1
The Effect of E-Cigarette Purchasing Age Restrictions on
Cigarette and Alcohol Use by Adolescents
Bachelor Thesis in Economics
University of Amsterdam
Faculty of Economics and Business
Author: Bud Schiphorst
Student number: 10763945
Date: January 29, 2017
Supervisor: Vadim Nelidov
1
2 Statement of Originality:
This document is written by Student Bud Schiphorst who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original
and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of
3 Abstract:
As of 30th of November 2014, 40 states had implemented e-cigarette purchasing age restrictions.
Because the bans serve to reduce public harm, it is important that potential effects on the usage of other harmful substances are analysed. This thesis examines the causal impact of e-cigarette
purchasing age restrictions on cigarette and alcohol use by adolescents using a fixed effects regression analysis. Implementation of an e-cigarette purchasing age restriction was estimated to increase recent smoking among adolescents by 0.829 percent point. This estimate is consistent with previous
literature, but was statistically insignificant. The estimated effects on regular and heavy cigarette use among adolescents were also similar to estimates in previous literature. It was however shown that, due to violation of an underlying assumption of the fixed effects model, claims about causality cannot be made. Implementation of an e-cigarette purchasing age restriction was estimated to cause a 1.029 percent point decrease in recent alcohol use by adolescents.
4 Introduction:
Electronic nicotine delivery systems (ENDS), including electronic cigarettes (e-cigarettes) and other devices such as electronic hookahs, electronic cigars, and vape pens, are battery-powered devices capable of delivering aerosolized nicotine and additives to the user (Marynak, et al., 2014, p. 1145). According to Riker, Lee, Darville and Hahn (2012, p. 160) ENDS were first introduced into American markets in 2007 and rapidly increased in use. They state that in December 2010 the US Court of Appeals decided that ENDS can be regulated as “tobacco products”. Absent a federal ban, states began implementing bans on sales of ENDS to minors. The first state to implement such a ban was New Jersey on the 12th of March 2010, and as of the 30th of November 2014, 40 states had
implemented such bans (Marynak, et al., 2014, p. 1148). Because the bans serve to reduce public harm, it is important that potential effects on the usage of other harmful substances (e.g. cigarettes and alcohol) are analysed.
Friedman (2015, p. 300) states that although several studies have examined the smoking-vaping2 relationship, potential confounders limit causal interpretation. Using data from the United
States, she found that implementing age restrictions resulted in a 0.9 percentage point increase in the adolescent smoking rate, relative to states without such age restrictions. She claims that the empirical findings in her paper provide the first causal evidence that e-cigarette access reduces teen smoking. Friedman used data from the National Survey on Drug Use and Health (NSDUH), but only up to 2013. Because 16 states implemented age restrictions in 2014 (Marynak, et al., 2014, p. 1148), it might be interesting to analyse whether findings using data up to 2015 are consistent with previous findings.
Previous studies have found a complementary relationship between cigarette and alcohol use (e.g., Dee, 1999). Hershberger, Karyadi, VanderVeen and Cyders (2016, p. 14) state that electronic cigarettes were created to approximate the look, feel and experience of using a cigarette. Because cigarette and alcohol use co-occur, they hypothesized that electronic cigarette use and alcohol use also co-occur. They tested this hypothesis by conducting a survey to assess the participants’ perceived likelihood of using electronic cigarettes and alcohol together. They found that electronic cigarette use is associated with higher problematic alcohol use. It might be interesting to analyse whether this perceived likelihood of co-use is consistent with the actual consumption of alcohol.
In this thesis, the following research question will be examined: What is the effect of e-cigarette purchasing age restrictions on e-cigarette and alcohol use by adolescents? This question will be examined by matching data on e-cigarette age restrictions with data on adolescent smoking and drinking rates from 2009 to 2015, provided by the Youth Behavior Risk Surveillance System (YRBSS). The effects will be analysed using fixed effects regression models, an approach similar to Friedman’s study. This thesis will contribute to the existing literature by using more recent data and
2Using ENDS is referred to as “vaping” opposed to “smoking”, throughout this thesis “vaping”, “electronic cigarettes” and
5 by also analysing the relationship between e-cigarettes and alcohol use. On top of that, this thesis will test the underlying assumptions of the fixed effect regressions, which will help to verify whether applying the method is actually statistically viable.
The next chapter will consist of a literature review, which will provide an overview of the relevant existing scientific literature. It will also verify the relevance of the research question and provide more insight on how this thesis will contribute to the existing scientific literature. The third chapter will discuss the methodology of this thesis and provide an overview of what data has been used. The results of the analysis and their validity will be discussed in the fourth chapter. The final chapter will give a conclusive answer to the research question and discuss any potential limitations of the thesis.
Literature review:
E-cigarettes and conventional cigarette use:
Because e-cigarettes are relatively new, there isn’t much known about the long-term health effects of vaping. While e-cigarettes are not without health risks, Farsalinos and Polosa (2014, p. 1) state that currently available evidence indicates that vaping is less harmful than smoking. That might be one of the reasons why e-cigarettes are thought of as a potential smoking cessation aid. Whether e-cigarettes are effective as a smoking cessation aid depends on whether e-cigarettes are perceived as substitutes by users.
Huang, Tauras and Chaloupka (2014) examined the relationship between e-cigarette
consumption and cigarette prices by estimating the cross-price elasticity using data from a retail store scanner in the US (the Nielsen Company). They did not find any consistent and statistically
significant relationship. They did however find a statistically significant estimate of the own-price elasticity of e-cigarettes of -1.9, which implies that e-cigarettes are very responsive to own-price changes. Grace, Kivell and Laugesen (2014) estimated the cross-price elasticity between e-cigarettes and cigarettes using experimental data. They estimated a cross-price elasticity of 0.16, which
according to them suggests that e-cigarettes are potentially substitutes for cigarettes and that their availability will reduce tobacco consumption. Existing literature on the cross-price elasticity between cigarettes and cigarettes is scarce, possibly because there is a lack of freely available data on e-cigarette prices and e-e-cigarette consumption. This thesis will avoid the use of price data and will instead examine the effect of increasing non-monetary costs (implementing purchasing age restrictions increases non-monetary costs of e-cigarette use by minors).
