Upload
judy-domingo
View
219
Download
0
Tags:
Embed Size (px)
DESCRIPTION
In this paper we evaluate the effectiveness of cigarette taxes as a mechanism to reducesmoking rates among adolescents. Using data from a nationally representative sample ofadolescents, we use an endogenity-corrected model with school-level fixed effects toobtain our estimates. Moving beyond the conventional definition of smoking to adefinition that recognizes the complex nature of addiction by categorizing smoking intovarious stages and also by controlling for peer and family effects together, we learn thatadolescents are not necessarily the most responsive to taxes. Influence from peers andfamily plays a more significant role in influencing adolescent smoking.
Citation preview
Electronic copy available at: http://ssrn.com/abstract=2181686
1
Social Learning Theory, Cigarette Taxes and Adolescent Smoking Behavior
Mir M Ali*
Abstract In this paper we evaluate the effectiveness of cigarette taxes as a mechanism to reduce smoking rates among adolescents. Using data from a nationally representative sample of adolescents, we use an endogenity-corrected model with school-level fixed effects to obtain our estimates. Moving beyond the conventional definition of smoking to a definition that recognizes the complex nature of addiction by categorizing smoking into various stages and also by controlling for peer and family effects together, we learn that adolescents are not necessarily the most responsive to taxes. Influence from peers and family plays a more significant role in influencing adolescent smoking. Forthcoming, Journal of Economics and Statistics. Keywords: Adolescent Smoking; Stages of Smoking; Social Learning Theory. Office of Regulations, Policy & Social Science, Food & Drug Administration, College Park, MD 20740, USA, phone: (240) 402-1784, fax: (301) 436-2637, e-mail: [email protected].
Electronic copy available at: http://ssrn.com/abstract=2181686
2
1. Introduction Youth smoking is an important public health policy concern today. According to
the latest Youth and Tobacco Use report released by the Center for Disease Control and
Prevention (CDC) , 4000 youths aged 12-17 try their first cigarette every day in the
United States. About 80% of adults who are smokers started smoking before they were
21 (Liang et al., 2001) and epidemiological evidence suggests that individuals who avoid
smoking in adolescence or young adulthood have significantly lower probabilities of
being addicted to smoking. Even though smoking prevalence has decreased steadily over
the years, this decrease is primarily attributable to increased quit rates rather than due to
fewer people starting to smoke (National Center for Health Statistics, 1995). Thus,
reducing smoking rates among adolescents remains a prime public health policy concern.
One of the major instruments that have emerged as the main policy tool in the
effort to reduce youth smoking is higher cigarette prices through increased cigarette
excise taxes. Studies advocating cigarette prices to curtail adolescent smoking found that
higher prices lead to lower levels of cigarette consumption (Chaloupka and Warner,
2000). Regarding adolescent smoking and excise taxes, a key policy concern is the
efficacy of tax policies aimed at changing the smoking behavior of adolescents who are at
an initial stage of smoking or have already made a transition into addiction. In other
words, are we only reducing smoking among adolescents for whom addiction is a less
likely outcome, suggesting that polices to increase prices are not target efficient?
Although an extensive literature has been devoted towards understanding the
addictive behavior of adolescents, the question regarding the target efficiency of taxes
has received relatively little attention. In most studies smoking has been defined as a
3
binary variable where smokers are defined as those individuals who smoked at least one
day out of the past thirty days. This definition fails to accurately account for the different
stages of progression a person goes through before becoming an addict, thus modeling
the different stages of addiction only as a choice rather than an outcome. If the youth at
the margin of deciding to take up smoking who respond to the tax policy are those with a
lower propensity for addiction and long-term use, policy gains will be exaggerated. In
this paper the effectiveness of price as a smoking reduction mechanism among
adolescents is examined by stage conceptualization of smoking, i.e. instead of relying on
the widely utilized binary indicator of smoking, we take into account the fact that an
individuals smoking behavior evolves through a sequence of several development stages
characterized by different smoking frequencies and intensities (Lloyd-Richardson et al.,
2002; Mayhew et al., 2000).
The influence of peers and parents on adolescent smoking behavior has been
discussed in the recent literature (Christakis and Fowler, 2008; Norton et al., 1998;
Powell et al., 2005; Powell and Chaloupka, 2005). Even though the studies found peers
and parents play a significant role in influencing smoking among adolescent, prices still
emerged as the main policy instrument. This could be due to the fact that most of the
studies to date do not control for the peer and family effects simultaneously. According to
the Social Learning Theory, adolescents acquisition behavior and values are based in
large part on a complex web of interpersonal social relations (Lloyd-Richardson et al.,
2002). In other words, an adolescent is influenced by his/her peer and family and such
influences are reflected by his/her engagement in certain activities. Thus, controlling for
both the peer and family effects (in terms of peer smoking status, parents smoking status,
4
parent-child relationship, living with both parents, etc.) together in addiction models will
enable us to more precisely estimate the influence of price (tax) on smoking, i.e. are we
overestimating the true impact of the price by failing to control for the peer and family
influence?
Despite the consistency of results across studies that have provided evidence in
terms of the significance of peer networks in smoking decisions, the magnitude of its
effect is still debated and questioned (Christakis and Fowler, 2008; Valente et al., 2005).
Skepticism regarding the influence of social network has been largely due to the fact that
much of the research was unable to address the issue of peer selection (Alexander et al.,
2001; Hall and Valente, 2007; Hoffman et al., 2007) and thus was unable to accurately
identify the endogenous or true social effect, i.e. the effect where an individuals
propensity to behave in some way varies with behavior of the reference group (Manski,
1993). Accurate estimation of this effect is relevant for policy because it suggests that
composition of peer networks is an important determinant of adolescent smoking and that
interventions that decline smoking propensities of individuals will also decrease the
smoking propensities of their peers (Fletcher, 2010). Building on this existing literature
we are able to extend our analysis in a couple of ways. First, our peer measures are drawn
not only from the nomination of close friends, but also from the individuals who attended
the same school as the respondents and were in the same grade with them as well, i.e.
classmates. This will allow us to identify the differences in effects that could be exerted
by different composition of the reference groups. Fletcher (2010) only examined the
influence of classmates but not close friends and also did not examine the impact of
excise taxes. Second, we are also able to build on this literature by improving upon the
5
methodology to purge the potential biases from the peer estimates, by correcting for
endogenity while accounting for various stages of smoking addiction. Even though
Fletcher (2010) was able to address the endogenity concerns of classmates smoking, it
utilized the standard dichotomous measure of smoking which ignores the various stages
of addiction that an adolescent goes through before becoming an addict.
