Do adolescents with emotional or behavioral problems respond to cigarette prices?
Tekin, Erdal ; Mocan, Naci ; Liang, Lan 等
1. Introduction
Cigarette smoking is the most preventable cause of morbidity and
mortality (McGinnis and Foege 1993; Peto et al. 1994; Mokdad et al.
2004). Medical research has established a strong link between chronic
cigarette use and lung cancer at least since the 1964 Surgeon
General's Report on Smoking and Health. Lung cancer accounts for
about 30% Of all cancer deaths, and about 85% of lung cancer deaths are
attributable to tobacco use (U.S. Department of Health and Human
Services 1989). Cigarette smoking is also one of the leading risk
factors of cardiovascular disease, which is the leading cause of death
in the United States at the turn of the 21st century. (1) Each year more
than 400,000 Americans die from cigarette smoking, which indicates that
20% of all deaths are cigarette related (Centers for Disease Control and
Prevention 2008). Youth smoking is particularly important in this
context, as the epidemiological evidence indicates that individuals who
avoid smoking in adolescence or in young adulthood are unlikely to ever
become smokers. In developed countries about 80% of adult smokers
started smoking in their teens, and across the world there is an
emergent trend toward initiation of smoking at younger ages (World Bank
1999).
Although adult smoking has been declining gradually in the United
States since the 1970s, there has been an increase in youth smoking
during the 1990s. According to the Monitoring the Future Survey (Johnson
et al. 2004), smoking rates peaked in 1996 for 8th and 10th graders,
where the rates were 21% and 30.4%, respectively. The smoking rate
peaked in 1997 for high school seniors, when more than one in three high
school seniors smoked. At these peak levels, the smoking rates were
about 50% higher for 8th and 10th graders and about 30% higher for 12th
graders in comparison to the corresponding rates that prevailed in 1991.
Even though the smoking prevalence among high school students has
declined since 1997, the newest information suggests that the rate of
decrease in smoking prevalence has declined, and nearly 25% of high
school seniors report current smoking (Johnston et al. 2004). Such
persistence in smoking rates among adolescents is noteworthy because it
pertains to a generation that has received the greatest amount of
smoking prevention messages and prevention interventions of any cohorts
in American history.
In January 2000 the Department of Health and Human Services
launched a comprehensive and nationwide health promotion agenda, known
as Healthy People 2010, which serves as a guide for improving the health
of the American population during the first decade of the 21st century.
One of the target areas highlighted in Healthy People 2010 is tobacco
use, with the goal of reducing the use of tobacco products among
adolescents to 21% by 2010 (Centers for Disease Control and Prevention
2000). Because academic research has identified a negative relationship
between cigarette prices and smoking (Lewit and Coate 1982; Chaloupka
and Grossman 1996), raising cigarette prices through the enactment of
higher cigarette excise taxes has received much attention among various
tobacco control strategies. The evidence on the extent of responsiveness
of teenagers and young adults to cigarette prices, however, is somewhat
mixed. Lewit, Coate, and Grossman (1981) find that among youths between
the ages of 12 and 17, an increase in cigarette prices has a fairly
substantial negative impact on smoking, with an elasticity of -1.44.
Although Chaloupka (1991) reports that young adults, ages 17 to 24, are
relatively insensitive to price changes, Chaloupka and Wechsler (1997)
estimate a statistically significant and substantially large price
elasticity for cigarette demand among college students. DeCicca, Kenkel,
and Mathios (2001) report that the price effect of smoking onset between
the 8th grade and the 12th grade is not significantly different from
zero. Gruber and Zinman (2001) conclude that the most effective policy
determinant of youth smoking, particularly among older teens, is the
price. Emery, White, and Pierce (2001) find that although established
adolescent smokers are responsive to price changes, experimenters are
not. (2)
Adolescents constitute an important age group to analyze from a
policy point of view because of the adverse health impacts over the life
cycle of early initiation of smoking. This age group is also an
important category to analyze in and of itself because of presumed
differences in risky behavior in comparison to adults; although, recent
research underscores their responsiveness to prices and incentives
(Levitt 1998; Gruber and Zinman 2001; Mocan and Rees 2005; Visser,
Harbaugh, and Mocan 2006).
This article focuses on a particular segment of the adolescent
population. Specifically, it investigates whether adolescents with
mental and behavioral problems, such as depression or delinquency,
respond to variations in cigarette prices. This is a question that also
has important policy implications. Studies show that adolescents with
mental health problems have much higher rates of smoking (McMahon 1999;
Saffer and Dave 2005), and this is particularly true for depression,
conduct disorder or delinquency, and attention deficit disorder. With
estimates of lifetime prevalence of depression through adolescents as
high as 20% (Rushton, Forcier, and Schectman 2002), and the strong
evidence on co-morbidity among depression, delinquency, and substance
use, it is important to develop an understanding of the efficacy of
cigarette price variations for this particular population. If
adolescents with mental health problems are not very responsive to
cigarette prices, then policy makers ought to find other ways to reduce
tobacco use among these adolescents, in addition to raising taxes. If,
on the other hand, they are responsive to cigarette prices, raising
prices through taxes and other supply reduction policies may be
considered an effective policy option.
There may be reasons for adolescents with emotional and behavioral
problems to behave differently from adults and from other adolescents
with no such problems. For example, the theory of rational addiction
(Becket and Murphy 1988) postulates that individuals maximize utility
over the life cycle by taking into account the implications of their
current actions on future utility. Specifically, utility depends on the
current consumption of the addictive good, nonaddictive good, and the
stock of past addictive consumption. The rational addict understands
that while his utility rises when he consumes more today, his long-run
utility is lower because consumption of the addictive good increases the
stock of past consumption, which has a negative marginal utility. In
this context Becket and Murphy (1988) and Becker, Grossman, and Murphy
(1991) show that price responsiveness is inversely related to time
preference. Individuals with higher discount rates are expected to be
more responsive to price in comparison with those who have low discount
rates. One testable implication is that younger and less educated
individuals are more price sensitive than others (Becker, Grossman, and
Murphy 1991). In our context this means that higher price sensitivity of
individuals with mental or behavioral problems is predicted if such
individuals have higher discount rates. In addition, these individuals
are expected to discount the future more heavily because future costs
are lower for them as their expected future wages are lower. There
exists research to indicate that individuals who engage in risky
behaviors have higher discount rates (Kirby and Petty 2004; Chesson et
al. 2006). Clinical psychology recognizes that a number of psychological
disorders, including depression, co-occur with various addictions and
risk-taking behaviors and that they involve some type of failure of
"self-regulation" (Greenbaum, Foster-Johnson, and Petrila
1996; Baumeister and Vohs 2004). The question of whether people with
emotional disorders have cognitive differences or whether they differ in
their judgments of reality is a subject of research (Dunning and Story
1991; Claypoole at al. 2007; Eisner, Johnson, and Carver 2008).
It should be noted that a rational addiction framework is not
necessary for differential price responsiveness to emerge between
individuals with and without emotional and behavioral problems. For
example, Saffer and Dave (2005) show that in a static model where
utility depends on the consumption of an addictive good, nonaddictive
good, and mental illness, the price elasticity of the addictive good can
be larger for mentally ill individuals. However, whether or not
individuals with emotional or behavioral problems react differently to
prices is an empirical question, which motivates us to analyze the
potential differences in price elasticity between individuals with and
without these problems, similar in spirit to the research that analyzes
the potential differences between males and females in their
responsiveness to cigarette prices (Cawley, Markowitz, and Tauras 2004).
Only one economic study to date has examined the interaction
between mental illness and demand for tobacco and other substances.