Pesko, Hughes and Faisal (2016) and Friedman (2015) found a statistically significant negative relationship between e-cigarette accessibility and cigarette use, also indicating that they are perceived as substitutes. Since these two studies have a very similar research method this thesis, they will be discussed in more detail in a later paragraph.
6 E-cigarette and alcohol use:
The relationship between alcohol and cigarette use has been well established in previous literature according to Hershberger et al. (2016, p. 14). For example, Dee (1999) found that higher cigarette taxes and reductions in teen smoking are associated with a lower prevalence of teen drinking, which indicates that cigarettes and alcohol are complements. Those findings are consistent with later literature (e.g., Cameron & Williams, 2001; Bask & Melkersson, 2004). From the relationship between conventional cigarettes and alcohol, rises the following question: “Is there also such a relationship between e-cigarettes and alcohol use?”.
Hershberger et al. (2016, p. 14) state that although recent research has investigated e-cigarette perceptions and use in substance dependent populations, little research has examined how e-cigarettes influence other addictive behaviours in adults, particularly alcohol use. Hershberger et al. (2016) conducted a survey to assess the perceived likelihood of co-use of e-cigarettes and alcohol among participants. They found that participants that identified as e-cigarettes users, reported significantly higher levels of alcohol use compared to non e-cigarette users. The study by Hershberger et al. seems to be the only published study that examined the relationship between e-cigarettes and alcohol use at this moment.
This thesis will contribute to the existing literature by examining whether a decrease in accessibility of e-cigarettes has a causal effect on alcohol use among adolescents.
The effects of bans on e-cigarette sales to minors.
Friedman (2015) examined the casual effect of e-cigarette accessibility on conventional cigarette use by adolescents, by using data from the National Survey on Drug Use and Health (NSDUH). She first analysed data on adolescent smoking rates from before 2010 (before any purchasing age restrictions were implemented). She found no statistically significant difference in the trends of states that would later implement an age restriction on the sales of e-cigarettes and of those what would not. To
estimate the effect of the bans, Friedman conducted fixed effects regressions with the smoking rate of 12 to 17 year olds in a certain two-year period as the outcome variable. As the regressor of interest she used a variable indicating whether a ban was effective during the period. She allowed for fixed state and year effects and included variables to control for demographics and smoking policies. By adding the smoking rates among 18 to 25 year olds she also controlled for policies that affect teen and young adult smoking for which state level data are unavailable. She estimated an effect of a 0.9 percentage point rise in adolescent smoking due to implementing e-cigarette purchasing age
restrictions. Friedman also did two falsification tests and a placebo test. For the first falsification test she added a regressor which indicated whether a ban would be active in the next period, she argues that this test verifies that the estimated effect of the ban isn’t driven by a time varying characteristic common to the states that are about to enact an age restriction (2015, p. 304). The second falsification test examined whether the bans on e-cigarette sales had an effect on smoking among adults, which
7 according to Friedman would implicate a driver other than the ban itself. The placebo test examined whether randomly assigned placebo bans would have a significant effect. The results of all three tests were consistent with the validity of her analysis. Friedman states that her paper has several
limitations, the paper for example doesn’t address the complementarities of e-cigarettes with other risky behaviours (e.g. alcohol consumption). She also adds that since the e-cigarette market is still quite young, her paper’s analyses may not reflect relationships at equilibrium.
Pesko et al. (2016) conducted a study similar to the study by Friedman. They examined the effect of e-cigarette purchasing age restrictions on other substance use, using data from the Youth Risk Behavior Surveillance System (YRBSS) from 2007 to 2013. Besides the effect on cigarette use, they also examined the effect on cigar, smokeless tobacco and marijuana use among adolescents. They used a fixed state and year effect regression model, while controlling for racial/ethnic composition of adolescents, cigarette excise taxes and indoor cigarette laws. They also examined association of e-cigarette purchasing age restrictions on outcomes 2 years before a restriction was in place. Pesko et al. (2016, p. 208) argue that if significant associations are found before the policy was implemented, that may indicate that restrictions were endogenously enacted in states with changing cigarette use relative to non-implementing states. Pesko et al. (2016, p. 208) also argue that if significant associations are not found in the period prior, then any associations found in the period in which the policy was implemented exhibit evidence of causality. They also examined the association that restrictions have in the 2 years after implementation to observe any lagged effects. They found that e-cigarette purchasing age restrictions are associated with an increase in regular and heavy adolescent cigarette use. Pesko et al. (2016, p. 211) state that their finding of a 0.8 increase in regular cigarette users in response to e-cigarette purchasing age restrictions is consistent with the study by Friedman. They found an association with an increase in recent and casual cigarette use as well, but they also found that recent and casual cigarette use was increasing the period before the purchase age restriction was implemented, preventing them from making claims of causality. They did not find evidence of e-cigarette purchasing age restrictions being associated with recent cigar use, recent smokeless tobacco use, or recent marijuana use.
Summary:
While several studies have examined the relationship between e-cigarettes and cigarettes, there is still a limited amount of causal empirical evidence. This is most likely due to the lack of freely available data on e-cigarette prices and consumption. Two previous studies provided causal empirical evidence for the relationship between e-cigarettes accessibility and cigarette use among adolescents (Friedman, 2015; Pesko et al., 2016). They only used data up to 2013, while 16 states implemented age
restrictions in 2014 (Marynak, et al., 2014, p. 1148). There is also a very limited amount of literature available on the relationship between e-cigarettes and alcohol use.
8 This thesis will examine the validity of previous findings on the relationship between e-cigarettes and cigarette use among adolescents. This will be done by trying to replicate the results using more recent data and testing the underlying assumptions of the fixed effects model. This thesis will also be one of the first empirical studies to examine the relationship between e-cigarettes and alcohol use among adolescents.
Methods
The following research question will be examined in this thesis: What is the effect of e-cigarette purchasing age restrictions on cigarette and alcohol use by adolescents? This research question actually consists of two sub-questions: What is the effect of the purchasing age restrictions on
cigarette use by adolescents and what is the effect of the purchasing age restrictions on alcohol use by adolescents? These two sub-questions will be analysed separately in this thesis.
