cigarette strength and smoking behavior · 2 cigarette strength and smoking behavior in june 2009,...
TRANSCRIPT
Cigarette Strength and Smoking Behavior
Esteban Petruzzello∗
Abstract
This paper analyzes the link between cigarette strength and smoking behavior. I es-
timate state dependence demand models that capture two main features of addiction:
reinforcement and the existence of withdrawal costs. I find that demand for lights is
more price sensitive than for regulars, and that lights are less addictive. Addition-
ally, I find that while smoking restrictions are ineffective in reducing regular cigarette
smoking, they reduce light cigarette purchases by about half a pack per month. I also
find that the quitting behavior of regular and light smokers is similar, though there is
some evidence that switching to lights contributes to quitting. These findings support
the idea that an establishment of maximum nicotine levels could be helpful to curtail
smoking.
JEL Codes: D03, D12, I18, L66.
∗Department of Economics, University of Miami - [email protected]. I would like to thank Aviv Nevo,Igal Hendel, and Robert Porter for their support and guidance. I would also like to thank German Bet,Ignacio Franceschelli, Melissa Gray, Frank Limbrock, Lee Lockwood, Fernando Luco, Guillermo Marshall,Cecilia Peluffo, Tiago Pires, and Ana Reynoso for their valuable comments, and the seminar participantsat Northwestern University. This research was funded in part by a cooperative agreement between theUSDA/ERS and Northwestern University, but the views expressed herein are those of the author and donot necessarily reflect the views of the U.S. Department of Agriculture.
1 Introduction
Tobacco smoking is the leading cause of preventable death in the United States. Given that
tobacco kills nearly half a million Americans each year (Surgeon General’s 2014 Report)
and is responsible for substantial health-related economic losses, regulations at all levels of
government have been made to address this public health issue. At the Federal level, the
2009 Family Smoking Prevention and Tobacco Control Act gave the Food and Drug Ad-
ministration (FDA) the power to regulate maximum levels of nicotine, the main psychoac-
tive substance in tobacco and the key causative agent of cigarette addiction. Moreover,
in 2011 tobacco products labeled or advertised with reduced harm claims, such as “light”
or “mild,” were banned by the FDA in an attempt to deter consumers from associating
such products with any health benefits. Light cigarettes have lower machine-measured
nicotine levels than regular strength cigarettes, but the actual levels provided by a light
cigarette vary from smoker to smoker and may even be equivalent to regular cigarettes in
some cases. It is, therefore, an empirical question to determine if light cigarettes are less
addictive than regular strength cigarettes, and if light cigarette smokers behave differently.
This paper contributes to the policy debate by analyzing the link between cigarette
strength and smoking behavior. I use detailed, high-frequency household level purchase
data for the 2006-2010 period to estimate demand models that take into account the ad-
dictive nature of smoking. My baseline set-up consists of a cigarette demand model that
captures one of the main features of addiction, reinforcement : greater past consumption
of an addictive good raises current consumption. I also incorporate a second feature of
addiction: the existence of withdrawal costs. When smokers’ nicotine levels in a given
period are zero (or substantially lower than the habitual levels), they can experience a
physical discomfort that affects their purchase decisions. I analyze this cost by estimating
a specification of the model that captures the effect of temporary smoking cessation on the
purchases of the following period. Additionally, I incorporate the establishment of public
smoking restrictions as a determinant of the purchased quantities.
I estimate the model for both regular and light cigarettes and find the following
results. First, demand for light cigarettes is more price sensitive than for regulars. On
average, a price increase of one dollar reduces regular cigarettes consumption by about
0.3 packs per month, while it reduces light cigarette consumption by more than a pack
per month. Second, light cigarettes are less addictive than regulars, as evidenced by the
2
estimations of the two features of addiction in terms of packs smoked. The reinforcement
effect for regulars is 30% (i.e., a pack smoked in the past month accounts for 0.3 packs
smoked in the current one), while it is only 21% for lights. Moreover, for regular smokers,
the withdrawal cost is responsible for the purchase of 0.8 packs on the month following
the temporary cessation, while that figure is 0.5 for light smokers. Third, public smoking
restrictions are ineffective in reducing regular cigarette smoking, though notably smoking
restrictions reduce average light cigarette purchases by about half a pack per month.
Additionally, I examine the determinants of smoking cessation. I estimate a
discrete-time hazard model of the probability of cessation that includes both time-invariant
household characteristics and time-varying determinants, controlling for duration depen-
dence. I find that higher prices are associated with a higher probability of quitting, and
that smoking restrictions do not affect the likelihood of quitting. There are no signif-
icant differences between regular and light smokers; both groups present similar rates
of quitting successfully and relapsing, though there is some evidence that switching to
lights contributes to quitting for some smokers. Overall, this paper’s findings support the
idea that an establishment of maximum nicotine levels could be helpful to curtail smoking.
The rest of the paper is organized as follows. Section 2 presents some history
on the regulation of cigarette nicotine strength, and a description of smoking behaviors
among consumers of light cigarettes. In Section 3, there is a brief description of the data
and a descriptive analysis of the main variables, including an analysis of switching behav-
ior and of smoking cessation. Section 4 presents the methodology and the main results,
while Section 5 concludes.