Saffer and Dave (2005) used the National Comorbidity Survey with
appended price data and estimated demand functions for individuals with
any lifetime mental illness. They concluded that individuals with a
history of mental illness are responsive to prices, and therefore,
higher excise taxes are effective even within this population. This is
an important study with interesting findings. However, some issues are
unexplored. Saffer and Dave (2005) use a sample that includes
individuals with ages ranging from 15 to 54. Theoretical and empirical
research suggests that youths respond to prices and policies differently
than adults (Lewit and Coate 1982; Chaloupka 1991). Therefore, a sample
of 15-to-54-year-olds does not allow for a differentiation between
youths and adults in their responsiveness to prices. Also, it is
possible that individuals with different types of psychological
disorders might respond differently to prices. To address these issues,
we use a large nationally representative sample of adolescents from
grades 7 to 12, where we are able to measure the extent of their mental
health and behavioral problems. By focusing on a younger age group and
employing measures of emotional and behavioral problems, we aim to
further our understanding of the interactions between youth smoking,
mental health, and tobacco control.
The article is organized as follows: Section 2 discusses the
empirical framework and the methodology, section 3 introduces the data
used in the analyses, the results are discussed in section 4, and
section 5 concludes.
2. Empirical Framework
Following the previous literature (Chaloupka and Wechsler 1997;
Czart et al. 2001; Gruber and Zinman 2001), we estimate two-part models
for individual cigarette demand. The participation equation estimates a
discrete choice model, where the dependent variable is dichotomous,
indicating whether the individual is a smoker. The second equation
specifies an ordinary least squares for the conditional demand by those
who are smokers. The dependent variable in the second equation is a
continuous measure of the number of cigarettes consumed per day. The
price elasticity of cigarette demand is computed as the sum of the price
elasticity of smoking participation obtained from the participation
equation and the price elasticity of conditional demand for cigarettes
obtained from the consumption equation. Since one of our goals is to
gauge the differential response to cigarette prices, we estimate the
models separately for adolescents with and without emotional or
behavioral problems. More specifically, we estimate models of the
following form:
[CS.sub.ijs] = [alpha] + [delta][P.sub.s] + [X.sub.ijs][beta] +
[[mu].subs] + [v.sub.j] + [[epsilon].sub.ijs], (1)
where [CS.sub.ijs] is a measure of the smoking indicator, or
smoking intensity of person i who lives in county j of state s.
[P.sub.s] stands for the price of cigarettes in state s; [X.sub.ijs]
represents personal and family attributes of the individual, such as
age, gender, race, and mother's education; [[mu].sub.s] stands for
unobserved state-level characteristics that may impact smoking behavior;
[v.sub.j] represents unobserved effects at the local (county) level; and
[[epsilon].sub.ijs] is a random error term.
Cigarette consumption is also likely to be influenced by the
behavior of the peers. Thus, Equation 1 can be revised as
[C.sub.ijs] = [alpha] + [delta][P.sub.s] + [X.sub.ijs][beta] +
[gamma][CPEER.sub.ijs] + [[mu].sub.s] + [v.sub.j] + [[epsilon.sub.ijs],
(2)
where [CPEER.sub.ijs] measures the extent of the smoking behavior
of person i's peers. Given that Equation 2 is applicable to the
behavior of each of these peers, it follows that
[CPEER.sub.ijs] = f ([P.sub.s], [X.sub.-ijs], [C.sub.ijs],
[[mu].sup.*.sub.s],[v.sup*.sub.j],[[epsilon].sup.*.sub.ijs]), (3)
where [X.sub.-ijs] stands for personal and family attributes of
one's peers (not including one's own), and [[mu].sup.*.sub.s]
and [v.sup.*.sub.s] are unobserved aggregate state and local effects,
respectively.
Equations 2 and 3 underline the endogeneity of the reflection
problem (Manski 1993; Sacerdote 2001), where both the individual's
own consumption and the consumption of her peers influence each other.
Furthermore, it is obvious that self-selection into a particular peer
group is endogenous, and a person's unobserved attributes that make
her more likely to be associated with a particular group of friends are
potentially correlated with her behavior. In other words, the error term
[[epsilon].sub.is] is likely to be correlated with [CPEER.sub.ijs] in
Equation 2. Studies that are primarily interested in the analysis of the
peer effects rely on random assignment of peers, such as roommates in
dormitories (Sacerdote 2001), to identify the impact of peers on
one's own behavior. In our case these complications are not as
important because our primary interest is the impact of price, and
substitution of Equation 3 into Equation 2 gives
[C.sub.ijs] = [eta] + [pi][P.sub.s] + [X.sub.ijs][PSI] +
[[lambda].sub.s] + [[omega].sub.j] + [[xi].sub.is], (4)
where the error term [???]is includes the variables pertaining to
the peers' exogenous attributes represented by X_0., in Equation 3,
but by definition, they are uncorrelated with price.
In this framework we will not attempt to identify the effect of the
peer's behavior on the price elasticity because the data set we
employ does not allow us to address either the selection or the
reflection issue. We will estimate versions of Equation 4, where the
coefficient [pi] captures the total impact of price on cigarette
consumption of the individual, which consists of multiple channels,
including the direct impact of price, and the impact of price that works
through peers' behavior. (3)
In Equation 4 the coefficient of price might be biased if
unobserved state-level determinants of smoking ([[lambda].sub.s]) are
correlated with price. Because price does not vary within a state in the
data we employ, we cannot control for these unobservables using state
fixed effects. Instead, we employ a large set of state-level variables
that gauge the sentiment toward tobacco consumption in the state. The
same concern motivated researchers to employ a state-level index as a
control variable to capture smoking sentiments in the state (Gruber and
Zinman 2001; DeCicca et al. 2006). Similarly, it may be important to
control for local-level unobservables ([[omega].sub.j]) that may be
correlated with both cigarette prices and smoking tendencies.
Consequently, empirical models also include an extensive array of county
variables that aim to gauge the socioeconomic conditions of the county
in which the individual resides. These variables are described in
section 3.
3. Data
The data used in the analysis are drawn from the first wave of the
National Longitudinal Study of Adolescent Health (Add Health). (4) Add
Health is a nationally representative study of adolescents in grades 7
through 12. An in-school questionnaire was administered to every student
who attended one of the sampled 132 U.S. schools on a particular day
during the period between September 1994 and April 1995. A random sample
of approximately 200 adolescents from each high school/feeder school
pair was selected for in-home interviews, which were conducted from
April 1995 to December 1995. (5) The in-home interviews constituted the
core sample. In addition to the core sample, several special samples
(for example, ethnic and genetic) were also drawn on the basis of
in-school interviews. The core and the special samples provide a total
number of 20,745 adolescents for Wave I. Data are gathered from
adolescents, from their parents, siblings, friends, romantic partners,
and fellow students, and from school administrators. The survey was
designed to provide detailed information on teen behavior, including
their emotional problems and substance use/abuse. The time period
covered by our data corresponds to a period of most interest for the
analyses of youth smoking because in the 1990s youth smoking took a
dramatic upswing (Gruber and Zinman 2001).
Survey administrators took several steps to maintain data security
and to minimize the potential for interviewer or parental influence.
First, respondents were not provided with any printed questionnaires.
Rather, all data were recorded on laptop computers. Second, for
sensitive topics, such as delinquent behavior and substance use/abuse,
the adolescents listened to prerecorded questions through earphones and
entered their answers directly on the laptops.
More detailed description of these data can be found in Mocan and
Tekin (2005, 2006).