Models:
The two sub-questions will be analysed by using fixed effects regression models similar to the model in Friedman’s study. The models used to analyse the effect on cigarette and alcohol consumption by adolescents are as follows:
(1) 𝐴𝑑𝑜𝑙𝑒𝑠𝑐𝑒𝑛𝑡𝑠𝑚𝑜𝑘𝑒𝑆𝑌 = 𝛼0+ 𝛼1𝐵𝑎𝑛𝑆𝑌 + 𝛼2𝐴𝑑𝑢𝑙𝑡𝑠𝑚𝑜𝑘𝑒𝑆𝑌+ 𝛼3𝐶𝑖𝑔𝑡𝑎𝑥𝑆𝑌+ 𝛼4𝐴𝑑𝑜𝑙𝑒𝑠𝑐𝑒𝑛𝑡𝑑𝑟𝑖𝑛𝑘𝑆𝑌+ 𝛼5𝑀𝑒𝑑𝑖𝑎𝑛𝑖𝑛𝑐𝑜𝑚𝑒𝑆𝑌+ 𝛼6𝑈𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡𝑆𝑌+ 𝜂𝑆𝑡𝑎𝑡𝑒𝑆+ 𝜆𝑌𝑒𝑎𝑟𝑌+ 𝜀𝑆𝑌 (2) 𝐴𝑑𝑜𝑙𝑒𝑠𝑐𝑒𝑛𝑡𝑑𝑟𝑖𝑛𝑘𝑆𝑌 = 𝛽0+ 𝛽1𝐵𝑎𝑛𝑆𝑌 + 𝛽2𝐴𝑑𝑢𝑙𝑡𝑑𝑟𝑖𝑛𝑘𝑆𝑌+ 𝛽3𝐴𝑙𝑐𝑜ℎ𝑜𝑙𝑡𝑎𝑥𝑆𝑌+ 𝛽4𝐴𝑑𝑜𝑙𝑒𝑠𝑐𝑒𝑛𝑡𝑠𝑚𝑜𝑘𝑒𝑆𝑌+ 𝛽5𝑀𝑒𝑑𝑖𝑎𝑛𝑖𝑛𝑐𝑜𝑚𝑒𝑆𝑌+ 𝛽6𝑈𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡𝑆𝑌+ 𝜃𝑆𝑡𝑎𝑡𝑒𝑆+ 𝜙𝑌𝑒𝑎𝑟𝑌+ 𝜐𝑆𝑌
𝐴𝑑𝑜𝑙𝑒𝑠𝑐𝑒𝑛𝑡𝑠𝑚𝑜𝑘𝑒𝑆𝑌 and 𝐴𝑑𝑜𝑙𝑒𝑠𝑐𝑒𝑛𝑡𝑑𝑟𝑖𝑛𝑘𝑆𝑌 represent the rate of adolescents in state S and year Y that smoked cigarettes or drank alcohol respectively. This thesis will, similarly to Pesko et al. (2016) use three measures for 𝐴𝑑𝑜𝑙𝑒𝑠𝑐𝑒𝑛𝑡𝑠𝑚𝑜𝑘𝑒𝑆𝑌: the recent, regular and heavy smoking rate3. Due to limited available data, 𝐴𝑑𝑜𝑙𝑒𝑠𝑐𝑒𝑛𝑡𝑑𝑟𝑖𝑛𝑘𝑆𝑌 will only be defined as the rate of adolescents that drank alcohol on at least one day during the 30 days before the survey.
𝐵𝑎𝑛𝑆𝑌 is a dummy variable indicating whether an age restriction on the sale of e-cigarettes was active at the end of the survey period. The coefficients 𝛼1 and 𝛽1 capture the casual effect of an e-cigarette purchasing age restriction on adolescent cigarette and alcohol use respectively.
3 The recent, regular and heavy smoking rate are defined as the rate of adolescents that smoked on at least one, twenty and
9 𝐴𝑑𝑢𝑙𝑡𝑠𝑚𝑜𝑘𝑒𝑆𝑌 and 𝐴𝑑𝑢𝑙𝑡𝑑𝑟𝑖𝑛𝑘𝑆𝑌 represent the rate of adults that recently smoked or drank alcohol respectively. Since consumption by adults shouldn’t be affected by the implementation of a
purchasing age restriction, these variables may be used as controls for other time-variant state-level factors that influence cigarette and alcohol consumption. Friedman (2015, p. 303) argues that
including the consumption rate of adults may control for further policies that impact consumption for which state level data are unavailable (e.g., advertising and anti-smoking campaigns).
𝐶𝑖𝑔𝑡𝑎𝑥𝑆𝑌 and 𝐴𝑙𝑐𝑜ℎ𝑜𝑙𝑡𝑎𝑥𝑆𝑌 represent inflation-adjusted total tax rates for cigarettes and alcohol respectively. Youth are more sensitive to price changes in cigarettes (Ding, 2003, p. 115) and alcohol (Chaloupka, Grossman, & Saffer, 2002) than adults. Therefore, the adult consumption trends do not truly capture the effect of changing tax rates. Including 𝐶𝑖𝑔𝑡𝑎𝑥𝑆𝑌 and 𝐴𝑙𝑐𝑜ℎ𝑜𝑙𝑡𝑎𝑥𝑆𝑌 in the regression model solves this problem.
𝑆𝑡𝑎𝑡𝑒𝑆 and 𝑌𝑒𝑎𝑟𝑌 represent state and year fixed effects respectively. According to Pesko et al. (2016, p. 208) including state and year fixed effects controls for state-level time-invariant
unobservable characteristics, and unobservable characteristics unique to a particular year. They argue that for example, anti-smoking sentiment and non-changing population characteristics unique to a particular state is controlled for by the inclusion of state fixed effects. They also state that he national introduction of ENDS over time is controlled for by year fixed effects.
The smoking and drinking rates of adolescents are highly correlated with each other and previous literature indicates that both may be affected by the accessibility of e-cigarettes. Not including the consumption rate of the other variable in the model will therefore introduce an omitted variable bias. To solve this 𝐴𝑑𝑜𝑙𝑒𝑠𝑐𝑒𝑛𝑡𝑑𝑟𝑖𝑛𝑘𝑆𝑌 is included in the regression model of
𝐴𝑑𝑜𝑙𝑒𝑠𝑐𝑒𝑛𝑡𝑠𝑚𝑜𝑘𝑒𝑆𝑌 and vice versa. These variables will also control for state-level changes in risky behaviour by adolescents across time.