3
2 Cigarette strength and smoking behavior
In June 2009, President Obama signed the Family Smoking Prevention and Tobacco Con-
trol Act into law. The main provision of this Act was to give the Food and Drug Ad-
ministration (FDA) the power to regulate the manufacture, distribution, and marketing
of tobacco products in order to protect public health. In particular, the FDA can require
standards for tobacco products, such as maximum tar and nicotine levels, though it cannot
reduce the nicotine levels to zero. Given that nicotine is the main psychoactive substance
in tobacco and the primary causative agent in terms of addiction, it is important to un-
derstand the role of cigarette strength in smoking behavior and addiction.
Prior to 2011, cigarettes were labeled “full” or “light” depending on their nicotine
strength. Light cigarettes, though now not explicitly labelled as such, are fitted with a
porous filter to diffuse the smoke with clean air. When tested by machine, these cigarettes
register lower levels of nicotine, tar, and carbon monoxide than regular cigarettes. How-
ever, health advocates argue that the machine measurements do not represent human
smoking behavior. Various studies have found that many smokers compensate for the low
strength of light cigarettes by changing their pattern of inhalation. This behavior is often
unconscious rather than deliberate, and includes actions such as increasing the drag time,
blocking the filter holes, inhaling deeper, and consuming more cigarettes (Sutton et al.,
1982; Benowitz et al., 1983). The actual nicotine levels provided by a light cigarette thus
vary from smoker to smoker and may even be equivalent to a regular strength cigarette in
some cases. It is, therefore, an empirical question to determine if light cigarettes are less
addictive than regular strength cigarettes.
Smokers choose light cigarettes for a variety of reasons. Some simply prefer the
flavor, which is less harsh than regular strength cigarettes. Others may switch from regu-
lars to lights in an attempt at the “switch down” method of cessation, in which smokers
switch to a lighter brand every three weeks to gradually decrease their nicotine consump-
tion. Interestingly, one study found that switching to lights was associated with a 58%
increase in cessation attempts, but a 60% lower chance of successfully quitting (Tindle et
al., 2010).
Other smokers may use light cigarettes as an alternative to cessation. Light
cigarettes were initially marketed as being healthier than regular strength cigarettes and
4
there is evidence that many smokers perceive them to be less harmful. However, numerous
studies have shown that regular and light cigarettes deliver similar amounts of tar and
nicotine and there are no health advantages to smoking lights instead of regulars (2004
Surgeon General’s Report). For this reason, in 2011 the FDA banned tobacco products
labeled or advertised with reduced harm claims, including “low,” “light,” “mild,” and
similar, misleading descriptors. In response, the tobacco industry replaced the labels with
colors; Marlboro Lights are now Marlboro Gold Pack, for example.
This research is especially relevant to the increased market for electronic cigarettes,
which are not currently regulated by the FDA. The nicotine levels in the cartridges of these
cigarettes can be controlled even more precisely than traditional cigarettes and many
brands already offer the option of no nicotine.
3 Data and descriptive analysis
The main data source for this paper is the Nielsen Homescan Data Set. This database
contains a national panel of households that register their purchases of several product
categories, including cigarettes, on a daily basis from 2006 to 2010. Households in this
panel use a barcode scanner provided by Nielsen to input information about their pur-
chases. Each record in the data shows the date and store where the transaction took
place, how much was purchased of each product (universal product code level), and the
price that was paid. There is also detailed information about product characteristics and
household demographics, both of which are updated annually.
If the transaction takes place at a store for which Nielsen already has store-
level price data, Nielsen obtains the price directly from store data. If the store is not
part of the Nielsen database, the household is asked to enter the price. Households have
incentives provided by Nielsen to join the panel and remain active in reporting their trans-
actions (such as monthly prize drawings and gift points).1 Einav et al. (2010) provide
a quality check for these data by comparing them with data from cash registers. The
authors conclude that the magnitude of the reporting error is similar to other commonly
used databases.
1In order to preserve the quality of the data, Nielsen filters households that do not report their trans-actions regularly, and periodically adds new households to replace the ones who leave (trying to keep thesample representative of national demographics).
5
It is important to emphasize that every cigarette transaction can be recorded, be
it from a big chain store (like Walmart), a small tobacco shop, or a gas station. Even mail
order, online shopping, and vending machines transactions are recorded.2 These data are
quite unique and to the best of my knowledge there is only one paper (Harding et al.,
2012) that makes use of two years of these cigarette data (2006 and 2007) to analyze how
cigarette taxes are passed through to consumer prices.
There are 26,630 households that purchase cigarettes during the 2006-2010 pe-
riod. Analyzing the total sample and the cigarette smokers subsample by year, I find that
about 20% of the households make cigarette purchases, a number consistent with national
figures. Data are observed at the household level; my baseline assumption is that there is
only one smoker per household.
There are four main categories of cigarettes according to their strength: reg-
ular, medium, light, and ultra-light. The two categories that comprise most of the market
are regular and light. For the purposes of this paper, I group the medium, light, and ultra-
light categories under the label “light”. Table A1 in the Appendix shows the nicotine, tar,
and carbon monoxide levels of the top 40 brands in terms of sales.
Most of the households in the sample (about 65%) purchase only regular cigarettes
or light cigarettes. The rest of the households purchase both, but generally with a clear
preference for one of the two. For the purposes of this paper, a household is identified as
a lights smoker if most of its cigarette expenditure is on light cigarettes, while the rest of
the households are considered regular cigarette smokers.
Table 1 presents descriptive statistics.3 Column (1) of Table 1 shows information
for the cigarette smokers sample. Columns (2) and (3) of Table 1 show the results splitting
the sample between regular smokers and light smokers.