Smoking Measures
Measures of smoking for each individual are created based on the
following question posed to each respondent: "During the past 30
days, on how many days did you smoke cigarettes?" We create a
binary indicator for current smoking, which takes on the value of 1 for
those who reported to have smoked at least one day during the past 30
days and a value of 0 otherwise. For these smokers the survey includes
information on how many cigarettes they smoked each day, which is used
to calculate the average daily number of cigarettes. We use the binary
indicator of smoking participation and the average daily number of
cigarettes smoked as two measures of smoking throughout our analysis.
Emotional and Behavioral Problems
Resnick et al. (1997) developed continuous measures of emotional
and behavioral problems for the Add Health data. Emotional distress is
measured using a 17-item depression scale. Fourteen of these questions
collect information on how frequently the respondents felt the following
negative emotions during the past seven days: being bothered by things
that usually do not bother them, having poor appetite, feeling blue,
having trouble concentrating, feeling depressed, feeling like a failure,
feeling fearful, being less talkative than usual, feeling lonely,
feeling people being unfriendly toward them, feeling sad, feeling
disliked by others, feeling hard to get started doing things, and
feeling life not worth living. The possible responses were 0 = never or
rarely, 1 = sometimes, 2 = a lot of times, and 3 = most of the time or
all of the time.
The additional three questions consider the frequency of the
following negative emotional states in the last 12 months: having
trouble relaxing, feeling moody, and crying frequently. The possible
responses to these questions were 0 = never, 1 = just a few times, 2 =
about 1 a week, 3 = almost every day, and 4 = every day. The measure of
emotional problems is obtained by summing up the individual's
response to each of these questions. The range of the variable is from 0
to 5 I. Behavioral problems are assessed by an 11-item delinquency
scale, also based on self-reported behaviors of the Add Health
respondents. These items include the frequency of the following
behaviors in the past 12 months: damaging property, lying to parents,
committing theft, getting into serious fights, running away from home,
driving a car without permission, committing robbery, selling illicit
drugs, skipping school, ever having been suspended, and being expelled
from school. Respondents' choices pertaining to the first eight of
these items include never, 1 or 2 times, 3 or 4 times, and 5 or more
times. The last three responses are binary. Each respondent is evaluated
on this behavioral problems scale by adding up the frequency of these
acts. The range of the variable in the sample is from 0 to 28.
From these continuous scales, we identify an individual as having
emotional (behavioral) problems if he or she is ranked in the top
quartile of the distribution of emotional (behavioral) problems scale.
Previous research suggests that this classification is correlated with
different stages of smoking uptake for the Add Health respondents
(Lloyd-Richardson et al. 2002). (6) The definitions and descriptive
statistics of these variables and personal and family background
characteristics are reported in Table 1. The sample means indicate that
smoking rates are significantly higher among adolescents with emotional
and behavioral problems (those who are in the top 25th percentile) in
comparison to those without these problems. The average number of
cigarettes smoked in a day in the previous month is also statistically
significantly higher for those in the upper quartile of the emotional
problems scale and the behavioral problems scale. The difference is
large in magnitude in the case of behavioral problems. For example,
individuals who are in the top 25% of the distribution for emotional
problems are almost three times more likely to smoke, and they smoke
about two more cigarettes per day.
To investigate the sensitivity of the results to the measurement of
the variables we implement the following strategy. First, instead of the
75th percentile, we use the 90th percentile as the cutoff to classify
individuals into mental or behavioral problem groups. Second, we
construct the emotional and behavioral problems scales in alternative
ways by assigning different weights to the items included. Specifically,
we classify a person as being depressed if he or she reported having
ever felt depressed, ever felt that life was not worth living in the
past week, or cried almost every day in the past 12 months, regardless
of his or her responses to the other questions. In the case of the
behavioral problems, we recalculate the index using a weighted sum,
where lying to parents, running away from home, driving a car without
permission, skipping school, and being suspended from school receive a
weight of one, property damage and getting into a serious fight receive
a weight of two, theft and selling drugs are assigned a weight of three,
and committing robbery receives a weight of four. Thus, we consider
certain behaviors as more serious than others in creating the behavioral
problems scale. (7)
Price and Other Explanatory Variables
Cigarette price data were collected by the Tobacco Institute before
it was dissolved. The consulting firm Orzechowski & Walker continued
to collect and publish this information in the annual volume of Tax
Burden on Tobacco. This publication provides cigarette price information
for all 50 states and the District of Columbia. There are two prices
reported for each state: average price over all brand name cigarettes
and average price over all cigarettes (including generics). Following
the literature, we use the price for the brand name cigarettes. Because
the published price pertains to November 1, a weighted average price for
the calendar year is computed. (8) Cigarette prices are deflated by a
cost-of-living index to account for geographic price variations.
We control for a rich set of socioeconomic and demographic
variables, including age, gender, race and ethnicity, religiosity,
parental education, mother's work status, allowance and earned
income of the adolescent, marital status of parents, total family
income, and whether either parent smokes. In addition to these
variables, the models include the body mass index of the adolescent and
self-reported health status. As the previous literature suggests, these
measures might be correlated with youth smoking (Cawley, Markowitz, and
Tauras 2004).
As mentioned in section 2, the average price of cigarettes in the
state might be correlated with state-level sentiments toward smoking. If
these sentiments also influence individuals' smoking behavior, the
estimated effect of cigarette price would be biased. To guard against
obtaining biased estimates of the price effect, we add eight variables
that aim to control for sentiment toward smoking in the state. They are
(i) the death rate due to smoking per 100,000 people; (ii) a dichotomous
variable to indicate if vending machines are banned from locations
accessible to youth but allowed only in businesses holding liquor
licenses; (iii) a dichotomous variable to indicate if tobacco marketing
is prohibited on billboards within 500 feet of schools and/or churches;
(iv) a dichotomous variable to indicate tobacco marketing is restricted,
such that free samples are prohibited through mail and prohibition on
free samples within 500 feet of schools; (v) a dichotomous variable to
indicate existence of a program enforcing citations or fines in state
for violations; (vi) the number of full-time-equivalent staff on tobacco
control per 100,000 people; (vii) total funds for tobacco control per
100,000 youth ages 17 and younger; and (viii) a dichotomous variable to
indicate whether the state offers in-service training on tobacco use
prevention to school health service staff.
We also control for the impact of contextual variables, such as
local unemployment rate, population density, and a dichotomous variable
to indicate if the person lives in an urban area. Also included are the
proportion of blacks in the county of residence in 1990, the proportion
of Hispanics in the county in 1990, the median household income in the
county in 1990, the proportion voting democratic in the 1992
presidential election in the county, and the proportion voting for Ross
Perot in the 1992 presidential election.
Table 1 demonstrates that personal and family characteristics
differ significantly between adolescents with emotional and behavioral
problems and those without these problems. For example, males are more
likely to have behavioral problems in comparison to females; whereas,
females are more likely to have emotional problems. The proportion of
blacks is higher in the group with emotional problems in comparison to
the full sample. The opposite is true for whites. Adolescents whose
parents are smokers are more likely to have both emotional and
behavioral problems.
4. Results
Table 2 displays the estimated coefficients from the binary
participation equation for adolescents with and without emotional
problems. Those with (without) emotional problems are the ones who are
ranked in the top 25% (bottom 75%) of the distribution of the emotional
problems scale in the sample. The results based on the 90%/10% cutoff
are discussed later. The reported coefficients are the marginal
probabilities obtained from probit regressions. According to the results
displayed in Table 2, a $1 increase in the real price of cigarettes
generates about a 3 percentage point decrease in the probability of
smoking for adolescents with no emotional problems, but the impact is
not statistically significant. For those with emotional problems (those
who are in the 75th percentile of the depression scale or higher), the
impact is 15.5 percentage points, about a 48% reduction from the sample
mean. The participation elasticity for those with emotional problems is
-1.04, as compared to the -0.33 for those without emotional problems.