Modeling addiction as a stage specific phenomenon and incorporating the social
learning theory will help us in identifying a number of important concerns that would
need to be addressed in devising policies aimed towards reducing long-term smoking
rates. From our analysis we would be able to identify the factors that play a more
significant role in adolescent smoking behavior. We would also be able to identify the
role that prices, state policies, peer effects and family influence play at different stages of
addiction. Although it might be correct to assume that it is irrelevant whether higher taxes
directly result in less tobacco consumption or whether the effect is transmitted through
the behavior of peers - this assumption would likely hold true for adult smokers only.
Most states in the U.S. prohibits sale of cigarettes to individual under the age of 18 (i.e.
adolescents), thus it is important from a policy perspective to identify the main
mechanism of influence so that other policies can be devised to complement excise taxes.
For example, if peer influence is identified to be a major contributor to smoking then it is
quite likely that an adolescent experimented with tobacco not because of the price of the
cigarette but because he/she might have seen their friends smoke.
In sum, in this paper we aim to (i) analyze the effectiveness of tobacco tax
policies as a tool to reduce smoking rates. In other words, we intend to find out if these
price measures are an overestimate and indicate target inefficiencies, i.e. reduce smoking
6
among adolescents for whom addiction is a less likely outcome; (ii) analyze how
different factors (price increase along with state policies, family influence, association
with friend) affect smoking behavior at various stages of addiction, and (iii) propose a
mechanism to obtain unbiased peer effect estimates.
2. Empirical Model and Data
2.1 Data
The data utilized in our study are drawn from The National Longitudinal Study of
Adolescent Health (Add Health). Add Health consists of data on adolescents in 132
schools nationwide between grades 7 to 12. The in-school portion of the first wave of the
survey (1994) contains cross-section data on about 90,000 adolescents. A subset of the
initial sample (20,745 respondents) was also interviewed in their home (in-home portion
of the data). The primary data for our analysis come from the first wave of the in-home
survey portion of Add Health. Parents were also interviewed in the first wave of the in-
home sample. This allows us to control for a wide range of parent-child relationship
measures as well as the smoking status of the parents.
A primary advantage of Add Health is that it asked respondents to nominate their
five closest male and five closest female friends and since these friends were also part of
the survey we were able to construct peer measures of smoking from the responses of the
friends themselves. The average number of nominated friends per individual is 2.54 and
approximately 85% of the friends are from the same school as the respondent. Thus, the
sample of our analysis with nominated peers consists of 6,549 adolescents with at least
one nominated friend interviewed in Add Health. The sample size of our grade-level peer
analysis consists of 19,988 individuals. Besides smoking information, Add Health
7
contains an extensive range of variables regarding smoking related state policies and
various demographic variables. Table 1 reports descriptive statistics from the first wave
(1994) of Add Health data.
2.1.1. Categorical Smoking Variables
Our main dependent variable in the study is the individual's smoking stage. The
smoking stages were created based on the individuals smoking frequency and recency .
Following Llyod-Richardson et al. (2002) we categorize people into the following stages
of smoking addiction:
(i) Never Smoker - Those respondents who denied ever trying a puff of cigarettes. Such
adolescents who have never smoked are either unaware of positive reasons to initiate
smoking or are ignoring or resisting pressure to smoke (Mayhew et al., 2000).
(ii) Experimental - Those who endorse trying cigarettes, although denied smoking within
the past 30 days or ever smoking regularly (i.e. daily smoking). This stage is marked by
adolescents trying their first few cigarettes.
(iii) Intermittent - Those who reported smoking between 1 and 29 out of the past 30 days.
This stage is characterized by a gradual increase in the frequency of smoking and an
increase in the variety of situations in which cigarettes are used. Adolescents
in this stage have progressed beyond sporadic smoking to smoke on a higher but still
infrequent basis.
(iv) Regular - Those who responded smoking on a daily basis within the past thirty days.
In this final stage adolescent may experience nicotine dependence, withdrawal symptoms
and may find it difficult to quit.
8
In the previous literature, the most widely used measure of smoking was a
dichotomous variable that categorizes smokers as anyone who reported smoking at least
one day out of the past thirty days or a continuous variable measuring the number of
cigarettes smoked (Chaloupka and Warner, 2000). Tax estimates on such variables might
be an overestimate since it would fail to control for the various stages of addiction that a
person might go through. From our definition of smoking stages we can see that a person
might have smoked zero days out of the past thirty days and yet could fall into the
category of someone who has experimented with cigarettes and has the possibility of
making a transition into being a regular smoker. Thus, an overestimation of price is quite
likely without distinctions being made among the various stages of addiction since the
experimental smokers will be categorized as non-smokers.
2.1.2. Peer Measure Variables
We construct two different measures of peer variables. One is based on the
respondents nomination of close friends and another is based on the people who were in
the same grade and school as the respondent. For the nominated friends we created a
variable pertaining to the percentage of friends who are smokers, i.e. they are either
experimental, intermittent or regular smokers. The school-level peer smoking measure is
the percentage of students (excluding the respondent) in the same grade and school as the
respondent who are smokers, i.e. they are either experimental, intermittent or regular
smokers.
2.1.3 Parent Measures
Since Add Health interviewed one of the parents, we can control for the parent-
child relationship via not only how the child perceives it to be, but also how the parent
9
perceives it. This will allow us to capture with more precision the role that parents play in
influencing an adolescents smoking behavior (Ennett at al., 2001). Besides the smoking
behavior of the parents, it is also the relationship that the child has with his/her parents
that is expected to play a vital role in influencing their smoking behavior. Apart from
controlling for the respondent's perception of whether he/she thinks that his/her parents
care, understand, and pay attention, we also control for the respondents' parents
perception of whether they think they get along well with their children, whether they
feel they can trust their children, etc.