2Impersonal sales have been almost entirely prohibited since June 2010 (FDA Consumer Fact Sheet).3Detailed descriptions of the variables can be found in the Appendix. I aggregate data by month of
purchase and weight prices according to the purchased volume. Therefore, a data point is a household-month observation. Also, all the products are standardized at the level of a pack (20 cigarettes), with acigarette length of 85mm (known as King size).
6
Table 1: Descriptive statistics
Entire Sample Regular Smokers Light Smokers
Number of Households 26,630 11,257 15,373
Income $51,894 $48,609 $54,300
Household Size 2.5 2.6 2.5
Income per HH Member $24,387 $22,511 $25,761
Age 49.9 50.0 49.9
Minority 18% 23% 15%
Some College 74% 72% 75%
Unemployed or Retired 24% 27% 22%
Packs per Month 8.3 7.6 8.7
Price per Pack $3.8 $3.8 $3.8
Source: Nielsen Homescan data.
The table displays some differences in characteristics; households that smoke light
cigarettes are wealthier and have a higher employment rate. Light smokers also smoke on
average one more pack per month than regular smokers. Table A2 in the Appendix shows
the expenditure shares across demographics of each type of cigarette according to their
strength.
I closely examined the purchasing behavior of about 100 households that switch
from smoking regular cigarettes to smoking lights. There are two patterns that emerge
with higher frequency. One of them is supportive of the “switch down” method of ces-
sation: these households switch to lights and gradually reduce the purchased amounts,
sometimes stopping afterwards. The other one is supportive of the “compensatory smok-
ing” phenomena: these households switch from regulars to lights and start buying more
cigarettes, potentially to compensate for the nicotine loss. Figures A1 and A3 in the Ap-
pendix show those patterns for selected households.
There is strong persistence in purchased brands; more than half of the house-
holds purchase only one brand, and for about 80% of the households the top brand has a
share higher than 70% of the total cigarette purchases. Additionally, cigarette purchases
are very heterogeneous across households. The median monthly purchase in the sample is
about two packs per month, while the 75th percentile is more than 10 packs.
7
Monthly household purchases present a fairly constant pattern. Figure A5 in the
Appendix shows the coefficient of variation (standard deviation - mean ratio) of monthly
purchases for each of the 26,630 households that make purchases in the 2006-2010 period,
with households in ascending order according to total purchases. Most of the sample has
a coefficient of variation lower than 1, with this phenomenon being starker for households
that purchase greater amounts.
For the purposes of this paper, I consider that purchases and consumption are
equal at the month level. This is unlikely to cause interpretation problems for two reasons.
First, cigarettes do not store well and are known to dry out and have a bad flavor after
some weeks. Second, there is evidence that many smokers try to only purchase the amount
they will consume in a relatively short time period as a way of exerting self-control. Hav-
ing an excess stock of cigarettes could tempt the smoker into smoking more, thus many
smokers opt for frequent, small purchases to regulate their consumption (Kim et al., 2006).
In my data, a household makes on average 2.7 cigarette transactions per month, while the
median is about 2 transactions; these figures support the aggregation by month of pur-
chase.
Data on the various increases in state cigarette excise taxes from 2006-2010
come from the Federation of American Tax Administrators. Information about state-
level restrictions comes from the Centers for Disease Control and Prevention. During the
2006-2010 period, several U.S. states (and the District of Columbia, which I treat as a
state for the purposes of this paper) established smoking restrictions in public places such
as bars, restaurants, and workplaces. Table 2 shows the timing of such restrictions.
8
Table 2: List of state smoking restrictions, 2006-2010
Month Year State Month Year State
April 2006 New Jersey October 2007 Minnesota
July 2006 Arkansas January 2008 Illinois
July 2006 Colorado February 2008 Maryland
November 2006 Hawaii July 2008 Iowa
November 2006 Nevada September 2008 Pennsylvania
December 2006 Ohio January 2009 Oregon
January 2007 District of Columbia June 2009 Nebraska
January 2007 Louisiana November 2009 South Dakota
January 2007 Utah December 2009 Virginia
May 2007 Arizona January 2010 North Carolina
June 2007 New Mexico May 2010 Michigan
July 2007 Tennessee July 2010 Kansas
September 2007 New Hampshire July 2010 Wisconsin
Source: Centers for Disease Control and Prevention. There are fourteen states with smoking
restrictions established before 2006: California, Delaware, Connecticut, Florida, Idaho, New York,
Maine, Massachusetts, Georgia, Montana, North Dakota, Rhode Island, Vermont, and Washington.
The state of Indiana established a restriction on 2012. The remaining ten states have no statewide
smoking restriction as of May 2015: Alabama, Alaska, Kentucky, Mississippi, Missouri, Oklahoma,
South Carolina, Texas, West Virginia, and Wyoming.
3.1 Smoking cessation
More than half of the adult smoking population tried to quit smoking in 2010 (Behavioral
Risk Factor Surveillance System Survey Data, 2011). Although the health benefits are
greater for people who stop at earlier ages, cessation has major health benefits at all ages
(Centers for Disease Control and Prevention Fact Sheet, 2012). Quitting is difficult and
relapse rates are high; several attempts are usually required to stop smoking.