For both groups, having a smoking parent is associated with an
increased likelihood of smoking. Having a married parent has a negative
impact on the propensity to smoke for both groups. Earned income
(Salary) of the adolescent has a small but positive impact on the
probability to smoke. This could be due to an income effect, or it could
be due to exposure to adults and older peers in a work environment with
smoking propensity. Interestingly, higher allowance is associated with
increased smoking for adolescents without emotional problems, while it
has a negative effect on the smoking propensity of those with emotional
problems. However, both effects are small in magnitude. Males with no
emotional problems are more likely to smoke than their female
counterparts, while there is no statistically significant difference
between the smoking propensity between males and females with emotional
problems. Adolescents with strong religious beliefs have a lower
propensity to smoke. Physically healthier adolescents have a lower
propensity to smoke regardless of their mental health status.
Table 3 presents the results of participation equations based on
the scores of the delinquency scale. As before, the adolescents are
classified into groups with and without behavioral problems based on the
75th percentile cutoff. The results are similar to those displayed in
Table 2. More specifically, adolescents with behavioral problems respond
to a $1 increase in the price of cigarettes by reducing their smoking
propensity by 19 percentage points. This represents about a 43% decrease
from the sample mean. The price elasticity for smoking participation for
those respondents with behavioral problems is -0.93, which is in sharp
contrast to the one obtained from the sample without behavioral problems
that is close to 0.
Tables 4 and 5 display the estimates from the conditional demand
equations for those with and without emotional problems and for those
with and without behavioral problems, respectively. As displayed in
Table 4, among adolescents with emotional problems, a $1 increase in the
real price of cigarettes reduces the number of cigarettes smoked per day
by 1.6; although, the coefficient is not statistically significant. For
those without emotional problems, the reduction is 2.5 cigarettes per
day. The corresponding price elasticities of conditional demand for
those with and without emotional problems are -0.53 and -0.89,
respectively. Table 5 shows that a $1 increase in the real price of
cigarettes decreases the average daily number of cigarettes by 2.2 among
those in the top 25% of the behavioral problems scale. The coefficient
is borderline significant with a p-value of 0.11. The corresponding
value for those who are in the bottom 75% of the behavioral scale is
about 1.7 cigarettes. The implied price elasticities of conditional
demand for those with and without behavioral problems are -0.65 and
-0.67, respectively.
The overall price elasticity of smoking for those adolescents with
and without emotional problems (calculated by adding up the elasticities
obtained from Tables 2 and 4) are -1.57 and -1.22, respectively. The
overall price elasticity of smoking for adolescents with behavioral
problems is -1.58 and is -0.72 for those without behavioral problems.
The price coefficients are not estimated with precision in some cases,
but the price elasticities of smoking obtained from these models are
consistent with those documented in other studies for this age,
suggesting that higher cigarette prices are deterrents for youth smoking
for all adolescents. (9,10)
The variables that are included to control for state-level
sentiment towards smoking impact participation, as well as consumption
behavior, in every regression. The following variables are individually
significant in most regressions: death rate due to smoking, the
indicator variable to signify whether tobacco marketing is prohibited on
billboards within 500 feet of schools or churches, the indicator
variable for restricted tobacco marketing, and per capita
full-time-equivalent staff on tobacco control. Furthermore the eight
tobacco sentiment variables are extremely significant determinants of
behavior as a group with p-values ranging from 0.03 to 0.00. Therefore,
these variables seem to represent smoking sentiment at the state-level
reasonably well. Some county-level variables, such as county median
household income, local unemployment rate, proportion Hispanic,
proportion black, proportion that voted Democrat, and population
density, were also significant in most regressions.
Robustness
We estimate the same models displayed in Tables 2-5 with one
modification in the classification scheme. Specifically, we consider an
individual as having emotional or behavioral problems if he or she
ranked in the top 10% of the relevant distribution. The results, which
are not reported in the interest of space, were very similar to those
displayed in Tables 2-5. The top panel of Table 6 displays the
calculated participation, consumption, and total price elasticities
obtained from both specifications (the 75th percentile cutoff, as well
as the 90th percentile cutoff). The bottom panel of Table 6 summarizes
the results obtained from estimating the same set of specifications
using the alternative measure of emotional and behavioral problems as
described in section 2. Sixteen elasticity comparisons can be made
between individuals with and without emotional and behavioral problems
based on participation and consumption equations. In nine of these
comparisons the participation or consumptions elasticities are larger
for those with emotional or behavioral problems. The magnitudes of the
elasticities are not very different between the two groups. The median
total elasticity for those with emotional or behavioral problems is
-1.58 and is -1.09 for those without emotional or behavioral problems.
These values are very similar to the widely cited total smoking
elasticity of- 1.31, reported by Chaloupka and Grossman (1996), as well
as the elasticity of -1.44, reported by Lewit, Coate, and Grossman
(1981).
To test more formally whether the participation and conditional
demand elasticities are statistically different between those with and
without emotional or behavioral problems, we estimate models by pooling
the data and adding dichotomous variables for emotional or behavioral
problems and an interaction term between the problem indicator dummy and
cigarette price. In 12 out of 16 cases the interaction terms were
negative, indicating that the point estimate for the price effect for
those with emotional or behavioral problems was larger in absolute value
in comparison to those with no such problems. This is consistent with
the elasticities summarized in Table 6. However, the coefficient of the
interaction term was statistically significant only in the participation
equation for emotional problems when the 75% cutoff was employed to
identify emotional problems.
Although peer effects cannot be identified properly in this study
because of data limitations, it is interesting to investigate how
controlling for endogenous peer effects would impact the results. (11)
The data set includes a question about the smoking behavior of friends.
Specifically, the question asks how many of the three best friends smoke
at least one cigarette a day. The answers to this question are used as a
measure of peer smoking and are included as an additional regressor.
Note that a response of zero means either none of the respondent's
three best friends are smokers or the respondent does not know a person
who qualifies as a best friend. The coefficient of the peer smoking
variable was positive and statistically highly significant in every
participation and consumption regression. Overall, the estimated price
coefficients in participation and consumption equations were similar to
those reported in Tables 2-5. The biggest difference was seen in the
consumption equations equivalent to Table 5, where the estimated price
coefficients were -2.36 (p = 0.101) and -1.33 (p = 0.104) for those with
and without behavioral problems, respectively, in models with peer
effects (the corresponding coefficients are -2.20 [p = 0.11], and -1.66
[p = 0.039] in models without peer effects, reported in Table 5).
Consequently, the elasticities obtained from the models with peer
effects were very similar to those reported in Table 6.
Clark and Loheac (2007) report that the structure of the peer
effects is complicated (for example, the impact of the prevalence of
school-level behavior is nonlinear, and age and gender-specific effects
are dissimilar). They also find that peer effects are stronger in the
case of alcohol than cigarettes, which is consistent with the results
reported in this article regarding the relative stability of the price
effect in inclusion/exclusion of peers' consumption. Although
peers' consumption is an endogenous variable, the fact that the
price elasticities based on Equation 4 are very similar to the ones that
are obtained from estimating Equation 2 may suggest that there are no
differential peer effects regarding the reaction to a change in the
price of cigarettes between adolescents with emotional or behavioral
problems and those without such problems. Of course, we cannot rule out
the possibility that the measures intended to represent the peer effects
in our model do not appropriately capture these effects due to data
limitations.