2.1.4 Policy Measures
Besides cigarette excise taxes (in dollars) an extensive array of state policy
measures are controlled for. Like taxes, state policy measures have been labeled as one of
the most effective policy instruments that could be utilized to curtail smoking among
adolescents. State policies such as banning cigarettes sales via vending machine,
marketing restrictions on billboards, prohibition of distribution of free samples as
promotional tools, state initiating dissemination of information regarding the adverse
effect of smoking etc. are all likely to contribute to preventing adolescents from taking up
smoking. Besides these, various other state and local enforcement programs are likely to
limit the availability of cigarettes among adolescents.
In addition to these measures, we also control for socio-demographic factors like
age, the grade they are in, family structure and gender.
2.1.5 Factor Analysis
Following DeCicca et al. (2007) and Kan and Tsai (2004), instead of relying on
multiple indicators to measure the parent-child relationship and state policy measure, we
10
created an index to control for that. The indicators of the parent-child relationship and
state policy were obtained after conducting exploratory (principal factor) factor analysis.
This is a data reduction technique conducted to minimize correlated regressor problems,
collinearity and improve interpretation of mutually exclusive components of responses.
Thus, factor analysis was utilized as a statistical technique to find an indicator(s) of
orthogonal common factor that will linearly summarize a set of original variables related
to the parent-child relationship and state policies (Kan and Tsai, 2004). Even though an
interpretation of these indexes as estasticities is not advisable, this would allow us to infer
from the estimates whether the parent-child relationship and state policies have a
significant position or negative effect on adolescent smoking behavior.
We utilized the Kaiser criterion since it is one of the most utilized methods in the
literature in deciding the number of factors to be retained. According to the Kaiser
criterion, only factors that have eigenvalues greater than 1.0 are retained. Table 2 through
4 reports the factor analysis results. The factor loadings and scoring coefficients of the
factors with eigenvalues greater than 1.0 are presented in Table 3 and 4. From Table 2,
we can see that for both the parent-child relationship and state policy variables there is
only one factor with an eigenvalue greater than 1. This suggests that only one factor is
sufficient to summarize all the variables pertaining to the parent-child relationship and
anti-smoking state policies. Thus in our regressions we include this single factor as an
index for parent-child relationship and anti-smoking state policies.
2.2 Estimation Framework
We estimate an ordered probit model of smoking behavior where the dependent
variable, *iY representing an individuals smoking stage (0 = never smoker; 1 =
11
experimental smoker; 2 = intermittent smoker; 3 = regular smoker) is specified as
follows:
iiiiiii SXCFPY 543210* (1) where Pi refers to the vector of peer measures (nominated friends and school-based) for
individual i, Fi refers to individuals vector of family characteristics such as parent-child
relationship, parents smoking status and easy access to cigarettes at home, Ci refers to
state level cigarette taxes faced by the individual and state level tobacco control policies.
Xi refers to the vector of demographic characteristics such as age, gender, race, education
etc. and finally Si is the school dummies that controls for unobserved school type
(school-level fixed effects) or school-level factors that are common to all individuals
within the same school. Finally, i refers to the error term. Given the standard normal assumption for i we simply compute each response probability as follows:
)()|()|()|0( 111* xxxPxYPxYP
*2 121( 1| ) ( | ) ( ) ( )P Y x P Y x x x
*3 232( 2 | ) ( | ) ( ) ( )P Y x P Y x x x
)(1)|()|3( 33* xxYPxYP
where 321 are threshold parameters and x is the vector of all of the explanatory variables.
The main coefficient of interest is the endogenous effect 1, which indicates the extent of peer influence on an individuals decision to smoke. If 1 is estimated to be positive, then any policy intervention that alters the smoking behavior of the individual
within a reference group or social network would have an effect on non-treated
12
adolescents smoking behavior that are in the same social network (Ali et al. 2011). A
standard linear regression using average contemporaneous measure by a reference group
as a proxy for social interactions is easy to estimate. However, such measures of peer
networks or social interactions have quite a few problems of interpretation (Ali et al.,
2011). Manski (1993, 2003) identifies and distinguishes between three different types of
peer or social effects:
a. Endogenous This effect occurs when individual behavior responds to the behavior of
others in the reference group.
b. Exogenous or Contextual effect This occurs when individual behavior responds to
the exogenous characteristics of the reference group.
c. Correlated effect This occurs when individuals in the same group behaves similarly
because they have similar unobserved characteristics or they face similar institutional
characteristics. Therefore, correlated effects do not signify social interactions, but
confound its estimation.
In the case of smoking, endogenous effect can take place if individuals are more
likely to smoke if their classmates or close friends smoke, i.e. their decision to participate
in smoking is interrelated. Contextual (or exogenous) effect can occur, for example, if an
individual is more likely to smoke if he or she is surrounded by peers who have parents
who smoke (Ali et al., 2011). Correlated effects can occur if individuals in the same
school or neighborhood choose to smoke because they each face a lower opportunity cost
of smoking, for example, low cigarette prices, easy availability etc. This could also occur
if individuals with similar taste or preference for smoking tend to be friends, i.e. peer
selection (Alexander et al., 2001; Hall and Valente, 2007; Hoffman et al., 2007).
13
Manski (1993, 2000) also notes that while both the endogenous and exogenous
effects signify social interactions, correlated effects, however, are a statistical non-social
phenomenon. As such, policy implications would differ significantly depending on the
type or category of the social effect that we are identifying. For example, if smoking in
schools is characterized by an endogenous effect then any policies that are designed to
reduce smoking in one particular group or an individual will produce a social multiplier
and affect individuals who were not directly targeted by the policy (Ali et al., 2011).
Exogenous effects on the other hand may not imply the same multiplier effect since the
policy does not alter the reference groups characteristics. Correlated effects, which are
not evidence of social interactions, can lead to misleading policy implications if they are
not identified or adequately controlled for. Standard regressions of individual propensity
to engage in a particular activity on group means are unable to distinguish between the
endogenous, exogenous and correlated effects and thus is unable to provide a causal
estimation of peer effects. This identification difficulty, coined as the reflection problem
by Manski, occurs because group behavior by definition is the aggregation of individual
behavior, i.e. group behavior affects individual behavior and vice versa due to the
simultaneity in choices. Thus, for the purpose of devising effective policy it is important
to purge these biases from peer effect estimates to identify whether peer influence is more
important than peer selection (Ali et al., 2011; Norton et al., 1998).