3.1.1 Quitting
I consider that households have ceased to smoke in month t if they made their last cigarette
purchase in month t− 1 and have been otherwise active on the Nielsen panel for at least
three more months. A rationale for the choice of three months as the minimum threshold
is that it has been shown that it takes between 1.5 and 3 months after quitting before
9
the number of nicotinic receptors in the brain of a smoker normalizes to non-smoker levels
(Cosgrove et al., 2009). Obviously, I have no way of confirming if this is in fact a per-
manent quitting situation, but this offers a reasonable approximation. According to this
definition of quitting, 51% of the households in the sample make a potentially successful
quitting attempt. This figure is virtually the same when comparing regular and light
smokers (51% and 50%, respectively).
3.1.2 Temporary quitting
Table A3 in the Appendix shows the number of quitting attempts of more than two months
per household, splitting the sample between regular smokers and light smokers. We can
see that there are no significant differences between the number of attempts made by those
two types of smokers.
3.1.3 Relapsing
I consider that a household has relapsed if:
• they were active in the panel for at least three consecutive months in which they did
not purchase cigarettes; and
• they purchased cigarettes sometime after that.
According to this definition, 63% of the households relapse at least once. This figure is
consistent with survey data. Splitting the sample by regular and light smokers, I find that
64% of regular smokers relapse, while 62% of light smokers do so.
4 Methodology and results
4.1 Baseline reinforcement model
I am interested in estimating the role of state dependence in the demand for cigarettes. I
do so by estimating the following equation:
qit = γ · pit + β · qi,t−1 + αi + µt + εit (1)
where qit are the packs purchased by household i in month t; pit is the average price per
pack that was paid for that amount; αi is household i’s fixed effect; µt is a month effect
10
common to all households; and εit is the error term, assumed to be independent across
households and serially uncorrelated.4
This equation is consistent with a boundedly-rational model of addiction in which
individuals recognize the impact of past decisions but can only choose their current con-
sumption based on current prices. I test the validity of this assumption in Petruzzello
(2015).5 State dependence is captured by adding a lag of the dependent variable as a re-
gressor. I analyze if, after controlling for time-invariant systematic differences (the αi’s),
the previous consumption helps to explain the current one. More specifically, β is a mea-
sure of the extent of reinforcement, one of the main features of addiction: the fact that
greater past consumption of an addictive good raises current demand. Thus, if a good is
addictive, we would expect β to be positive; the degree of reinforcement is greater when
the coefficient β is larger.
Results for the estimation of Equation (1) are shown in Table 3. Columns (1),
(3), and (5) of Table 3 show the results of OLS fixed-effects estimation. There is a po-
tential endogeneity bias that could arise from regressing the purchased packs of cigarettes
on the price paid. Therefore, I also explore specifications in which I employ state excise
taxes as instruments for the potentially endogenous prices; columns (2), (4), and (6) of
Table 3 show the results of 2SLS estimation.6 State excise taxes vary widely across states;
for example, the excise tax in Missouri is $0.17, while that figure is $4.35 in New York
(the median tax rate is $1.57).7,8 State taxes also vary through time; most of the states
raise their cigarette excise taxes one or more times in the 2006-2010 period, and this is the
key source of exogenous price variation. Gruber and Koszegi (2001) report that state tax
increases are responsible for about 80% of the within state-year variation in prices. Table
A5 in the Appendix shows all the excise tax increases, new rates, and date of application
for any increase between January 2006 and December 2010. Table A6 in the Appendix
shows first stage estimation results.
4I address the validity of this assumption in Petruzzello (2014).5An alternative approach would be to include a lead of the dependent variable; this would be consistent
with the standard rational addiction model specification, which relies on forward-looking behavior.6Cigarette excise taxes have been employed as an instrument for cigarette prices in similar contexts
(Becker et al., 1994; Gruber et al., 2003; Adda and Cornaglia, 2013).7Very few counties and cities in the U.S. establish their own excise taxes.8The federal tax rate had been $0.39 since 2002 until April 2009, when it was raised to $1.01 to provide
funding for the 2009 expansion of the Children’s Health Insurance Program.
11
Table 3: Estimation of the baseline model
Dep. variable: packs purchased per month
All Regular Light
(1) (2) (3) (4) (5) (6)
OLS 2SLS OLS 2SLS OLS 2SLS
Price -0.1997∗∗∗ -1.2904∗∗∗ -0.1086∗∗∗ -0.8673∗∗∗ -0.7027∗∗∗ -1.1615∗∗∗
(0.0406) (0.2178) (0.0212) (0.2808) (0.0680) (0.2613)
Packs 0.2607∗∗∗ 0.2572∗∗∗ 0.2999∗∗∗ 0.2978∗∗∗ 0.2128∗∗∗ 0.2111∗∗∗
t-1 (0.0111) (0.0112) (0.0249) (0.0248) (0.0058) (0.0058)
F-stat 1639.00 631.14 1355.85
R2 0.6787 0.0549 0.6627 0.0681 0.6880 0.0745
N 300032 300032 139092 134089 194165 189127
Robust standard errors are in parenthesis. All the regressions include month dummies
and household fixed effects. Columns (1), (3), and (5) show the results of OLS estima-
tion, while Columns (2), (4), and (6) present 2SLS estimation results. The R squared
reported for 2SLS estimations is the within R squared. F-stat is the statistic for the first-
stage F test of excluded instruments. *Statistically different from 0 at 10% significance;
**Statistically different from 0 at 5% significance; ***Statistically different from 0 at 1%
significance.
These results show that the demand for light cigarettes is more price sensitive
than for regular cigarettes; this suggests that public intervention in the form of higher
cigarette taxes could be more successful to curb lights smoking. Additionally, light
cigarettes display a lower state dependence than regulars: the reinforcement effect for
regulars is 30% (a pack smoked in the past month accounts for 0.3 packs smoked in the
current one), while that figure is only 21% for lights.