5. Conclusion
There exists a considerable literature on the impact of prices on
cigarette consumption. A different line of recent research on youth
risky behavior demonstrates that, similar to adults, adolescents respond
to prices and incentives. However, only limited research exists on the
behavior of adolescents with emotional or behavioral problems. In this
article we investigate whether cigarette consumption of adolescents with
emotional or behavioral problems responds to cigarette prices. The issue
is important from an academic, as well as public policy, point of view.
Whether adolescents behave rationally and respond to prices and
incentives in the domain of risky behavior is an important question for
economists to model and explain human behavior. From a policy
perspective, the extent of risky behavior of adolescents is critically
important for their future well-being and the potential financial
burdens imposed on society. In the case of smoking, it is well known
that smoking as a youth is strongly correlated with smoking as an adult
(Gruber and Zinman 2001). Given that smoking-related illness is the
leading preventable cause of death in the United States, and that
smokers on average live about six years less than those who never smoked
(Cutler et al. 2002), it is important to explore potential mechanisms
that may lead to a change in youth smoking behavior.
In this article we analyze an understudied and particularly
vulnerable segment of the population: adolescents with mental health and
behavioral problems. Using a nationally representative data set of
adolescents, we estimate standard two-part models, controlling for
personal and family attributes, local (county) characteristics, and a
large set of variables that aim to control for the sentiment toward
smoking at the state level. Sentiment variables explain half of the
variation in cigarette prices; and controlling for price, local area
characteristics, and smoking sentiment variables influences both
participation and consumption behaviors. Nevertheless, in the absence of
a credible natural experiment that would exogenously move the cigarette
price, one can never be completely certain of the unbiasedness of the
estimated price effect.
Our results show that adolescents with emotional or behavioral
issues, a group that is generally thought to be the less rational,
respond to cigarette prices. Furthermore, the estimated participation
and consumption elasticities are usually larger in absolute value in
comparison to those adolescents with no emotional or behavioral
problems; although, the difference is generally not statistically
significant. The estimated elasticities are similar to those obtained by
previous research using data from the general adolescent population.
Thus, the results underscore the significance of prices as a potential
device to modify adolescent behavior, including adolescents with
emotional or behavioral problems.
Received March 2007; accepted August 2008.
References
Baumeister, Roy, and Kathleen Vohs, eds. 2004. Handbook of
self-regulation: Research. theory, and applications. New York: Guilford
Press.
Becker, Gary, Michael Grossman, and Kevin Murphy. 1991. Rational
addiction and the effect of price on consumption. American Economic
Review 81:237-41.
Becker, Gary, Michael Grossman, and Kevin Murphy. 1994. An
empirical analysis of cigarette addiction. American Economic Review
84:396-418.
Becker, Gary, and Kevin Murphy. 1988. A theory of rational
addiction. Journal of Political Economy 96:675-700.
Cawley, John, Sara Markowitz, and John Tauras. 2004. Lighting up
and slimming down: The effects of body weight and cigarette prices on
adolescent smoking initiation. Journal of Health Economics 23:293-311.
Centers for Disease Control and Prevention. 2000. Tobacco use. In
Healthy People 2010: Understanding and Improving Health. 2000. 2nd
edition. Washington, DC: U.S. Government Printing Office. Accessed June
2006. Available at http://www.healthypeople.gov/Document/pdf/Volume2/27Tobacco.pdf.
Centers for Disease Control and Prevention. 2008. Health effects of
cigarette smoking. Accessed March 2009. Available at http://www.cdc.gov/
tobacco/data-statistics/fact-sheets/health_effects/health_effects.htm.
Chaloupka, Frank J. 1991. Rational addictive behavior and cigarette
smoking. Journal of Political Economy 99:722-42.
Chaloupka, Frank J., and Michael P. Grossman. 1996. Tobacco control
policies and youth smoking. NBER Working Paper No. 5740.
Chaloupka, Frank J., and Kenneth E. Warner. 2000. The economics of
smoking. In Handbook of health economics, edited by A. J. Culyer and J.
P. Newhouse. Amsterdam: Elsevier, pp. 153-1627.
Chaloupka, Frank J., and Henry Wechsler. 1997. Price, tobacco
control policies and smoking among young adults. Journal of Health
Economics 16:359-73.
Chesson, Harold, Jami Leichliter, Gregory Zimet, Susan Rosenthal,
David Bernstein, and Kenneth Fife. 2006. Discount rates and risky sexual
behaviors among teenagers and young adults. Journal of Risk and
Uncertainty 32:217-30.
Clark, Andrew, and Youenn Loheac. 2007. "It wasn't me, it
was them!" Social influence in risky behavior by adolescents.
Journal of Health Economics 26:763-84.
Claypoole, Keith, Carolyn Noonan, Roderick Mahurin, Jack Goldberg,
Tom Erickson, and Dedra Buchwald. 2007. A twin study of cognitive
function in chronic fatigue syndrome: The effects of sudden illness
onset. Neuropsychology 21:507-13.
Cutler, David, Jonathan Gruber, Raymond S. Hartman, Mary Beth
Landrum, Joseph P. Newhouse, and Meredith B.
Rosenthal. 2002. The economic impacts of the tobacco settlement.
Journal of Policy Analysis and Management 21(1):1-19.
Czart, Christina, Rosalie Liccardo Pacula, Frank Chaloupka, and
Henry Wechsler. 2001. The impact of prices and control policies on
cigarette smoking among college students. Contemporary Economic Policy
19(2):135-49.
DeCicca, Philip, Donald Kenkel, and Alan Mathios. 2001. Putting out
the fires: Will higher taxes reduce youth smoking?
Journal of Political Economy 116:1261-1303.
DeCicca, Philip, Donald Kenkel, Alan Mathios, Yoon-Jeong Shin, and
Jae-Young Lira. 2006. Youth smoking, cigarette prices, and anti-smoking
sentiment. NBER Working Paper No. 12458.
Dunning, David, and Amber Story. 1991. Depression, realism, and the
overconfidence effect: Are the sadder wiser when predicting future
actions and events? Journal of Personality and Social Psychology
61:521-32.
Eisner, Lori, Sheri Johnson, and Charles Carver. 2008. Cognitive
responses to failure and success relate uniquely bipolar depression
versus mania. Journal of Abnormal Psychology 117:154-63.
Emery, Sherry, Martha White, and John Pierce. 2001. Does cigarette
price influence adolescent experimentation? Journal of Health Economics
20:261-70.
Greenbaum, Paul, Lynn Foster-Johnson, and Amelia Petrila. 1996.
Co-occurring addictive and mental disorders among adolescents:
Prevalence research and future directions. American Journal of
Orthopsychiatry 66(1):52-60.
Gruber, Jonathan, and Jonathan Zinman. 2001. Youth smoking in the
United States: Evidence and implications. In Risky behavior among youth,
edited by Jonathan Gruber. Chicago: University of Chicago Press, pp.
69-120.
Johnston, L. D., P. M. O'Malley, J. G. Bachman, and J. E.
Schulenberg. 2004. Monitoring the future, national results on adolescent
drug use: Overview of key findings, 2004. NIH Publication No. 06-5882.
Bethesda, MD: National Institute on Drug Abuse.
Kirby, Kris, and Nancy Petry. 2004. Heroin and cocaine abusers have
higher discount rates for delayed rewards than alcoholics or
non-drug-using controls. Addiction 99:461-71.
Levitt, Steven. 1998. Juvenile crime and punishment. Journal of
Political Economy 106:1156-85.
Lewit, Eugene, and Douglas Coate. 1982. The potential for using
excise taxes to reduce smoking. Journal of Health Economics 1:121-45.
Lewit, Eugene, Douglas Coate, and Michael Grossman. 1981. The
effects of government regulation on teenage smoking. Journal of Law and
Economics 4:545-69.