𝑀𝑒𝑑𝑖𝑎𝑛𝑖𝑛𝑐𝑜𝑚𝑒𝑆𝑌 and 𝑈𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡𝑆𝑌 represent median income and unemployment rates respectively. These are included in both models as controls for time-variant state-level
demographics that may influence cigarette and alcohol use. Other demographics were also considered (e.g., population size and ethnic composition), but because these are mostly time-invariant and thus already mostly captured by the fixed state and year effects.
Finally, 𝜀𝑆𝑌 and 𝜐𝑆𝑌 represent stochastic error terms. Assumptions:
According to Stock and Watson (2014, p. 412) there are four important assumptions for the fixed effects regression model:
1. 𝑢𝑖𝑡 has conditional mean zero: E[𝑢𝑖𝑡|𝑋𝑖1, … , 𝑋𝑖𝑇, 𝜂𝑖] = 0, ∀𝑡
2. (𝑋𝑖1, … , 𝑋𝑖𝑇, 𝑢𝑖1 , … , 𝑢𝑖𝑇 ) are independent and identically distributed draws from their joint distribution.
10 4. There is no perfect multicollinearity.
Where 𝑢𝑖𝑡 represents the stochastic error in state i and year t, 𝑋𝑖𝑡 represents a vector of all the regressors and 𝜂𝑆 represents the fixed state effects.
According to Stock and Watson (2014, p. 411) assumption 1 implies that there is no omitted variable bias. They state that assumption 1 implies that the conditional mean of the error term is independent from all of the observations of the regressors, even the observations of the regressors in different time periods. They state that the second assumption is that the variables of one state are distributed identically to, but independently of, the variables of another state. Although assumption 2 holds that the variables are independent across states, it allows the variables to be correlated over time within a state.
Violation of assumption 1 to 3 will lead to inconsistent estimates and therefore make the fixed effects model inappropriate to use. Violation of assumption 4 will make it impossible to derive the fixed effect regression estimates.
Falsification tests:
Assumption 1 rules out feedback between the outcome variable and future variables. I.e. the following is assumed within fixed effect regression: E[𝑢𝑖𝑡|𝑋𝑖𝑡+1 ] = 0. Where 𝑢𝑖𝑡 represents the error term in state i and year t and 𝑋𝑖𝑡+1 represents a vector with all the regressors in state i and year t+1. However, one may for example argue that whether a state implements a purchasing age restriction is correlated with the cigarette and/or alcohol consumption by adolescents in the preceding period. Whether there is feedback will be tested by adding lead-regressors (𝑋𝑖𝑡+1) to the regression models 1 and 2 and then testing the joint significance of those lead-regressors with a partial F-test. Joint significance of the lead-regressors will indicate that there’s feedback between the dependent and independent variables, which means that assumption 1 doesn’t hold.
A test similar to Friedman’s second falsification test will also be conducted in this thesis. In her second falsification test she examined whether age restrictions impact cigarette consumption by adults. She argues that that would implicate a driver other than the age restriction itself (e.g., greater information about smoking’s risks). This will be tested in this thesis by running the following regression models (3) and (4) with adult consumption as the dependent variable and testing the significance of 𝐵𝑎𝑛𝑆𝑌.
(3) 𝐴𝑑𝑢𝑙𝑡𝑠𝑚𝑜𝑘𝑒𝑆𝑌 = 𝛾0+ 𝛾1𝐵𝑎𝑛𝑆𝑌 + 𝛾2𝐶𝑖𝑔𝑡𝑎𝑥𝑆𝑌+ 𝛾3𝐴𝑑𝑢𝑙𝑡𝑑𝑟𝑖𝑛𝑘𝑆𝑌+ 𝛾4𝑀𝑒𝑑𝑖𝑎𝑛𝑖𝑛𝑐𝑜𝑚𝑒𝑆𝑌+ 𝛾5𝑈𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡𝑆𝑌+ 𝜌𝑆𝑡𝑎𝑡𝑒𝑆+ 𝜎𝑌𝑒𝑎𝑟𝑌+ 𝜖𝑆𝑌
(4) 𝐴𝑑𝑢𝑙𝑡𝑑𝑟𝑖𝑛𝑘𝑆𝑌= 𝛿0+ 𝛿1𝐵𝑎𝑛𝑆𝑌 + 𝛿2𝐴𝑙𝑐𝑜ℎ𝑜𝑙𝑡𝑎𝑥𝑆𝑌+ 𝛿3𝐴𝑑𝑢𝑙𝑡𝑠𝑚𝑜𝑘𝑒𝑆𝑌+
𝛿4𝑀𝑒𝑑𝑖𝑎𝑛𝑖𝑛𝑐𝑜𝑚𝑒𝑆𝑌+ 𝛿5𝑈𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡𝑆𝑌+ 𝜓𝑆𝑡𝑎𝑡𝑒𝑆+ 𝜋𝑌𝑒𝑎𝑟𝑌+ 𝜉𝑆𝑌 Where 𝜖𝑆𝑌 and 𝜉𝑆𝑌 are stochastic error terms and the other variables as previously defined in regression 1 and 2.
11 Regular standard errors assume heteroskedasticity and no serial correlation of the regression errors. Whether there is autocorrelation of the error terms will be tested with the following regression: (Û𝑖𝑡− Û𝑖𝑡−1) = 𝜚(Û𝑖𝑡−1− Û𝑖𝑡−2) + 𝑒𝑖𝑡. Where Û𝑖𝑡 represents the residuals of regression model 1 or 2. If the estimate of 𝜚 is significantly different from -0.5, then that indicates that there is autocorrelation (Drukker, 2003, 169). Clustered standard errors will be used if significant autocorrelation is found. Stock and Watson (2014, p. 413) state that clustered standard errors allow for heteroskedasticity and for the regression errors to be autocorrelated within a state, but treat the error terms as uncorrelated across states. They also state that the clustered robust standard errors, in combination with assumption 2, provide valid standard errors.