4.1.1 Incorporating withdrawal costs
The specification I have explored incorporates one of the key components of addiction:
reinforcement. I also explore a specification that incorporates withdrawal costs. These
manifest as a physical discomfort that affects the current decision when the levels of nico-
tine in the body are substantially lower than the habitual levels.9 This effect is related to
9These effects have been widely documented; Harris (1993) provides a summary.
12
past cigarette consumption but is only triggered for individuals that try to reduce their
consumption. Withdrawal costs have an asymmetric nature; they are not triggered by
increases in consumption, only by reductions. As pointed out by Suranovic et al. (1999),
the development of adjustment costs associated with cessation is sufficient to explain why
smoking is habit-forming and is a critical element in making cigarettes addictive.10
I estimate a variation of the baseline model in which I include a withdrawal
cost binary variable, Wit. That variable is equal to 1 in period t for household i if that
household purchased cigarettes on period t − 2 but did not purchase cigarettes in t − 1
(while still purchasing some other product in t−1).11 If withdrawal costs were not a factor,
we would expect that temporary cessation in the past month has no significant effect on
current consumption, holding the rest of the variables fixed. Table A7 in the Appendix
presents the results of the estimation, which show that the withdrawal cost coefficients are
statistically and economically significantly different from zero. Table A8 in the Appendix
shows first stage estimation results.
For regular smokers, the withdrawal cost is responsible for an average of 0.8
packs on the month following the temporary cessation. This figure is 0.5 packs for light
smokers. Therefore, there is evidence of a lower impact of the withdrawal cost for light
smokers. For this specification, there is no significant difference between the reinforcement
effects for regular and light smokers.
4.2 Differential impact of smoking restrictions
I now examine the effectiveness of public smoking restrictions in curbing smoking and
analyze whether the impact is different for lights and regulars. The main purpose be-
hind the enactment of these types of restrictions is to reduce the exposure of non-smokers
to secondhand smoke. Nonetheless, according to the World Health Organization, smok-
ing restrictions may reduce the demand for cigarettes by creating an environment where
smoking becomes increasingly more difficult and also by shifting social norms. By directly
prohibiting smoking in certain areas, these restrictions may deter social smokers by elimi-
nating common smoking settings. Additionally, these restrictions affect habitual smokers
10“An addiction is the compulsive need for and use of a habit-forming substance characterized bytolerance and by well-defined physiological symptoms upon withdrawal” (Merriam-Webster dictionary).
11For this specification it is crucial to consider the months with no cigarette purchases. To define aprice for the months with no purchases, I assign the last price paid and update this value with any increasein state or federal taxes if applicable.
13
by increasing the hassle cost associated with smoking, as they must move out of restricted
areas in order to smoke.
However, it could be the case that some smokers maintain their cigarette con-
sumption even in the presence of restrictions. Adda and Cornaglia (2010) find some evi-
dence that restrictions simply displace smoking to areas that are not restricted. Owyang
and Vermann (2012) do not find correlation between smoking restrictions and smoking
behavior examining survey data, while Irvine and Nguyen (2011) find that smoking re-
strictions have an effect on individuals at the top of the income distribution (they employ
the 2003 Canadian Community Health Survey).
I benefit from the fact that there are states with no smoking restrictions during
the 2006-2010 period and that the states that established restrictions did so at different
moments of time. This variability is exploited to identify the causal effect of smoking
restrictions on cigarette purchases, controlling for time-invariant household characteristics
and temporary shocks that are common across households. Specifically, households in
states that lack restrictions serve as controls for those in states that have smoking restric-
tions. I define the smoking restriction binary variable Bit, which is equal to 1 in period t
for household i if the state where the household resides established a smoking restriction
on t or before. Table 4 shows the results of incorporating this smoking restriction time-
varying dummy on the specification of Equation (1) (Table A9 in the Appendix presents
first stage results).
14
Table 4: Impact of state smoking restrictions
Dep. variable: packs purchased per month
All Regular Light
(1) (2) (3) (4) (5) (6)
OLS 2SLS OLS 2SLS OLS 2SLS
Smoking -0.4200∗∗ -0.4572∗∗∗ -0.0529 -0.1660 -0.4958∗∗ -0.3939∗∗
restriction (0.1690) (0.1737) (0.2323) (0.2516) (0.2013) (0.2000)
Price -0.1467∗∗∗ -1.6517∗∗∗ -0.0876∗∗∗ -0.6908 -0.6740∗∗∗ -2.0289∗∗∗
(0.0301) (0.3445) (0.0159) (0.4397) (0.0672) (0.4092)
Packs 0.2874∗∗∗ 0.2835∗∗∗ 0.3432∗∗∗ 0.3415∗∗∗ 0.2249∗∗∗ 0.2213∗∗∗
t-1 (0.0140) (0.0142) (0.0303) (0.0303) (0.0069) (0.0069)
F-stat 712.14 177.64 1255.90
R2 0.6746 0.0169 0.6568 0.1063 0.6868 0.0653
N 217715 217715 99547 96016 142737 139159
Robust standard errors are in parenthesis. All the regressions include month dummies
and household fixed effects. Columns (1), (3), and (5) show the results of OLS estima-
tion, while Columns (2), (4), and (6) present 2SLS estimation results. The R squared
reported for 2SLS estimations is the within R squared. F-stat is the statistic for the first-
stage F test of excluded instruments. *Statistically different from 0 at 10% significance;
**Statistically different from 0 at 5% significance; ***Statistically different from 0 at 1%
significance.