Lloyd-Richardson, Elizabeth, George Papandonatos, Alessandra
Kazura, Cassandra Stanton, and Raymond Niaura. 2002. Differentiating
stages of smoking intensity among adolescents: Stage-specific
psychological and social influences. Journal of Consulting and Clinical
Psychology 70:998-1009.
Manski, Charles. 1993. Identification and endogenous social
effects: The reflection problem. Review of Economic Studies
60(3):532-42.
Mcginnis, J. Michael, and William Foege. 1993. Actual causes of
death in the United States. Journal of American Medical Association
270:2207-12.
McMahon, Robert. 1999. Child and adolescent psychopathology as risk
factors for subsequent tobacco use. Nicotine & Tobacco Research
1:45-50.
Miller, Laurie, Gail Wasserman, Richard Neugebauer, Deborah
Gorman-Smith, and Dimitra Kamboukos. 1999. Witnessed community violence
and antisocial behavior in high-risk urban boys. Journal of Clinical
Child Psychology 28:2-11.
Mocan, Naci, and Daniel Rees. 2005. Economic conditions, deterrence
and juvenile crime: Evidence from micro data. American Law and Economics
Review 7:319-49.
Mocan, Naci, and Erdal Tekin. 2005. Drug use and juvenile crime:
Evidence from a panel of siblings and twins. In Substance use:
Individual behavior, social interactions, markets, and politics, edited
by Bjorn Lindgren and Michael Grossman. Amsterdam: Elsevier, pp. 91-120.
Mocan, Naci, and Erdal Tekin. 2006. Guns and juvenile crime.
Journal of Law and Economics 49(2):507-31.
Mokdad All, James Marks, Donna Stroup, and Julie Gerberding. 2004.
Actual causes of death in the United States, 2000. Journal of the
American Medical Association 291:1238-45.
Peto, Richard, Alan Lopez, Jillian Boreham, Michael Thun, and Clark
Heath. 1994. Mortality from smoking in developed countries, 1950-2000.
Oxford, UK: Oxford University Press.
Resnick, Michael, Peter Bearman, and Robert Blum, et al. (1997).
Protecting adolescents from harm: Findings from the National
Longitudinal Study on Adolescent Health. Journal of the American Medical
Association 278:823-32.
Rushton, Jerry, Michelle Forcier, and Robin Schectman. 2002.
Epidemiology of depressive symptoms in the National Longitudinal Study
of Adolescent Health. Journal of the American Academy of ChiM and
Adolescent Psychiatry 41:199-205.
Sacerdote, Bruce. 2001. Peer effects with random assignment:
Results from Dartmouth roommates. Quarterly Journal of Economics
116:681-704.
Saffer, Henry, and Dhaval Dave. 2005. Mental illness and the demand
for alcohol, cocaine, and cigarettes. Economic Inquiry 43:229-46.
U.S. Department of Health and Human Services. 1989. Reducing the
health consequences of smoking." 25 years of progress. A report of
the Surgeon General. Rockville, MD: U.S. Department of Health and Human
services, Public Health Service Centers for Disease Control and
Prevention, Centers for Chronic Disease Prevention and Health Promotion,
Office on Smoking and Health.
Visser, Michael, William Harbaugh, and Naci Mocan. 2006. An
experimental test of criminal behavior among juveniles and young adults.
NBER Working Paper No. 12507.
World Bank. 1999. Curbing the epidemic: Governments and the
economics of tobacco control. Tobacco Control 8:196-201.
(1) Cigarette smoking triples the risk of dying from heart disease
(Centers for Disease Control and Prevention 2008).
(2) Additional references can be found in the detailed survey paper
by Chaloupka and Warner (2000).
(3) For example, a person's cigarette consumption may decrease
in response to a rise in cigarette prices. But cigarette consumption may
also decrease because the person may have fewer opportunities to obtain
cigarettes from a peer as the peer's quantity demanded goes down in
response to rising cigarette prices. Rather than identifying these
channels separately, our empirical model yields the overall effect of
cigarette prices on one's cigarette consumption.
(4) Add Health is 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 to 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, 123W. Franklin Street, Chapel Hill, NC 27516-2524, USA
(
[email protected]).
(5) Participating high schools were asked to identify junior high
or middle schools that were expected to provide at least five students
to the entering class of the high school. These schools are called
feeder schools. Their probability of selection was proportional to the
percentage of the high school's entering class that came from that
feeder.
(6) Lloyd-Richardson et al. (2002) find that for the Add Health
respondents, being in a higher quartile of the depression and
delinquency scales is associated with higher odds of being an
experimental smoker as compared to being a nonsmoker, being an
intermittent smoker as compared to being a nonsmoker or being a
experimental smoker, and being a regular smoker as compared to being in
all the earlier stages of smoking uptake.
(7) In the case of behavioral problems, there are other potential
ways individuals can be classified. For example, we employed a third
classification scheme, where an adolescent was classified as being
delinquent if he or she ever committed theft or burglary or sold illicit
drugs, regardless of his or her responses to the other questions.
Results from this classification were consistent with the ones reported
based on other classifications of behavioral problems.
(8) The average price for the calendar year is computed by
subtracting state and federal excise taxes from the current year's
price and the previous year's price and weighting the pre-tax
prices accordingly (4/12 previous year and 8/12 current year). We then
add the current federal and state excise tax back to the average pre-tax
price calculated above.
(9) See Chaloupka and Warner (2000) for a review of some of the
relevant literature.
(10) One potential caveat is that in the sample of adolescents with
behavioral problems, although the coefficient of cigarette price was
always highly statistically significant, it was not always significant
in the consumption equation at conventional significance levels.
(11) This is because we do not have random or quasi-random
assignment of peers, nor do we have a plausible instrument for peer
smoking that can be excluded from consumption equations.
Erdal Tekin, Georgia State University and NBER, Andrew Young School
of Policy Studies, Georgia State University, P.O. Box 3992, Atlanta, GA
30302-3992, USA; E-mail
[email protected]; corresponding author.
Naci Mocan, Department of Economics, Louisiana State University,
2119 Taylor Hall, Baton Rouge, LA 70803-6306, USA; E-mail
[email protected].
Lan Liang, Agency for Healthcare Research and Quality, Center for
Financing, Access, and Cost Trends Agency for Healthcare Research and
Quality 540 Gaither Road, Room 5236 Rockville, MD 20850, USA; E-mail
[email protected].
This research was initially conducted when Lan Liang was at the
University of Illinois at Chicago and was supported in part by a grant
from the Robert Wood Johnson Foundation Substance Use Policy Research
Program to the University of Illinois at Chicago (Grant No. 048567). We
thank participants of the 2005 American Economic Association Meetings
and the 5th International Health Economics Association World Congress
for useful suggestions. The views expressed in this article are those of
the authors. No official endorsement by the U.S. Department of Health
and Human Services, the Agency for Healthcare Research and Quality, or
the National Bureau of Economic Research is intended or should be
inferred. We thank Paul Mahler, Emre Unlu, and Duha Altindag for
research assistance and Michael Grossman for helpful comments.