Estimating our models with Si, the school-level fixed effects, potentially mitigates
the correlated effects. However, an instrumental variable regression is also necessary in
this empirical analysis because of the reflection problem. The reflection problem, as
mentioned previously, arises because peer behavior affects individual behavior and vice
14
versa. Manski demonstrated that most estimates of 1 are not identified without utilizing instrumental variables or other similar methodologies (Manski, 1993, 2000). The
instrumental variable solves the reflection problem by identifying variables that are
related to the group level variable but not to the individual level variable. To utilize the
instrumental variable technique or IV one must first find variables or instruments that
have two properties. First, they have to be correlated with (causes variation in) the
variable whose effect we want to know about. Here it is the peer measure. Second, these
instruments must have no direct effect on the outcome measure (Yi in 1). For our
instrument we propose three variables:- (i) the percentage of peers who have parents who
smoke, (ii) the percentage of peers who live with both biological parents and (iii) the
percentage of peers who have easy access to cigarettes at home. These peer level
variables do not predict individual behavior but rather predicts the peer level behavior.
The intuition behind the instruments is that, while individuals who have parents who
smoke are more likely to smoke, the proportion of individuals friend who have smoking
parents will only directly affect the friend but not the individual. Similar intuition applies
to the other instruments; they affect the individuals friends decision to smoke but not
that of the individual himself or herself. The first stage will thus estimate the following
equation -
iiiiii XCFZP 43210 (2) where iZ refers to the instruments and i refers to the residual. However, we
cannot implement the two-stage least squares estimation due to the ordered and
categorical nature of the dependent variable (Terza et al., 2008). We thus implement a
control function approach in the ordered probit regression to address the issue of
15
endogenity (Lee, 2007; Terza et al., 2008; Fang et al., 2009). In the first stage of the
control function approach, a reduced form regression is estimated and the result is used to
generate predicted value of the endogenous variable (peer smoking in our case). This
predicted value is then used to obtain the fitted residual, which is then included as an
additional regressor (along with the endogenous variable) in the second-stage estimation.
Terza et al. (2008), strongly advises about the inclusion of this residual in estimation of
nonlinear models with endogenous regressor and provides evidence that this approach
leads to more consistent results.
It is possible that the first stage of this control function approach could be an
ordinary least square (Lee, 2007; Fang et al., 2009). After obtaining the estimated
coefficients, we can predict the dependent variable in Equation (2) as P . We then obtain
the fitted residual in Equation (2) by subtracting the predicted value P from P. The second stage inserts the fitted residual into Equation (1) and estimates the following
model by ordered probit regression:
iiiiiii SXCFPY ^
6543210* (3)
Due to the two-step feature of the model, we adjust the standard errors in the
second step by nonparametric bootstrap techniques using 200 replications. Combined
with the school-level fixed effects, the control function approach will enable us to obtain
evidence suggestive of a causal effect of social interaction on individual smoking
behavior. We also undertake several tests to verify the validity of our instruments.
3. Empirical Results
Table 5 presents our ordered probit model without the social learning mechanism.
From the table we can see that even without controlling for social networks, cigarette
16
taxes do not have a significant effect on all smoking stages and affects only the never
smokers and regular smokers. This suggests that the dichotomous measure of smoking
status could lead to an upward bias in price estimates. Experimental smokers could
progress on to becoming regular smokers and an increase in cigarette excise tax alone
might not be able to prevent the progression to a higher addictive stage. The primary
reason for adolescents being an important group for policy focus is to prevent them from
initiating smoking and reduce eventual addiction. This may not be accomplished with
reliance on prices alone. Also, since the effect of tax on experimental and intermittent
smokers are statistically insignificant, it could imply that taxes alone may not be able to
prevent the progression of such smokers into the full addictive stage.
The policy index is also significant for the never smokers and regular smokers
only. Thus, state level policy measure might induce a reduction in smoking by the regular
smokers or induce a quitting behavior. It might also prevent the never smokers from
taking up smoking. The taxes and policy index being significant and greater in magnitude
for the never smokers confirms our prior hypothesis that tax policies in particular might
not be target efficient, i.e. unable to reduce smoking among those with a higher
propensity for addiction. Never smokers are the ones with the lowest propensity of
transitioning into an addictive stage and taxes have the greatest effect on them.
Demographic variables, parents education and work status has the expected effects.
Older individuals and those who are in a higher grade are less likely to be never smokers;
whites are more likely to be smokers than blacks and Hispanics. Being religious
(frequency of religious service attendance) is negatively associated with smoking.
17
Parents education is also negatively related with smoking status, whereas having parents
who work full-time outside the home is positively related to smoking.
From Tables 6 and 7 we can see that the effect of taxes and state policy index
becomes statistically insignificant after controlling for peer and family smoking (i.e.
social learning theory). After the inclusion of peer smoking, parent smoking, parent-child
relationship index and other variables related to the social learning mechanism, cigarette
excise tax is no longer significant for any stage of smoking. The peer measure in Table 6
pertains to close friends and the peer measure in Table 7 pertains to classmates. In both
cases, peer and family factors appears to play a significant role, with the classmates peer
measure being larger in magnitude compared with close friends peer measure. However,
as mentioned previously in Section 2.2, the peer measures are likely to be endogenous.
We adopt a control function approach to account for that in Tables 8 and 9. For the
purpose of brevity we show the results here for our main variables of interest only, since
the effect of the other control variables are quite similar to those presented in Tables 5 to
7.Appendix 1 provides estimates of our first stage regression of the control function
approach.
From Table 8 we see that the peer effects are only statistically significant for
never smokers The estimated peer coefficient means that a 10% increase in friends who
are smokers decreases the probability of being a never smoker by 3%. The statistically
insignificant effect of peer smokers underscores the importance of controlling for
endogenity of peer smoking. However, parental characteristics such as parental smoking
status, parent-child relationship, easy access to cigarettes exert a significant influence on
adolescent smoking behavior as hypothesized. These results are in line with our
18
hypothesis that the social learning mechanism will have different effects on different type
of smokers and that taxes might not be as effective a policy instrument as the previous
literature has estimated it to be. Particularly, in this case, our results may imply that
parental smoking behavior might be more influential, especially for experimental,
intermittent and regular smokers
It is important to mention that models that regress individual smoking stage on
friends smoking stage without accounting for schoollevel fixed effects produce a
larger peer effect estimates for all stages of smoking. However, such peer effect estimates
are biased since they are unable to purge the correlated effects from the peer measures to
identify the true (endogenous) peer effect. Estimating models with school-level fixed
effects reveals that individuals that are smokers might be more highly influenced by
unobserved environmental factors in their surroundings and this argument is further
advanced by the positive and significant effect of easy access to cigarettes on all stages of
smoking. In this regard, we might expect the more exogenous school-level peer measure
to exert a greater influence on being a smoker. Conceptually, school-level peer measures
are likely to operate through different causal mechanism than nominated peers. For
example, such broader measures of peer social network are likely to operate through the
establishment of smoking norms rather than operating through influence. Thus
unobserved school-level factors that are common to all individuals within the same
school could be more important to being a smoker.