Hypotheses:
𝛼1 and 𝛽1 correspond to the causal effect of age restrictions on e-cigarettes on the use of cigarettes and alcohol by adolescents respectively. The implementation an e-cigarette purchasing age restriction decreases the accessibility of e-cigarettes to adolescents. The decreased accessibility might lead adolescents to substitute vaping with smoking. On the other hand, the decreased accessibility might reduce gate-way effects from e-cigarettes to cigarettes. The decreased accessibility might also lead adolescents to substitute vaping with drinking alcohol. If e-cigarettes are perceived as complements to alcohol, the decreased accessibility might reduce alcohol use by adolescents. The ambiguity of the effects from a theoretical point is reflected in the following hypotheses:
Hypotheses for the effect of age restrictions on cigarette use by adolescents: 𝐻0: 𝛼1 = 0 𝐻1: 𝛼1≠ 0
Hypotheses for the effect of age restrictions on alcohol use by adolescents: 𝐻0: 𝛽1= 0 𝐻1: 𝛽1≠ 0
Data:
Data from the Youth Risk Behavior Surveillance System (YRBSS) will be used to provide measures for cigarette and alcohol consumption by adolescents. Brener et al. (2013, p. 1) state “YRBSS data are collected from multiple sources, including a national school-based survey conducted by Centers for Disease Control and Prevention (CDC) as well as school-based state, territorial, tribal, and large urban school district surveys conducted by education and health agencies”. They also state that each state survey uses a two-stage, cluster sample design to produce a representative sample of students in grades 9-124 in its jurisdiction. The national YRBSS is conducted during February to May of each
odd-numbered year (Brener, et al., 2013, p. 7)5. The YRBSS doesn’t provide complete information for
every state in every year. Complete information on adolescent cigarette and alcohol use from 2009 to 2015 is provided for 28 states.
4 Grades 9 to 12 correspond to the ages 14 to 18.
5 There is no available information on in which month exactly the data has been collected in each state. The end-point of the
12 Data on implemented e-cigarette purchasing age restrictions up to November 20146 is provided by
Marynak et al. (2014, p. 1148). Marynak et al. only provide data on state-wide bans, while some in some states age restrictions were implemented at a lower level. Following Pesko et al. (2016, p. 208), Massachusetts will be excluded from the analysis because it implemented age restrictions in most of its region at county level at different points in time.
Data on cigarette and alcohol use by adults is provided by the Behavioral Risk Factor
Surveillance System (BRFSS). The BRFSS consists of telephone surveys that collect state data about U.S. residents regarding their health-related risk behaviours. The rate of 18 to 24 year olds that currently smoke will be used as the measure for 𝐴𝑑𝑢𝑙𝑡𝑠𝑚𝑜𝑘𝑒𝑆𝑌. These rates shouldn’t be affected by the age restrictions, yet may still capture the effect of other policies that impact both teens and young adults according to Friedman (2015, p. 303). The rate of adults that recently drank alcohol7 will be
used as the measure for 𝐴𝑑𝑢𝑙𝑡𝑑𝑟𝑖𝑛𝑘𝑆𝑌. The BRFSS doesn’t provide complete information in every year on the recent drinking rate per age-group. Therefore, the recent drinking rate of all adults will be used, opposed to the recent drinking rate of just 18 to 24 year olds.
The State Tobacco Activities Tracking and Evaluation (STATE) system maintained by the CDC provides data on state-level cigarette excise taxes. Following Pesko et al. (2016, p. 208), federal excise taxes8 will be added to the state-level excise taxes. Chaloupka et al. (2002) state: “Most of the
studies employ the price of beer or the State excise tax on beer as a measure of the cost of alcohol because beer is the beverage of choice among Americans, particularly among youths”. In this thesis state-level beer taxes will be used as a measure for 𝐴𝑙𝑐𝑜ℎ𝑜𝑙𝑡𝑎𝑥𝑆𝑌. Data on state-level beer taxes is obtained from the Tax Foundation. Both the cigarette and alcohol taxes will be adjusted for inflation. Data on inflation rates is provided by the Bureau of Labor Statistics (BLS).
Data on unemployment rates per state from 2009 to 2015 is also provided by the Bureau of Labor Statistics (BLS). And finally, state-level data on median income per household is obtained from the Census Bureau.
Results:
Table 1. shows that cigarette and alcohol use by adolescents consistently decreased from 2009 to 2015 in the states with complete information. Recent cigarette use by adolescents declined from 18.92% to 11.42%, a decrease of 39.64%. Regular cigarette use by adolescents declined from 7.62% to 3.60%, a decrease of 52.76%. Heavy cigarette use by adolescents declined from 5.64% to 2.67%, a decrease of 52.66%. Recent alcohol use declined from 39.13% to 28.98%, a decrease of 25.94%.
6 Data on age restrictions between November 2014 and June 2015 was provided by Americans for Non-smokers’ Rights
[accessed on Jan 24, 2017].
7 The recently drinking rate again refers to drinking alcohol on at least one day in the last 30 days. 8 Federal excise taxes were $1.01 from 2009 to 2015.
13 Table 1. Summary statistics
2009 2011 2013 2015
Adolescents
Recent cigarette use 18.92% 17.28% 14.40% 11.42%
Regular cigarette use 7.62% 6.90% 5.18% 3.60%
Heavy cigarette use 5.64% 5.17% 3.72% 2.67%
Recent alcohol use 39.13% 36.27% 32.40% 28.98%
Adults
Recent alcohol use 51.73% 54.01% 52.03% 51.75%
Recent cigarette use 23.18% 24.80% 21.43% 17.42%
Taxes
Cigarette excise tax ($)* 2.26 2.37 2.34 2.34
Beer excise tax ($)* 0.32 0.31 0.34 0.34
Demographics
Unemployment rate 8.46% 8.08% 6.78% 5.12%
Median household income ($) 54,329 53,236 52,418 54,616
States that implemented a ban 0 1 6 22
(3.7%) (22.2%) (81.5%)
N 27 27 27 27
Notes: Observations are means from all states that have complete information from 2009 to 2015. *The excise taxes are
adjusted for inflation with 2007 as base year.