Table 1. Definitions and Descriptive Statistics
Variable Definition Full Sample
Smoker = 1 if smoked during past 30 0.228 (0.420)
days, = 0 otherwise
Smokenum * = Number of cigarettes usually 6.303 (7.588)
smoked per day during past 30
days (if greater than zero)
Behav. Problem = Behavioral problems scale 3.141
Emot. Problem = Emotional problems scale 8.591
Price * = State-specific average price of 2.204 (0.229)
cigarettes adjusted by cost of
living index
Male =1 if male, = 0 otherwise 0.496 (0.500)
White (b) =1 if white, = 0 otherwise 0.624 (0.484)
Black =1 if black, = 0 otherwise 0.216 (0.411)
Hispanic =1 if Hispanic ethnicity, = 0 0.159 (0.365)
otherwise
BMI = Body mass index (weight/height- 22.478 (4.381)
squared)
Healthl (c) = 1 if better than good health 0.703 (0.457)
Health2 = 1 if good health, = 0 otherwise 0.239 (0.427)
Age = Age in years 15.507 (1.692)
Religion =1 if adheres to any religion, = 0.888 (0.315)
0 otherwise
Allowance = Allowance per week adjusted by 7.870 (11.922)
cost of living index
Salary = Earned income of the adolescent 44.313 (76.987)
per week adjusted by cost of
living index
Mothered1 (d) = 1 if mother's education is less 0.165 (0.371)
than high school, = 0 otherwise
Mothered2 = 1 if mother's education up to 0.257 (0.437)
high school, = 0 otherwise
Mothered3 = 1 if mother's education is up 0.037 (0.188)
to GED, = 0 otherwise
Mothered4 = 1 if mother's education is 0.302 (0.459)
between GED and college, =
0 otherwise
Pmarried = 1 if parents married, = 0.749 (0.433)
0 otherwise
Psmoke = 1 if any of the parents smoke, 0.731
= 0 otherwise
Workmom = 1 if mother works, = 0.773 (0.419)
0 otherwise
Number of obs. 13,399
Emotional No Emotional
Variable Problems (a) Problems
Smoker 0.325 *** (0.468) 0.194 (0.395)
Smokenum * 6.742 *** (8.135) 6.045 (7.237)
Behav. Problem 4.487 *** 2.669
Emot. Problem 17.531 *** 5.452
Price * 2.203 (0.230) 2.208 (0.227)
Male 0.373 *** (0.484) 0.539 (0.499)
White (b) 0.570 *** (0.495) 0.643 (0.479)
Black 0.236 *** (0.425) 0.209 (0.406)
Hispanic 0.184 *** (0.387) 0.150 (0.357)
BMI 22.855 *** (4.519) 22.346 (4.325)
Healthl (c) 0.562 *** (0.496) 0.752 (0.432)
Health2 0.325 *** (0.469) 0.209 (0.407)
Age 15.772 *** (1.635) 15.414 (1.703)
Religion 0.879 ** (0.327) 0.892 (0.311)
Allowance 8.431 *** (12.879) 7.673 (11.562)
Salary 50.088 *** (83.830) 42.285 (74.335)
Mothered1 (d) 0.213 *** (0.409) 0.149 (0.356)
Mothered2 0.257 (0.437) 0.257 (0.437)
Mothered3 0.044 ** (0.206) 0.034 (0.181)
Mothered4 0.292 (0.455) 0.305 (0.460)
Pmarried 0.702 *** (0.458) 0.766 (0.423)
Psmoke 0.786 *** 0.712
Workmom 0.754 *** (0.430) 0.780 (0.414)
Number of obs. 3482 9917
Behavioral No Behavioral
Variable Problems (a) Problems
Smoker 0.439 *** (0.496) 0.158 (0.365)
Smokenum * 7.345 *** (8.092) 5.350 (6.963)
Behav. Problem 7.916 *** 1.563
Emot. Problem 11.416 *** 7.657
Price * 2.205 (0.228) 2.202 (0.232)
Male 0.603 *** (0.489) 0.461 (0.498)
White (b) 0.575 *** (0.494) 0.641 (0.480)
Black 0.224 (0.417) 0.213 (0.410)
Hispanic 0.202 *** (0.401) 0.144 (0.351)
BMI 22.826 *** (4.315) 22.363 (4.398)
Healthl (c) 0.619 *** (0.486) 0.730 (0.444)
Health2 0.295 *** (0.456) 0.221 (0.415)
Age 15.774 *** (1.586) 15.418 (1.718)
Religion 0.854 *** (0.353) 0.900 (0.301)
Allowance 8.122 (12.607) 7.787 (11.687)
Salary 54.996 *** (85.160) 40.781 (73.751)
Mothered1 (d) 0.198 *** (0.399) 0.154 (0.361)
Mothered2 0.236 *** (0.425) 0.264 (0.441)
Mothered3 0.043 ** (0.202) 0.035 (0.183)
Mothered4 0.321 *** (0.467) 0.295 (0.456)
Pmarried 0.683 *** (0.465) 0.771 (0.420)
Psmoke 0.813 *** 0.704
Workmom 0.778 (0.415) 0.772 (0.420)
Number of obs. 3329 10,070
Standard deviations are in parentheses. ***, ***, and * indicate that
the difference between the two groups is statistically significant
difference between columns 2 and 3, and 4 and 5, at the 0.01, 0.05,
and 0.10 levels, respectively.
(a) "Emotional problems" and "Behavioral problems" denote those
observations falling into the top 25% of the distribution of emotional
problems or behavioral problems scale.
(b) Omitted category is other race.
(c) Omitted category is less than good health.
(d) Omitted category is more than college.
Table 2. Marginal Effects from the Participation Equation
(Emotional Problems)
Without Emotional Problems (Bottom 75%)
Variables Coefficient Standard Error
Price -0.030 (0.038)
Salary 0.0003 *** (0.00005)
Allowance 0.001 *** (0.0003)
BMI -0.0004 (0.001)
Psmoke 0.083 *** (0.008)
Healthl -0.117 *** (0.020)
Healthl -0.038 *** (0.013)
Age 0.273 *** (0.038)
Age2 -0.008 *** (0.001)
Male 0.020 * (0.010)
Hispanic -0.016 (0.013)
White 0.037 ** (0.015)
Black -0.104 *** (0.013)
Religion -0.045 *** (0.011)
Pmarried -0.026 ** (0.011)
Motheredl 0.005 (0.011)
Motheredl -0.004 (0.009)
Mothered3 0.023 (0.021)
Mothered4 0.024 *** (0.009)
Workmom -0.006 (0.009)
County characteristics (a) Yes
State-level smoking Yes
sentiment variables (b)
N 9917
With Emotional Problems (Top 25%)
Variables Coefficient Standard Error
Price -0.155 *** (0.062)
Salary 0.000 3*** (0.0001)
Allowance -0.002 *** (0.001)
BMI -0.002 (0.002)
Psmoke 0.075 *** (0.019)
Healthl -0.153 *** (0.024)
Healthl -0.052 * (0.026)
Age 0.414 *** (0.075)
Age2 -0.013 *** (0.002)
Male 0.018 (0.020)
Hispanic -0.004 (0.027)
White -0.006 (0.022)
Black -0.228 *** (0.017)
Religion -0.055 ** (0.023)
Pmarried -0.046 *** (0.017)
Motheredl -0.014 (0.029)
Motheredl -0.003 (0.021)
Mothered3 0.085 * (0.050)
Mothered4 0.031 (0.024)
Workmom 0.014 (0.022)
County characteristics (a) Yes
State-level smoking Yes
sentiment variables (b)
N 3482
Robust standard errors, clustered at the state-level, are in
parentheses. ***, **, and 'indicate statistical significance at the 1%,
5%, and l0% levels, respectively.
(a) These include the county-level variables as described in section 3.
(b) These include the state-level variables as described in section 3.