With peers defined as students in the same grade and school as the respondent,
excluding the respondent, we see from Table 9 that the peer measures are greater in
magnitude compared to the nominated peer measures, except for experimental smokers,
19
where peer effects are not statistically significant. This greater effect of classmates over
nominated peers imply that school based intervention strategies to prevent smoking might
be quite effective since a change in behavior in one adolescent might spread to all the
classmates regardless of whether they are close friends or not. Cigarette excise tax along
with state-level policy index, are once again not significant, implying that for adolescents
the social learning mechanism could be more important, even though its effect could vary
in magnitude depending on the stage of addiction. Another important thing to note from
these sets of results is that the coefficient of our fitted residual is significant at the 1%
level for both nominated friends and classmates, indicating the importance of accounting
for the endogenity of peer smoking.
4. Conclusion
Prices or an increase in excise taxes have emerged as the main policy instrument
to reduce adolescent smoking. The vast majority of the literature has found prices to have
a significant negative impact on the probability of being a smoker. In this study we
analyze the effectiveness of prices (tax) as a smoking reduction mechanism by estimating
models that incorporate the stage conceptualization models of smoking and Social
Learning Theory. Biases in peer social networks have been accounted for by using
school-level fixed effects in the instrumental variable or the two-stage least square
analysis and two alternative definitions of peer, nominated friends and school-level peers.
Moving beyond the conventional definition of smoking to a definition that
recognizes the complex nature of addiction by categorizing smoking into various stages,
we learn that prices are necessarily not the most effective instrument to curtail smoking.
20
From the stage conceptualization of smoking model specification we learn that the effect
of prices across the different stages is not homogeneous even when we do not control for
the social learning mechanism. Prices remain insignificant after controlling for the social
learning mechanism under both peer measures. A possible avenue for future research
could be to evaluate how excise tax influence adult smokers using the stage
conceptualization of smoking rather than a binary indicator to measure smoking behavior.
A double-hurdle model (Jones 1989) could be adopted to examine this issue since it
would allow for the separation of non-starters and quitters. As with any empirical
strategy, our approach is subject to criticism and thus it is prudent to regard our results as
demonstrating a strong association between peer smoking and individuals smoking
behavior at various stages of addiction rather than demonstrating a causal relationship. If
more evidence from future studies confirms our findings, the resulting body of literature
may lead readers to infer causality. Our study only points in that direction provided that
our assumptions hold.
Adolescent smoking behaviors are not only dependent on their own propensity to
be an addict, but are also dependent on an extensive array of learning factors that are
influenced by their peer and parents. The subjective judgment on the cost and
benefits of smoking plays a very significant role in smoking decision. An individuals
prior belief on risk and benefits from smoking is a crucial factor in the smoking decision,
which depends largely on his demographic and other socio-economic characteristics
(Lahiri and Song, 2000). The Social Learning Theory states that the interpersonal social
relationship in which the individual is fixed is of primary importance in adolescent
acquisitions of behaviors and values.
21
From our results we can see that relying on taxes alone is not sufficient to reduce
smoking. Recent findings by Katzman et al. (2007) on the existence of a social market
through which adolescents can obtain cigarettes further reinforce our conclusion.
Katzman et al. (2007) also concludes that prices will be ineffective in influencing
smokers who obtain cigarettes through the social market. Increases in prices will not
achieve target efficient reductions of adolescent smoking among those who continue to
be around peers who smoke and live in households where parents smokes and access to
cigarettes is easy. Thus, increased social awareness regarding the adverse effect of
smoking is more likely to contribute towards a reduction in smoking.
Acknowledgement
The author would like to thank Debra Dwyer (Stony Brook University), John Rizzo (Stony Brook University) and Mark Montogomery (Stony Brook University) for their helpful comments and suggestions. The views expressed in this paper are those of the author and do not necessarily reflect the view of the Food & Drug Administration. This research uses data from Add Health, a program project designed by J. Richard Udry Peter S. Bearman, and Kathleen Mullan Harris, and funded by a grant P01- HD31921 from the National Institute of Child Health and Human Development, with cooperative funding from 17 other agencies. Special acknowledgment is due Ronald R. Rindfuss and Barbara Entwisle for assistance in the original design. Persons interested in obtaining data files from Add Health should contact Add Health, Carolina Population Center, 123 W. Franklin Street, Chapel Hill, NC 27516-2524 ([email protected]).
22
References
Alexander, C., Piazza, M., Mekos, D., and Valente, T. (2001), Peers, Schools, and Adolescents Cigarette Smoking. Journal of Adolescent Health, 29(1): 22 30. Ali. MM., Amialchuk, A., and Dwyer, DS. (2011), The Social Contagion Effect of Marijuana Use among Adolescents. PLoS ONE 6(1): e16183. Chaloupka, F., and Warner, K. (2000), The Economics of Smoking. In Newhouse J. Culyer, A. (eds), Handbook of Health Economics Volume 1B. North- Holland, 2000. Christakis, N., and Fowler J. (2008), The Collective Dynamic of Smoking in a Large Social Network. The New England Journal of Medicine, 358: 2249 - 2258. DeCicca, P., Kenkel, D., Mathios, A., Shin, Y-J., Lim, J-Y. (2007), Youth Smoking, Cigarette Prices and Anti-Smoking Sentiments. Health Economics, 17: 733 - 749. Ennett, S., Bauman, K., Foshee, V., Pemberton, M., and Hicks, K. (2001), Parent-child communication about adolescent tobacco and alcohol use: What do parents say and does it affect youth behavior?. Journal of Marriage and Family, 63:4862. Eisenberg, D., and Quinn, B. (2006), Estimating the Effect of Smoking Cessation on Weight Gain: An Instrumental Variable Approach. Health Services Research, 41(6): 2255 2266. Fang. H., Ali, MM., Rizzo, JA. (2009), Does Smoking Affect Body Weight and Obesity in China. Economics and Human Biology, 7(3): 334 350. Fletcher J. (2010), Social Interactions and Smoking Decisions: Evidence Using Multiple Cohorts, Instrumental Variables and School Fixed Effects. Health Economics, 19(4): 466 484.. Hall, J.A., and Valente, T.W. (2007), Adolescent smoking networks: The effects of influence and selection on future smoking. Addictive Behaviors, 32(12), 3054- 3059. Hoffman, B.R., Monge, P.R., Chou, C.-P., and Valente, T.W. (2007), Perceived peer influence and peer selection on adolescent smoking. Addictive Behaviors, 32(8), 1546- 1554. Jones, A. (1989), A Double-Hurdle Model of Cigarette Consumption. Journal of Applied Econometrics, 4: 23 39.