These results show that public smoking restrictions are inadequate to reduce
regular cigarette smoking. Notably, this is not the case for light cigarettes, as smoking
restrictions reduce average light cigarette purchases by about half a pack per month.
4.2.1 Examination of household heterogeneity
In order to examine the role of heterogeneity in addiction, I split the sample by median
age and median income per household member (also splitting the regular and light smok-
ers subsamples). I then estimate the baseline model with the state smoking restrictions
dummy for each of those cells. Table A10 in the Appendix shows the estimations of
15
the reinforcement coefficient for each cell. We can see that the reinforcement coefficient
for regular smokers is much smaller for younger and for wealthier households, while the
coefficients for light smokers do not present significant differences across cells.
I also investigate heterogeneity in the effects of smoking restrictions. Table A11
in the Appendix shows the coefficients on the state restriction dummy variable for each
cell.12 We can see that the restrictions are effective for younger and wealthier households
that smoke light cigarettes, but not for the rest of the cells.
4.3 Differential cessation and its determinants
In order to analyze the determinants of quitting, I estimate a discrete-time hazard logit
model of the probability of cessation. This model presents both time-invariant household
characteristics (minority status, age, and education level in 2006) and time-varying deter-
minants (the establishment of state smoking restrictions, price, new births, and income),
and it includes a sixth-degree time polynomial to control for duration dependence.13 I
employ the definition of quitting previously stated in Section 3.
Formally,
P (dit = 1) =exp(z)
1 + exp(z)(2)
where dit is the binary survival variable that takes a value of one if household i stops
purchasing in month t and zero otherwise, and
z ≡ a0 + g · Vi,t + h · Fi + P (t) + ιit (3)
where the V ’s are time-varying variables, the F ’s are fixed household demographics, P (t)
is a sixth-order polynomial in time included to control for duration dependence, and ιit is
the error term. The results are reported in Table 5.
12The coefficients from Tables A10 and A11 correspond to 2SLS estimation.13For this model I need to define the price for the quitting month and for previous months when there
were no purchases of cigarettes but the household was active in some other category. For those months Iassign the last price paid and I update this value with any increase in state or federal taxes if applicable.
16
Table 5: Estimation of the cessation hazard model
Dep. variable: survival dummy (cessation)
(1) (2) (3)
All Regular Light
Time-varying :
Smoking restriction 0.0103 -0.0288 0.0265
(0.0228) (0.0353) (0.0300)
Price 0.0025∗ 0.0029∗∗ 0.0089∗∗
(0.0014) (0.0013) (0.0041)
Births 0.4722∗∗∗ 0.4886∗∗∗ 0.4754∗∗∗
(0.0600) (0.0965) (0.0766)
log of Income 0.1677∗∗∗ 0.2001∗∗∗ 0.1521∗∗∗
(0.0153) (0.0230) (0.0208)
Time-invariant :
Age -0.0036∗∗∗ -0.0021 -0.0044∗∗∗
(0.0010) (0.0015) (0.0013)
College 0.0719∗∗∗ 0.1040∗∗∗ 0.0529
(0.0243) (0.0369) (0.0324)
Minority 0.0290 0.0192 0.0114
(0.0276) (0.0389) (0.0400)
R2 0.0122 0.0147 0.0108
N 375420 144933 222679
Standard errors are in parentheses. All the specifications include a sixth-degree time polynomial to
control for duration dependence. *Statistically different from 0 at 10% significance; **Statistically
different from 0 at 5% significance; ***Statistically different from 0 at 1% significance.
The likelihood of quitting is higher when the income is higher and when follow-
ing the birth of a child, as expected. We can also see that younger smokers are more
likely to quit. Education only has a positive impact on the probability of quitting for
regular smokers. In terms of smoking policy variables, higher prices are associated with a
higher probability of quitting for both types. Additionally, the establishment of smoking
restrictions does not have a significant impact on the probability of quitting for either
type.
17
5 Conclusion
I analyze the link between cigarette strength and smoking behavior by employing house-
hold level purchase data, examining cessation and switching patterns, and estimating state
dependence demand models that take into account the addictive nature of smoking.
The results show that light cigarettes are less addictive than regular cigarettes
and that demand for lights is more price sensitive than for regulars. Moreover, public
smoking restrictions reduce light cigarette purchases by about half a pack per month,
though they are ineffective in reducing regular cigarette smoking. Additionally, higher
prices are associated with a higher probability of quitting successfully, but public smoking
restrictions do not affect the likelihood of cessation for either type of smoker. Lastly, there
is some evidence that switching to lights contributes to quitting for some smokers.
Overall, this paper’s findings support the idea that an establishment of maxi-
mum nicotine levels could be helpful to curtail smoking, with two caveats. First, it could
be the case that the difference in relevant demographics between light and regular cigarette
smokers plays a role in the difference between their behaviors. Second, some of the house-
holds that switch to lights present evidence of compensatory behavior; in those cases, the
establishment of maximum nicotine levels would not generate the desired effect.
18
References
[1] ADDA, J. and CORNAGLIA, F. (2010): The Effect of Bans and Taxes on Passive Smoking,
American Economic Journal: Applied Economics, 2: 1-32.