Fig. 1. Recent cigarette and alcohol rates among adolescents in states with and without e-cigarette purchasing age
restrictions before June 2013. Notes: ‘Recently used’ refers to consumption on at least one day in the 30 days preceding the survey. Data on all states has been used, also states without complete information. The vertical line represents the date of the first implementation of a purchasing age restriction in March 2010, in New Jersey.
14
Fig. 1. displays the trends in recent cigarette and alcohol use among adolescents in states with and without e-cigarette purchasing age restrictions before June 2013. The recent cigarette and alcohol use rates are on average higher in states that did not enact a purchasing age restriction before June 2013. The rates in states with and without a restriction seem to follow the same trends before the first restrictions were actually implemented. The last observation might imply that the rates of states with and without restrictions would still have followed the same trends in the absence of any purchasing age restrictions. The similar trends indicate that the differences in the rates is caused by time-invariant factors, which are controlled for in the regression models by the fixed state effects. Between 2011 and 2013 the rates of recent cigarette use seem to decline at the same rate in states with and without restrictions, while in the same period the rates of recent alcohol use seem to decline faster in states that implemented a purchasing age restriction. Most states that didn’t implement a purchasing age restriction before 2013 did implement a ban between June 2013 and June 20159. Therefore, not much
can be deduced from the trends about the period between 2013 and 2015.
Fig. 2. Regular and heavy cigarette use rates among adolescents in states with and without e-cigarette purchasing age
restrictions before June 2013. Notes: ‘Regularly used’ and ‘heavily used’ refer to consumption on at least 20 and 30 days respectively in the 30 days preceding the survey. Data on all states has been used, also states without complete information. The vertical line represents the date of the first implementation of a purchasing age restriction in March 2010, in New Jersey.
Fig. 2. displays the trends in regular and heavy cigarette use among adolescents in states with and without e-cigarette purchasing age restrictions before June 2013. Again, the states without a purchasing age restriction before June 2013 seem to have higher cigarette use among adolescents. Focussing on the period before the first implementation of a purchasing age restriction, the smoking rates don’t seem to follow the same trends in states that implemented such a restriction before June 2013 and those that didn’t. This indicates that the difference between states that implemented a purchasing age restriction before June 2013 and those that didn’t are caused by time-variant factors. Fixed state and year effects do not control for these state-specific time-variant factors, but adult
15 smoking rates and cigarette taxes might. It is important to verify that the purchasing age restrictions were not endogenously enacted in states with changing smoking rates relative to states that didn’t implement age restrictions.
Table 210 displays the results of regression 1 and falsification regression 3. The test for
autocorrelation indicated that there might be autocorrelation of the regression errors of model 1 and 2 (F1,100=1.79, p=0.1836 and F1,100=2.86, p=0.094). Therefore, clustered robust standard errors were
used in the regressions in this thesis. The estimated coefficient of 𝐵𝑎𝑛𝑆𝑌 in the regression on the recent smoking rate of adolescents of 0.829 is consistent with the estimate by Friedman (2015) of 0.9. The estimate is however statistically insignificant (p= 0.108>0.1). The estimates of the coefficients of 𝐵𝑎𝑛𝑆𝑌 in the regressions on regular and heavy cigarette use are 0.613 and 0.527 respectively. These estimates are both statistically significant at 1% and close to the estimates of Pesko et al. (2016) of 0.8 and 0.65. These results imply that the implementation of an e-cigarette purchasing age restriction has no significant effect on the recent smoking rate among adolescents, but increases the regular and heavy smoking rates by 0.613 and 0.527 percentage point respectively. The estimated effect of a purchasing age restriction on adult cigarette consumption is, as was expected, highly insignificant. This result helps verify that the purchasing age restrictions have a causal effect on the cigarette consumption by adolescents.
Table 3 Displays the results of regression 2 and falsification regression 4. The estimated effect of a purchasing age restriction on the consumption of alcohol by adolescents is -1.023 and is statistically significant at 10%. This result implies that the implementation of an e-cigarette
purchasing age restriction decreases alcohol consumption among adolescents by 1.023 percent point. The estimated effect of a purchasing age restriction on the consumption of alcohol by adults is, as was expected, highly insignificant. Which helps verify that the purchasing age restrictions have a causal effect on the alcohol consumption by adolescents.
Table 4 displays the results of the regressions for testing whether the adolescent consumption rates are independent of future observations of the regressors. The method for testing this was
discussed in the section with falsification tests. The results indicate that there is no significant feedback between the recent smoking rate and the independent variables in regression model 1
(F6,26=1.19, p=0.3430>0.1). The results do however indicate significant feedback with the regular
smoking rate (F6,26=2.72, p=0.0346< 0.1) and the heavy smoking rate (F6,26=7.16, p = 0.0001<0.1).
This evidence for feedback between the regular and heavy smoking rates and the independent variables, indicate that assumption 1 doesn’t hold. This violation of the made assumption hurts the validity of conclusions made using the fixed effects model. For both the regular and heavy smoking rates is the estimate of 𝐵𝑎𝑛𝑆𝑌+1 significant at 1%. This indicates that whether a purchasing age restriction was implemented in a certain period, was dependent on the regular and heavy smoking
16 rates of the preceding period. This makes claims of a causal effect of age restrictions on regular and heavy smoking rates invalid.
The lead-regressors (𝑋𝑖𝑡+1) in the regression on recent alcohol use by adolescents are not jointly significant (F6,26=1.75, p=0.1488> 0.1). This indicates that there is no significant evidence that
assumption 1 is violated, but the low p-value may still raise some concern. The results indicate that especially 𝐴𝑑𝑢𝑙𝑡𝑑𝑟𝑖𝑛𝑘𝑆𝑌+1 and 𝑈𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡𝑆𝑌+1 have a significant relationship with the recent alcohol use by adolescents. Including these two variables in regression 2 only slightly changes the estimated coefficient of 𝐵𝑎𝑛𝑆𝑌 (from -1.023 to -0.869), which indicates that any potential bias shouldn’t have a large effect on the estimate. The estimate of the coefficient of 𝐵𝑎𝑛𝑆𝑌+1 is very small and highly insignificant, which indicates that whether a purchasing age restriction was implemented, was independent of the recent alcohol use among adolescents in the preceding period. This last observation strengthens the claim that the implementation of e-cigarette purchasing age restrictions have a causal effect on alcohol use by adolescents.