23
K. Kan, W-D Tsai. (2004), Obesity and Risk Knowledge. Journal of Health Economics, 23(5): 907-934. B. Katzman, S. Markowitz, and K A. McGreory. (2007), An Empirical Investigation of the Social Market for Cigarettes. Health Economics, 16: 1025-1039. Lee. S. (2007), Endogeneity in quantile regression models: a control function approach. Journal of Econometrics 141: 1131-1158. K. Lahiri and J. Song. (2000), The effect of smoking on health using a sequential self- selection model. Health Economics, 9:491511. L. Liang, F. Chaloupka, M. Nichter, and R. Clayton. (2001), Prices, policies and youth smoking. Addiction, 98(1):105122. E. Lloyd-Richardson, G. Papandonatos, A. Kazura, C. Stanton, and R. Niaura. (2002), Differentiating stages of smoking intensity among adolescents: Stage-specific psychological and social influences. Journal of Consulting and Clinical Pyschology, 70(4):9981009. Manski, C. (1993), Identification of Endogenous Social Effects: The Reflection Problem. Review of Economic Studies, 60(3): 531 542. Manski, C. (2000), Economic Analysis of Social Interactions. Journal of Economic Perspectives. 14(3):115 136. Mayhew, K., Flay, B., and Mott, J. (2000), Stages in the development of adolescent smoking Drug and Alcohol Dependence, S9(1):S61S81. Newhouse, J., and McClellan, M. (1998), Econometrics in Outcomes Research: The Use of Instrumental Variables. Annual Review of Public Health, 19:1734. Norton, E., Lindrooth, R., and Ennett, S. (1998), Controlling for the endogenity of peer substance use on adolescent alcohol and tobacco use. Health Economics, 7:439 453. L. Powell and F. Chaloupka. (2005), Parents, public policy and youth smoking. Journal of Policy Analysis and Management, 24(1):93112. L. Powell, J. Tauras, and H. Ross. (2005), The importance of peer effects, cigarette prices and tobacco control policies for youth smoking behavior. Journal of Health Economics, 24:950968. Schoenbaum, M., Unutzer, J., McCaffrey, D., Duan, N., Sherbourne, C., and Wells, K. (2002), The Effects of Primary Care Depression Treatment on Patients Clinical Status and Employment. Health Services Research, 37(5):114558.
24
Terza, JV., Basu, A., Rathouz, PJ. (2008), Two Stage Residual Inclusion Estimation: Addressing Endogenity in Health Econometric Modeling. Journal of Health Economics, 27(3): 531 543. Valente, T., Unger, J., Johnson, A. (2005), Do Popular Students Smoke? The Association among Popularity and Smoking among Middle School Students. Journal of Adolescent Health, 37(4): 323 329.
25
Table 1: Descriptive Statistics from Wave I (1994)
Variables Mean Std. Dev. Min Max
Smoking Stage: - - Never Smoker 0.49 0.49 0 1
Experimental Smoker 0.25 0.34 0 1 Intermittent Smoker 0.17 0.37 0 1
Regular Smoker 0.09 0.29 0 1 Tax 0.33 0.16 0.025 0.75
Policy Index 1.35 0.95 -0.79 1.93 Age 15.15 1.74 11 20
Grade 9.67 1.63 7 12 Male 0.49 0.50 0 1 White 0.61 0.48 0 1 Black 0.23 0.42 0 1
Hispanic 0.17 0.37 0 1 Religious 2.73 2.01 1 9
Parent Smoke 0.29 0.45 0 1 Easy Access to
Cigarettes 0.31 0.46 0 1
Lives with both Parents 0.49 0.50 0 1 Mother College 0.25 0.43 0 1 Father College 0.21 0.41 0 1 Work Fulltime 0.29 0.45 0 1
N: 19,988
26
Table 2: Factor Analysis Results - Eigenvalues
Eigenvalue Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6 Factor 7 Factor 8
State Policy 3.47424 0.98259 0.34966 0.13239 -0.05573 -0.10363 -0.16013 -0.21682 Parent-Child Relationship
2.25837 0.85097 -0.00106 -0.06302 -0.08920 -0.11075 -0.15053 -0.22606
Table 3: Factor Loadings and Scoring Coefficients State Policies Factor Loadings Scoring Coefficients Factor 1 Factor 1 Vending Machines Banned in location accessible to youths
0.5769 0.07586
Billboard Prohibited within 500 feet of school
0.7056 0.12497
Free Sample Prohibited 0.0363 -0.01104 Marketing Restrictions 0.8654 0.36874 Local Enforcement of Youth Access
