of Minimum Legal Drinking Age Laws on Alcohol Consumption, Smoking, and Marijuana Use: Evidence from a Regres- sion Discontinuity Design Using Exact Date of Birth.” Journal of Health Economics, 30(4), 2011, 740 – 52.

. “The Impact of Minimum Legal Drinking Age Laws on Alcohol Consumption, Smoking, and Marijuana Use Revisited.” Journal of Health Economics, 32(2), 2013, 477 – 79.

ERTAN YÖRÜK & YÖRÜK: TOBACCO PURCHASE AGE LAWS 429

SUPPORTING INFORMATION

Additional Supporting Information may be found in the online version of this article:

Table S1. The Effect of the MLTPA on Smoking: Alter- native Models

Figure S1. Distribution of the interview day, birth day, and running variable. Notes: In panels A and B, number of observations for each possible birth and interview day are plotted. In panel C, number of observations for each day, 30 day before and after the MLTPA is plotted

Figure S2. Trends in selected observable covariates before and after the MLTPA. Notes: Mean of the observable covariates for 30 day intervals are plotted. The solid lines are a second-order polynomial fitted on individual observations on either side of the MLTPA as reported in Table 2

Figure S3. Predicted smoking trends before and after the MLTPA. Notes: Means of the outcome variables for 30 day intervals are plotted. The solid lines are a second-order polynomial fitted on individual observations on either side of the MLTPA cutoff without any control variables as reported in specifications 1, 4, 7, and 10 of Table 3

Figure S4. The effect of the MLTPA on smoking: alter- native age bandwidths. Notes: Coefficient estimates and 95% confidence intervals of the discontinuity at the MLTPA from models that are estimated using alternative age bandwidths are plotted for each outcome variable. All models are esti- mated using a quadratic polynomial of the running variable and its interaction with the treatment term. All models con- tain birthday celebration dummy, year fixed effects, and set of control variables as discussed in the text

Copyright of Contemporary Economic Policy is the property of Wiley-Blackwell and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.