Conclusion:
This thesis analysed the effect of implementing e-cigarette purchasing age restrictions on cigarette and alcohol use among adolescents. It was estimated that implementing such an age restriction increases recent cigarette use by 0.829 percent point, which is consistent with the estimate by Friedman (2015) of 0.9. However, the estimate in the thesis was statistically insignificant at 10%. The implementation of a purchasing age restriction was estimated to decrease recent alcohol use among adolescents by 1.023 percent point. The estimated effects on regular and heavy cigarette use (an increase of 0.613 and 0.527 percent point respectively) are consistent with previous findings by Pesko et al. (2016).
The validity of these estimates and their interpretation as causal effects depend on the underlying assumptions of the fixed effects model. A significant relationship between the
implementation of purchasing age restrictions and regular and heavy smoking rates in the preceding period was found, which indicates violation of an important assumption of the model. The estimates of the effects on regular and heavy cigarette use are therefore considered invalid and claims about causality cannot be made. No significant relationship was found between the implementation of purchasing age restrictions and recent alcohol use rates in the preceding period, which is consistent with the assumptions made. However, it is also assumed that alcohol use among adolescents is independent with alcohol use among adolescents in other states. It may be the case that alcohol use is affected by alcohol use in nearby states, which would hurt the validity of the estimated effect.
Whether alcohol use is independent of alcohol use in nearby states may be a topic for further research. Friedman (2015) argued that the implementation of e-cigarette purchasing age restrictions can be seen as an increase in the non-monetary costs of e-cigarette use among minors. This, combined with the found effect of the age restrictions on alcohol use, indicate a complementary relationship between e-cigarettes and alcohol use. This relationship should be examined further when more data on prices of e-cigarettes becomes available.
17 Bibliography
Americans for Non-smokers’ Rights Foundation (2017). Accessed Jan 24 2017, at http://no-smoke.org/aboutus.php?id=436
Bask, M., & Melkersson, M. (2004). Rationally addicted to drinking and smoking? Applied Economics,
36, 373-381.
Brener, N. D., Kann, L., Shanklin, S., Kinchen, S., Eaton, D. K., Hawkens, J., & Flint, K. H. (2013). Methodology of the Youth Risk Behavior Surveillance System — 2013. MMWR Recomm Rep,
62(1), 1-20.
Bureau of Labor Statistics (2016). (Accessed Jan 24 2017, at https://www.bls.gov/.
Cameron, L., & Williams, J. (2001). Cannabis, alcohol and cigarettes: Substitutes or complements?
The Economic Record, 77(236), 19-34.
Census Bureau, U.S. (2017). Table H-8A. Median Household Income by State - 2 Year Average: Retrieved from (
http://www2.census.gov/programs-surveys/cps/tables/time-series/historical-income-households/h08.xls) (Jan 24 2017).
Centers for Disease Control and Prevention (2017). Youth Risk Behavior Surveillance
System (YRBSS). (Accessed Jan 24 2017, at http://www.cdc.gov/HealthyYouth/yrbs/. Centers for Disease Control and Prevention (2017). BRFSS Prevalence & Trends Data.
(Accessed Jan 24 2017, at https://www.cdc.gov/brfss/brfssprevalence/.
Centers for Disease Control and Prevention (2017). State Tobacco Activities Tracking and
Evaluation (STATE) System. (Accessed Jan 24 2017, at https://www.cdc.gov/statesystem/. Chaloupka, F. J., & Pacula, R. L. (1998). An examination of gender and race differences in youth
smoking responsiveness to price and tobacco control policies. National Bureau of Economic
Research, 373-377.
Chaloupka, F. J., Grossman, M., & Saffer, H. (2002). The Effects of Price on Alcohol Consumption and Alcohol-Related Problems. Alcohol research and health, 26(1), 22-34.
Dee, T. S. (1999). The complementarity of teen smoking and drinking. Journal of Health Economics,
18(6), 769-793.
Ding, A. (2003). Youth Are More Sensitive to Price Changes in Cigarettes than Adults. Yale journal of
Biology and Medicine, 76, 115-124.
Drukker, D. M. (2003). Testing for serial correlation in linear panel-data models. Stata Journal, 3(2), 168-177.
Farsalinos, K. E., & Polosa, R. (2014). Safety evaluation and risk assessment of electronic cigarettes as tobacco cigarette substitutes: a systematic review. Therapeutic Advances Drug Safety, 5(2), 67-86.
Friedman, A. S. (2015). How does electronic cigarette access affect adolescent smoking? Journal of
18 Grace, R. C., Kivell, B. M., & Laugesen, M. (2014). Estimating cross-price elasticity of e-cigarettes
using a simulated demand procedure. Nicotine & Tobacco Research, 592-598.
Haung, J., Tauras, J., & Chaloupka, F. J. (2014). The impact of price and tobacco control policies on.
Tobacco Control, 23(3).
Hershberger, A. R., Karyadi, K. A., VanderVeen, D. J., & Cyders, M. A. (2016). Combined expectancies of alcohol and e-cigarette use relate to higher alcohol use. Addictive Behaviors, 52, 13-21. Marynak, 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 WeeklyReport, 63(49), 1445-1550. Pesko, M. F., Hughes, J. M., & Faisal, F. S. (2016). The influence of electronic cigarette age purchasing
restrictions on adolescent tobacco and marijuana use. Preventive Medicine, 87, 207-212. Riker, C. A., Lee, K., Darville , A., & Hahn, E. J. (2012). E-Cigarettes: Promise or Peril? Nursing Clinics of
North America, 47(1), 159-171.