[2] ADDA, J. and CORNAGLIA, F. (2013): Taxes, Cigarette Consumption, and Smoking Inten-
sity: Reply, American Economic Review, 103, 3102-3114.
[3] AHIJEVYCH, K. and GARRETT, B.E. (2004): Menthol Pharmacology and Its Potential
Impact of Cigarette Behavior, Nicotine and Tobacco Research, 6, S17-28.
[4] ALVAREZ, J. and ARELLANO, M. (2003): The Time Series and Cross-Section Asymptotics
of Dynamic Panel Data Estimators, Econometrica, 71, 1121-1159.
[5] ARELLANO, M. and BOND, S. (1991): Some Tests of Specification for Panel Data: Monte
Carlo Evidence and an Application to Employment Equations, Review of Economic Studies,
58, 277-297.
[6] BECKER, G., GROSSMAN, M., and MURPHY, K. (1994): An Empirical Analysis of
Cigarette Addiction, American Economic Review, 84, 396-418.
[7] BECKER, G. and MURPHY, K. (1988): A Theory of Rational Addiction, Journal of Political
Economy, 96, 675-700.
[8] Behavioral Risk Factor Surveillance System Survey Data, (2011).
[9] BOND, S. Dynamic Panel Data Models: A Guide to Micro Data Methods and Practice The In-
stitute for Fiscal Studies, Department of Economics, UCL cemmap working paper CWP09/02
[10] Centers for Disease Control and Prevention Fact Sheet, (2012).
[11] CHALOUPKA, F.J. (1991): Rational Addictive Behavior and Cigarette Smoking, Journal of
Political Economy, 99, 722-42.
[12] COSGROVE, K.P., BATIS, J., BOIS, F., MACIEJEWSKI, P.K., ESTERLIS, I., KLOCZYN-
SKI, T., STIKLUS, S., KRISHNAN-SARIN, S., O’MALLEY, S., PERRY, E., TAMAGNAN,
G., SEIBYL, J.P., and STALEY, J.K. (2009): Beta2*-Nicotinic Acetylcholine Receptor Avail-
ability during Acute and Prolonged Abstinence from Tobacco Smoking Archives of General
Psychiatry, 66, 666-676.
[13] DELNEVO, C.D., GUNDERSEN, D.A., and HRWYNA, M. (2010): Examining the Relation-
ship Between Menthol Smoking and Cessation Using Data from the 2003 and 2006/7 Tobacco
Use Supplement, submission to the U.S. Food and Drug Administration.
[14] EINAV, L., LEIBTAG, E., and NEVO, A., (2010): Recording Discrepancies in Nielsen Home-
scan Data: Are They Present and Do They Matter? Quantitative Marketing and Economics,
8, 207-239.
19
[15] FAGAN, P., MOOLCHAN, E.T., HART, A., ROSE, A., LAWRENCE, D., SHAVERS, V.L.,
and GIBSON, J.T. (2010): Nicotine Dependence and Quitting Behaviors among Menthol and
Non-Menthol Smokers with Similar Consumption Patterns Addiction, 105, 55-74.
[16] GRUBER, J. and KOSZEGI, B. (2001): Is Addiction “Rational”? Theory and Evidence,
Quarterly Journal of Economics, 116, 1261-303.
[17] GRUBER, J., SEN, A., and STABILE, M., (2003): Estimating Price Elasticities When There
Is Smuggling: The Sensitivity of Smoking to Price in Canada Journal of Health Conomics, 22,
821-842.
[18] GUL, F. and PESENDORFER, W. (2007): Harmful Addiction, Review of Economic Studies,
74, 147-172.
[19] HARDING, M., LEIBTAG, E., and LOVENHEIM, M.F. (2012): The Heterogeneous Geo-
graphic and Socioeconomic Incidence of Cigarette Taxes: Evidence from Nielsen Homescan
Data, American Economic Journal: Economic Policy, 4, 169?198.
[20] HARRIS, J., (1993): Deadly Choices: Coping with Health Risks in Everyday Life. Basic
Books, New York.
[21] HARTMAN, A.M. (2011): What Menthol Smokers Report They Would Do If Menthol
Cigarettes Were No Longer Sold, Presentation to the FDA Tobacco Products Scientific Advi-
sory Committee Meeting.
[22] IRVINE, I.J. and NGUYEN, H.V. (2011): Toxic Choices: The Theory and Impact of Smoking
Bans, Forum for Health Economics and Policy, 14.
[23] KEELER, T.E., HU, T-W., BARNETT, P.G., and MANNING, W.G. (1993): Taxation,
Regulation, and Addiction: A Demand Function for Cigarettes Based on Time-Series Evidence,
Journal of Health Economics, 12, 1-18.
[24] KIM, A.E., RIBISL, K.M., DELNEVO, C.D., and HRYWNA, M. (2006): Smokers’ Beliefs
and Attitudes about Purchasing Cigarettes on the Internet Public Health Rep., 121, 594-602.
[25] LAWRENCE, D., ROSE, A., FAGAN, P., MOOLCHAN, E.T., GIBSON, J.T., and
BACKINGER, C.L. (2010): National Patterns and Correlates of Mentholated Cigarette Use
in the United States Addiction, 105, 13-31.
[26] LEVY, D.T., BLACKMAN, K., TAURAS, J., CHALOUPKA, F.J., VILLANTI, A., and
NIAURA, R.S. (2011): Quit Attempts and Quit Rates among Menthol and Non-Menthol
Smokers in the United States, American Journal of Public Health, 101, 1241-1247.