-0.7305 -0.16956
Dissemination of Information 0.8028 0.26638 Education Measurement for Compliance
-0.4308 -0.05026
Schools Required to Offer Tobacco use Prevention Programs
-0.7277 -0.13654
Table 4: Factor Loadings and Scoring Coefficients Parent-Child Relationship
Factor Loadings Scoring Coefficients Factor 1 Factor 1
Respondents Survey: Parents Care 0.2117 0.05434
Parents Understand 0.2168 0.06921 Parents Pay Attention 0.2747 0.09283
Communicate 0.2728 0.06833 Parents Survey:
Get Along 0.7941 0.36525 Understand 0.6504 0.19047
Trust 0.7360 0.26783 Decision 0.6491 0.18838
27
Table 5: Ordered Probit Estimates (Marginal Effects) without Social Learning Theory
Variables LR Chi2: 2085.60
Log Likelihood: -25360.05 N: 19,988
Never Smoker Experimental Smoker Intermittent Smoker Regular Smoker
Tax 0.039** (0.019)
-0.008 (0.010)
-0.017 (0.020)
-0.014** (0.007)
Policy Index 0.017** (0.005)
0.003 (0.005)
-0.001 (0.003)
-0.010** (0.004)
Parent Characteristics
Mom College 0.080*** (0.008)
-0.017 (0.020)
-0.025 (0.030)
-0.028*** (0.003)
Dad College 0.021*** (0.004)
-0.005 (0.004)
-0.009 (0.008)
-0.007*** (0.001)
Work Fulltime -0.015*** (0.004)
0.003*** (0.001)
0.006 (0.004)
-0.005 (0.005)
Demographics Age -0.025***
(0.005) 0.005*** (0.001)
0.011*** (0.002)
0.009*** (0.002)
Male 0.004 (0.006)
-0.001 (0.001)
-0.002 (0.003)
-0.001 (0.002)
White -0.074*** (0.010)
0.017*** (0.002)
0.033*** (0.004)
0.025*** (0.003)
Black 0.149*** (0.012)
-0.041*** (0.004)
-0.064*** (0.005)
-0.045*** (0.003)
Hispanic 0.059*** (0.010)
-0.015*** (0.003)
-0.026*** (0.004)
-0.019*** (0.003)
Grade -0.020*** (0.005)
0.004*** (0.001)
0.009*** (0.002)
0.007*** (0.002)
Religious 0.094*** (0.006)
-0.019*** (0.001)
-0.042*** (0.003)
-0.034*** (0.002)
Notes: Include school fixed effects. Robust Standard Errors in parenthesis. *** - sig. at the 1% level; ** - sig. at the 5% level; * - sig. at the 10% level.
28
Table 6: Ordered Probit Estimates (Marginal Effects) with Social Learning Theory
Variables LR Chi2: 6404.44
Log Likelihood: -23200.63 N: 6,549
Never Smoker Experimental Smoker Intermittent Smoker
Regular Smoker
Tax -0.007
(0.013) 0.002
(0.003) 0.003
(0.007) 0.002
(0.003) Policy Index 0.005
(0.004) -0.001 (0.001)
-0.003 (0.002)
-0.001 (0.001)
Peer Smoking Nominated -0.193***
(0.003) 0.048*** (0.002)
0.101*** (0.002)
0.04*** (0.001)
Parent Characteristics
Lives with both parents
0.055*** (0.008)
-0.014*** (0.002)
-0.029*** (0.004)
-0.013*** (0.002)
Easy Access to Cigarettes
-0.010*** (0.007)
0.021*** (0.001)
0.053*** (0.004)
0.027*** (0.002)
Parent Smoke -0.029*** (0.008)
0.007*** (0.002)
0.016*** (0.004)
0.007*** (0.002)
Parent-Child Relationship Index
0.036*** (0.004)
-0.009*** (0.001)
-0.019*** (0.002)
-0.009*** (0.001)
Mom College -0.001 (0.004)
0.000 (0.001)
0.000 (0.002)
0.000 (0.001)
Dad College 0.003 (0.004)
-0.000 (0.000)
-0.000 (0.002)
-0.000 (0.001)
Work Fulltime -0.46*** (0.014)
0.007 (0.010)
-0.003 (0.019)
0.017*** (0.007)
Demographics Age 0.002
(0.005) -0.001 (0.001)
-0.001 (0.002)
-0.001 (0.001)
Male 0.003 (0.006)
-0.001 (0.002)
-0.002 (0.003)
-0.001 (0.002)
White -0.073*** (0.010)
0.019*** (0.003)
0.038*** (0.005)
0.017*** (0.002)
Black 0.084*** (0.012)
-0.024*** (0.004)
-0.043*** (0.006)
-0.018*** (0.002)
Hispanic 0.030** (0.010)
-0.008** (0.003)
-0.016** (0.005)
-0.007** (0.002)
Other -0.076** (0.025)
0.013*** (0.003)
0.042** (0.014)
0.022** (0.009)
Grade -0.032*** (0.005)
0.008*** (0.001)
0.017*** (0.003)
0.008*** (0.001)
Urban 0.019** (0.007)
-0.005** (0.002)
-0.010** (0.004)
-0.005** (0.002)
Religious 0.043*** (0.007)
-0.010*** (0.002)
-0.022*** (0.003)
-0.010*** (0.002)
Notes: Include school fixed effects. Robust Standard Errors in parenthesis. *** - sig. at the 1% level; ** - sig. at the 5% level; * - sig. at the 10% level.