Table 3. Marginal Effects from the Participation Equation
(Behavioral Problems)
Without Behavioral Problems (Bottom 75%)
Variables Coefficient Standard Error
Price -0.003 (0.033)
Salary 0.0003 *** (0.00004)
Allowance 0.0003 (0.0003)
BMI -0.00002 (0.001)
Psmoke 0.056 *** (0.007)
Healthl -0.095 *** (0.016)
Healthl -0.032 *** (0.012)
Age 0.276 *** (0.047)
Age2 -0.008 *** (0.001)
Male -0.019 ** (0.009)
Hispanic -0.023 * (0.012)
White 0.041 *** (0.014)
Black -0.103 *** (0.010)
Religion -0.035 *** (0.012)
Pmarried -0.025 ** (0.011)
Motheredl 0.018 ** (0.009)
Motheredl 0.005 (0.008)
Mothered3 0.046 ** (0.021)
Mothered4 0.026 *** (0.007)
Workmom -0.002 (0.009)
County characteristics (a) Yes
State-level smoking Yes
sentiment variables (b)
N 10,070
With Behavioral Problems (Top 25%)
Variables Coefficient Standard Error
Price -0.188 *** (0.064)
Salary 0.0003 *** (0.0001)
Allowance 0.0001 (0.001)
BMI -0.003 (0.002)
Psmoke 0.094 *** (0.020)
Healthl -0.175 *** (0.030)
Healthl -0.055 * (0.029)
Age 0.160 (0.119)
Age2 -0.005 (0.004)
Male -0.039 (0.025)
Hispanic -0.034 (0.026)
White -0.004 (0.026)
Black -0.230 *** (0.021)
Religion -0.023 (0.020)
Pmarried 0.006 (0.019)
Motheredl -0.062 (0.038)
Motheredl -0.018 (0.030)
Mothered3 0.004 (0.056)
Mothered4 -0.0001 (0.028)
Workmom -0.011 (0.022)
County characteristics (a) Yes
State-level smoking Yes
sentiment variables (b)
N 3329
Robust standard errors, clustered at the state-level, are in
parentheses. ***, **, and * indicate statistical significance at the
1%a, 5%, and 10% levels, respectively.
(a) These include the county-level variables as described in section 3.
(b) These include the state-level variables as described in section 3.
Table 4. OLS Results: Consumption Equations for Smoking
(Emotional Problems)
Without Emotional Problems
(Bottom 75%)
Variables Coefficient Standard Error
Price -2.474 *** (0.734)
Salary 0.007 ** (0.003)
Allowance -0.002 (0.012)
BMI -0.081 *** (0.028)
Psmoke 1.507 *** (0.365)
Healthl -3.515 *** (0.836)
Healthl -2.114 ** (0.856)
Age 1.337 (1.329)
Age2 -0.013 (0.044)
Male 1.368 *** (0.323)
Hispanic -1.905 *** (0.405)
White 0.485 (0.464)
Black -2.854 *** (0.867)
Religion -1.181 *** (0.427)
Pmarried -1.193 *** (0.420)
Motheredl 1.552 *** (0.510)
Motheredl 1.202 ** (0.482)
Mothered3 4.592 *** (1.421)
Mothered4 0.885 ** (0.344)
Workmom 0.482 (0.391)
County characteristics (a) Yes
State-level smoking Yes
sentiment variables (b)
N 1924
With Emotional Problems
(Top 25%)
Variables Coefficient Standard Error
Price -1.643 (1.642)
Salary 0.006 ** (0.003)
Allowance 0.018 (0.030)
BMI 0.064 (0.044)
Psmoke 2.095 *** (0.535)
Healthl -2.410 *** (0.620)
Healthl -1.008 (0.772)
Age 4.112 * (2.110)
Age2 -0.111 (0.070)
Male 2.366 *** (0.383)
Hispanic -1.340 ** (0.632)
White 0.635 (0.588)
Black -4.075 *** (0.737)
Religion -0.715 (0.728)
Pmarried 0.157 (0.383)
Motheredl 2.232 *** (0.712)
Motheredl 2.156 *** (0.503)
Mothered3 2.908 *** (0.909)
Mothered4 1.771 *** (0.434)
Workmom -0.318 (0.537)
County characteristics (a) Yes
State-level smoking Yes
sentiment variables (b)
N 1131
Robust standard errors, clustered at the state-level, are in
parentheses. ***, **, and * indicate statistical significance at the
1%, 5%, and 10% levels, respectively.
(a) These include the county-level variables as described in section 3.
(b) These include the state-level variables as described in section 3.
Table 5. OLS Results: Consumption Equations for Smoking
(Behavioral Problems)
Without Behavioral Problems (Bottom 75%)
Variables Coefficient Standard Error
Price -1.656 ** (0.770)
Salary 0.004 (0.003)
Allowance 0.010 (0.015)
BMI -0.073 * (0.037)
Psmoke 1.329 *** (0.360)
Healthl -2.470 *** (0.528)
Healthl -0.824 (0.643)
Age 1.249 (1.549)
Age2 -0.009 (0.050)
Male 1.245 *** (0.287)
Hispanic -0.927 (0.572)
White 0.680 (0.534)
Black -2.866 *** (0.544)
Religion -1.145 ** (0.560)
Pmarried -0.595 ** (0.293)
Motheredl 2.286 *** (0.521)
Motheredl 1.455 *** (0.436)
Mothered3 2.923 *** (0.993)
Mothered4 1.441 *** (0.405)
Workmom -0.240 (0.489)
County characteristics (a) Yes
State-level smoking Yes
sentiment variables (b)
N 1595
With Behavioral Problems (Top 25%)
Variables Coefficient Standard Error
Price -2.200 (1.355)
Salary 0.009 ** (0.004)
Allowance -0.001 (0.018)
BMI 0.051 (0.045)
Psmoke 1.950 *** (0.409)
Healthl -2.752 *** (0.565)
Healthl -1.804 *** (0.634)
Age 3.297 ** (1.605)
Age2 -0.080 (0.054)
Male 1.197 *** (0.414)
Hispanic -2.755 *** (0.665)
White 0.612 (0.503)
Black -4.305 *** (1.073)
Religion -0.710 (0.442)
Pmarried -0.702 (0.521)
Motheredl 1.418 ** (0.596)
Motheredl 1.676 *** (0.438)
Mothered3 4.874 *** (1.213)
Mothered4 0.985 ** (0.474)
Workmom 0.408 * (0.236)
County characteristics (a) Yes
State-level smoking Yes
sentiment variables (b)
N 1460
Robust standard errors, clustered at the state level, are in
parentheses. ***, **, and * indicate statistical significance at 1%,
5%, and 10% levels, respectively.
(a) These include the county-level variables as described in section 3.
(b) These include the state-level variables as described in section 3.
Table 6. Estimated Elasticities
First Measure
75-25% Classification 90-10% Classification
With Without With Without
Emotional Emotional Emotional Emotional
Problems Problems Problems Problems
Participation -1.04 -0.33 -0.66 -0.53
Consumption -0.53 -0.89 -0.24 -0.95
Total elasticity -1.57 -1.22 -0.90 -1.48
With Without With Without
Behavioral Behavioral Behavioral Behavioral
Problems Problems Problems Problems
Participation -0.93 -0.05 -1.16 -0.27
Consumption -0.65 -0.67 -0.53 -0.69
Total elasticity -1.58 -0.72 -1.69 -0.95
Second Measure
75-25% Classification 90-10% Classification
With Without With Without
Emotional Emotional Emotional Emotional
Problems Problems Problems Problems
Participation -1.16 -0.29 -0.91 -0.49
Consumption -0.58 -0.89 -0.64 -0.84
Total elasticity -1.74 -1.18 -1.55 -1.34
With Without With Without
Behavioral Behavioral Behavioral Behavioral
Problems Problems Problems Problems
Participation -1.10 0.03 -1.48 -0.19
Consumption -0.73 -0.59 -0.27 -0.90
Total elasticity -1.83 -0.56 -1.75 -1.09