Stock, J. H., & Watson, M. W. (2014). Introduction to Econometrics. Harlow, England: Pearson. Tax Foundation (2017). (Accessed Jan 24 2017, at
19 Appendix:
Table 2. Fixed effects regression analysis on cigarette use from 2009 to 2015
Adolescents
Adults
Recent smoking Regular smoking Heavy smoking Recent smoking
𝐵𝑎𝑛𝑆𝑌
0.829
0.613***
0.527***
0.323
(0.498)
(0.179)
(0.160)
(0.950)
𝐴𝑑𝑢𝑙𝑡𝑠𝑚𝑜𝑘𝑒𝑆𝑌0.0433
0.0325
0.0292
(0.0562)
(0.0275)
(0.0263)
𝐶𝑖𝑔𝑡𝑎𝑥𝑆𝑌-0.126
0.364
0.425
-5.295***
(0.792)
(0.414)
(0.350)
(1.883)
𝑈𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡𝑆𝑌0.0567
-0.000524
-0.00327
-0.0492
(0.225)
(0.104)
(0.0994)
(0.317)
𝑀𝑒𝑑𝑖𝑎𝑛𝑖𝑛𝑐𝑜𝑚𝑒𝑆𝑌
3.59e-05
2.48e-05
-1.10e-05
0.000357**
(9.33e-05)
(5.08e-05)
(3.62e-05)
(0.000166)
𝐴𝑑𝑜𝑙𝑒𝑠𝑐𝑒𝑛𝑡𝑑𝑟𝑖𝑛𝑘𝑆𝑌
0.458***
0.178***
0.153***
(0.0891)
(0.0418)
(0.0358)
𝐴𝑑𝑢𝑙𝑡𝑑𝑟𝑖𝑛𝑘𝑆𝑌0.111
(0.166)
Constant
-1.970
-2.183
-1.328
10.49
(5.660)
(2.747)
(2.454)
(13.27)
Observations
108
108
108
108
R-squared
0.858
0.870
0.823
0.612
Number of states
27
27
27
27
State fixed effects
YES
YES
YES
YES
Year fixed effects
YES
YES
YES
YES
Notes: Robust standard errors in parentheses, standard errors are clustered per state. Recent use refers to
consumption on at least one day in the 30 days preceding the survey, regular use refers to at least 20 days and heavy use refers to at least 30 days. Only states with information from 2009 to 2015 were included in the regression. *** p<0.01, ** p<0.05, * p<0.1
20
Table 3. Fixed effects regression analysis on alcohol use from 2009 to 2015
Adolescents
Adults
Recent alcohol use
Recent alcohol use
𝐵𝑎𝑛𝑆𝑌
-1.023*
-0.286
(0.572)
(0.489)
𝐴𝑑𝑢𝑙𝑡𝑑𝑟𝑖𝑛𝑘𝑆𝑌0.0362
(0.103)
𝐴𝑙𝑐𝑜ℎ𝑜𝑙𝑡𝑎𝑥𝑆𝑌0.943
-1.837
(1.712)
(1.089)
𝑈𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡𝑆𝑌-0.197
-0.436
(0.195)
(0.272)
𝑀𝑒𝑑𝑖𝑎𝑛𝑖𝑛𝑐𝑜𝑚𝑒𝑆𝑌-6.46e-05
9.71e-05
(7.90e-05)
(9.28e-05)
𝐴𝑑𝑜𝑙𝑒𝑠𝑐𝑒𝑛𝑡𝑠𝑚𝑜𝑘𝑒𝑆𝑌0.563***
(0.1000)
𝐴𝑑𝑢𝑙𝑡𝑠𝑚𝑜𝑘𝑒𝑆𝑌0.0266
(0.0557)
Constant
31.24***
49.76***
(7.151)
(4.254)
Observations
108
108
R-squared
0.895
0.411
Number of states
27
27
State fixed effects
YES
YES
Year fixed effects
YES
YES
Notes: Robust standard errors in parentheses, standard errors are clustered per state. Recent use refers to
consumption on at least one day in the 30 days preceding the survey. Only states with information from 2009 to 2015 were included in the regression. *** p<0.01, ** p<0.05, * p<0.1
21
Table 4. Fixed effects regression analysis on adolescent consumption rates including lead
regressors.
Recent
smoking rate
Regular
smoking rate
Heavy
smoking rate
Recent alcohol
use
𝐵𝑎𝑛𝑆𝑌+10.838
0.893***
0.838***
0.0709
(0.648)
(0.316)
(0.219)
(0.787)
𝐴𝑑𝑢𝑙𝑡𝑠𝑚𝑜𝑘𝑒𝑆𝑌+10.0132
-0.0280
-0.0479
(0.123)
(0.0520)
(0.0638)
𝐶𝑖𝑔𝑡𝑎𝑥𝑆𝑌+10.0522
-0.767
-1.343***
(1.030)
(0.461)
(0.447)
𝑈𝑛𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡𝑆𝑌+10.0705
0.0310
-0.204
-0.774*
(0.355)
(0.171)
(0.153)
(0.404)
𝑀𝑒𝑑𝑖𝑎𝑛𝑖𝑛𝑐𝑜𝑚𝑒𝑆𝑌+1
1.19e-05
3.50e-05
3.40e-05
-0.000140
(0.000107)
(4.94e-05)
(4.88e-05)
(0.000118)
𝐴𝑑𝑜𝑙𝑒𝑠𝑐𝑒𝑛𝑡𝑑𝑟𝑖𝑛𝑘𝑆𝑌+10.153*
-0.0154
-0.0127
(0.0790)
(0.0447)
(0.0548)
𝐴𝑑𝑢𝑙𝑡𝑑𝑟𝑖𝑛𝑘𝑆𝑌+10.434**
(0.158)
𝐴𝑙𝑐𝑜ℎ𝑜𝑙𝑡𝑎𝑥𝑆𝑌+11.265
(1.504)
𝐴𝑑𝑜𝑙𝑒𝑠𝑐𝑒𝑛𝑡𝑠𝑚𝑜𝑘𝑒𝑆𝑌+1-0.206
(0.167)
Constant
-0.820
3.858
9.443
13.04
(12.71)
(6.307)
(6.802)
(13.46)
Observations
81
81
81
81
R-squared
0.831
0.843
0.813
0.844
Number of states
27
27
27
27
Original regressors
YES
YES
YES
YES
State fixed effects
YES
YES
YES
YES
Year fixed effects
YES
YES
YES
YES
Notes: Robust standard errors in parentheses, standard errors are clustered per state. Recent use refers to
consumption on at least one day in the 30 days preceding the survey, regular use refers to at least 20 days and heavy use refers to at least 30 days. Original regressors refers to the regressors used in regression model 1 and 2. Only states with information from 2009 to 2015 were included in the regression. *** p<0.01, ** p<0.05, * p<0.1