29
Table 7: Ordered Probit Estimates (Marginal Effects) with Social Learning Theory
Variables LR Chi2: 2100.93
Log Likelihood: -18347.56 N: 19,988
Never Smoker Experimental Smoker
Intermittent Smoker
Regular Smoker
Tax 0.001
(0.015) -0.002 (0.004)
-0.004 (0.007)
-0.002 (0.004)
Policy Index 0.003 (0.004)
-0.001 (0.001)
-0.002 (0.002)
-0.001 (0.001)
Peer Smoking Classmates
-0.519***
(0.036) 0.130*** (0.010)
0.240*** (0.017)
0.149*** (0.011)
Parent Characteristics
Lives with both parents
0.056*** (0.009)
-0.014*** (0.002)
-0.026*** (0.004)
-0.016*** (0.003)
Easy Access to Cigarettes
-0.122*** (0.008)
0.026*** (0.002)
0.058*** (0.004)
0.040*** (0.003)
Parent Smoke -0.046*** (0.010)
0.011*** (0.002)
0.022*** (0.005)
0.014*** (0.003)
Parent-Child Relationship Index
0.055*** (0.004)
-0.014*** (0.001)
-0.026*** (0.002)
-0.016*** (0.001)
Mom College 0.003 (0.016)
0.006 (0.011)
-0.019 (0.020)
0.004 (0.008)
Dad College 0.021 (0.017)
-0.022* (0.012)
0.021 (0.024)
-0.004 (0.009)
Work Fulltime -0.046*** (0.014)
0.007 (0.010)
-0.003 (0.019)
0.017** (0.007)
Demographics Age -0.016**
(0.006) 0.004** (0.001)
0.008** (0.003)
0.005** (0.002)
Male -0.001 (0.007)
0.000 (0.002)
0.001 (0.003)
0.000 (0.002)
White -0.089**** (0.012)
0.023*** (0.003)
0.041*** (0.005)
0.025*** (0.003)
Black 0.117*** (0.014)
-0.034*** (0.005)
-0.053*** (0.006)
-0.030*** (0.003)
Hispanic 0.032** (0.012)
-0.009** (0.003)
-0.015** (0.005)
-0.009** (0.003)
Other -0.077** (0.029)
0.014* (0.004)
0.037** (0.015)
0.026** (0.012)
Grade -0.014** (0.006)
0.003** (0.001)
0.006** (0.003)
0.004** (0.002)
Urban 0.025** (0.008)
-0.006** (0.002)
-0.012** (0.004)
-0.007** (0.002)
Religious 0.061*** (0.008)
-0.014*** (0.002)
-0.029*** (0.004)
-0.018*** (0.002)
Notes: Include school fixed effects. Robust Standard Errors in parenthesis. *** - sig. at the 1% level; ** - sig. at the 5% level; * - sig. at the 10% level.
30
Table 8: Control Function Ordered Probit Estimates (Marginal Effects)
Variables LR Chi2: 7509.06 Log Likelihood: -22648.32
N: 6,549 Never
Smoker Experimental
Smoker Intermittent
Smoker Regular Smoker
Tax -0.007
(0.005) 0.004
(0.015) 0.005
(0.011) 0.001
(0.003) Policy Index 0.001
(0.001) -0.001 (0.004)
0.005 (0.003)
-0.004 (0.003)
Peer Smoking Nominated -0.306***
(0.018) 0.204* (0.107)
0.045 (0.041)
0.057 (0.064)
Parent Characteristics
Easy Access to Cigarettes
-0.078*** (0.001)
0.028*** (0.009)
0.022*** (0.007)
0.028*** (0.003)
Parent Smoke -0.035*** (0.001)
0.024** (0.010)
0.006*** (0.001)
0.005** (0.002)
Parent-Child Relationship Index
0.032*** (0.001)
-0.004*** (0.001)
-0.021*** (0.003)
-0.007*** (0.001)
Residual 0.492*** (0.087)
-0.114*** (0.021)
-0.238*** (0.042)
-0.145*** (0.025)
Notes: Estimates of only main variables of interest provided. Include school fixed effects. Robust Standard Errors in parenthesis. *** - sig. at the 1% level; ** - sig. at the 5% level; * - sig. at the 10% level. Instruments: % of peers whose parents smoke; % of peers with easy access to cigarettes at home; % of peers who lives with both parents.
31
Table 9: Control Function Ordered Probit Estimates (Marginal Effects)
Variables LR Chi2: 2597.79 Log Likelihood: -18099.13
N: 19,988 Never
Smokers Experimental
Smoker Intermittent
Smoker Regular Smoker
Tax -0.006
(0.005) 0.002
(0.017) 0.003
(0.013) 0.001
(0.005) Policy Index 0.001
(0.001) 0.002
(0.005) 0.005
(0.004) -0.008 (0.005)
Peer Smoking Classmates -0.735***
(0.017) 0.227
(0.405) 0.256*** (0.024)
0.252*** (0.018)
Parent Characteristics
Easy Access to Cigarettes
-0.094*** (0.001)
0.021*** (0.010)
0.031*** (0.008)
0.042*** (0.004)
Parent Smoke - 0.039*** (0.003)
0.010** (0.004)
0.020** (0.009)
0.009* (0.004)
Parent-Child Relationship Index
0.053** (0.001)
-0.009*** (0.002)
-0.030*** (0.004)
-0.014*** (0.001)
Residual 0.208*** (0.013)
-0.048*** (0.004)
-0.100*** (0.007)
-0.061*** (0.001)
Notes: Estimates of only main variables of interest provided. Include school fixed effects. Robust Standard Errors in parenthesis. *** - sig. at the 1% level; ** - sig. at the 5% level; * - sig. at the 10% level. Instruments: % of peers whose parents smoke; % of peers with easy access to cigarettes at home; % of peers who lives with both parents.
32
Appendix 1: First Stage Estimation (OLS) of Control Function Approach
Variables Nominated Peers Classmates
Tax -0.002
(0.008) -0.001 (0.001)
Policy Index -0.006 (0.005)
-0.002 (0.001)
Instruments Peer Parent Smoke 0.030***
(0.006) 0.219*** (0.008)
Peer Easy Access to Cigarettes 0.136*** (0.014)
0.294*** (0.009)
Peer Lives with both parents -0.041*** (0.012)
-0.021*** (0.006)
Parent Characteristics Lives with both parents -0.014
(0.011) 0.003
(0.002) Easy Access to Cigarettes 0.021
(0.017) 0.006
(0.002) Parent Smoke 0.026
(0.018) 0.006
(0.002) Parent-Child Relationship
Index -0.030***
(0.014) -0.001 (0.001)
Mom College 0.006 (0.013)
0.003 (0.002)
Dad College 0.004 (0.012)
0.005** (0.002)
Work Fulltime 0.028*** (0.012)
0.006 (0.004)
Demographics Age 0.011
(0.008) 0.006*** (0.001)
Male 0.020*** (0.008)
0.001 (0.002)
White 0.074*** (0.016)
0.015*** (0.002)
Black -0.090*** (0.019)
-0.036 (0.023)
Hispanic -0.011 (0.015)
-0.046 (0.026)
Other 0.021 (0.046)
0.023 (0.036)
Grade 0.023*** (0.009)
0.033*** (0.001)
Urban -0.005 (0.003)
-0.006 (0.004)
Religious -0.010 (0.004)
-0.012 (0.009)
Adjusted R2 0.085 0.407 F Statistic 31.20 690.55
N 6,549 19,988 Notes: Include school fixed effects. Robust Standard Errors in parenthesis. *** - sig. at the 1% level; ** -
sig. at the 5% level; * - sig. at the 10% level.