[27] OWYANG, M.T. and VERMANN, E.K. (2012): Where There‘s a Smoking Ban, There‘s Still
Fire, Federal Reserve Bank of St. Louis Review, 94, 265-86.
[28] PETRUZZELLO, E. (2014): Measuring the Effect of Policy on the Demand for Menthol
Cigarettes: Evidence from Household Level Purchase Data, working paper.
20
[29] PETRUZZELLO, E. (2015): Testing for Forward-Looking Behavior: Evidence from the En-
actment of Smoking Restrictions, working paper.
[30] Surgeon General’s Report (2014).
[31] SURANOVIC, S.M., GOLDFARB, R.S., and LEONARD, T.C. (1999): An Economic Theory
of Cigarette Addiction Journal of Health Economics, 18, 1-29.
[32] The Industry Menthol Report (2001): Menthol Cigarettes: No Disproportionate Impact on
Public Health.
[33] TAURAS, J.A., LEVY, D.T, CHALOUPKA, F.J., VILLANTI, A., NIAURA, R.S., VAL-
LONE, D., and ABRAMS, D.B. (2010): Menthol and Non-menthol Smoking: the Impact of
Prices and Smoke-Free Air Laws, Addiction, 105, 115-123.
[34] Tobacco Products Scientific Advisory Committee (2011): Menthol Cigarettes and Public
Health: Review of the Scientific Evidence and Recommendations.
21
A Description of the variables
• College dummy : 1 if maximum education level attained in the household is any
college attendance. This includes “Some College,” “Graduated College,” and “Post
College Grad.”
• Age: Age on December 2005 (if more than one adult, average age).
• Minority dummy : 0 if the respondent answered “White” and “No” to the “Race”
and “Hispanic” entries, respectively.
• Income: Mid-point of the corresponding annual income interval (annual income is
categorized in 17 brackets between $5,000 and $200,000).
• Unemployed or Retired : 1 if no household head is employed.
• Births: Birthsit is equal to 1 if there was a birth in household i on month t or before,
but after January 2006; and is equal to 2 if there was a second birth in household i
on month t or before, but after January 2006. It takes the value zero otherwise.
22
B Characteristics of top brands
Table A1: Top 40 selling brands
Brand Firm Menthol Strength Tar Nicotine COMarlboro PM Light 11 0.9 13Marlboro PM Regular 16 1.2 16Newport L Y Regular 14 1.2 15Marlboro PM Ultra light 5 0.5 8Parliament PM Light 11 0.9 12Camel R Light 9 0.8 10Marlboro PM Medium 13 1 12Winston R Light 11 0.9 13Winston R Regular 16 1.3 15Kool R Y Regular 16 1.1 15Basic PM Light 11 0.8 15Salem R Y Light 8 0.7 10Marlboro PM Y Light 9 0.7 11Camel R Regular 17 1.3 15Basic PM Regular 15 1 18Merit PM Ultra light 5 0.5 7Basic PM Ultra light 5 0.4 8Virginia Slims PM Light 9 0.7 10Virginia Slims PM Y Light 9 0.7 10Salem R Y Regular 17 1.3 16Carlton R Ultra light 1 0.1 1Marlboro PM Y Regular 14 1 13Doral R Light 10 0.8 11Newport L Y Light 12 1 12Merit PM Regular 9 0.8 12Virginia Slims PM Ultra light 6 0.5 6Basic PM Y Light 11 0.8 15Virginia Slims PM Y Ultra light 6 0.5 6Pall Mall R Regular 14 1.2 12Salem R Y Ultra light 6 0.5 9Doral R Ultra light 4 0.4 6Capri R Light 9 0.8 6Virginia Slims PM Regular 14 1 12Virginia Slims PM Y Regular 14 1 12Doral R Regular 14 1.1 13Kool R Y Mild 11 0.9 10Winston R Ultra light 4 0.5 6Misty Slims R Y Light 9 0.7 10Capri R Y Light 9 0.8 6Misty Slims R Light 8 0.7 9
Source: Nielsen Homescan data and Federal Trade Commission. Nicotine,tar, and carbon monoxide are measured in milligrams per cigarette. PMstands for Philip Morris, R for Reynolds, and L for Lorillard.
23
C Expenditure shares
Table A2: Cigarette expenditure shares by demographics
Race/Ethnicity Regular Medium Light Ultra-light
White 39% 3% 44% 15%
African American 57% 5% 31% 7%
Hispanic 44% 3% 41% 12%
Asian 35% 7% 38% 21%
Income
Lowest fourth 46% 2% 39% 12%
Second fourth 42% 2% 42% 14%
Third fourth 38% 3% 44% 15%
Highest fourth 32% 3% 46% 19%
Education
Grade school 51% 1% 38% 10%
High school 45% 2% 41% 12%
Some college 41% 3% 42% 14%
College grad 34% 3% 45% 18%
Source: Nielsen Homescan data.
24
D Switching patterns
Figure A1: Switch down behavior examples
Source: Nielsen Homescan data. �: regular cigarettes; ♦: light
cigarettes; ×: no purchase of cigarettes but active in panel
25
Source: Nielsen Homescan data. �: regular cigarettes; ♦: light
cigarettes; ×: no purchase of cigarettes but active in panel
26
Figure A3: Compensatory behavior examples
Source: Nielsen Homescan data. �: regular cigarettes; ♦: light
cigarettes; ×: no purchase of cigarettes but active in panel
27
Source: Nielsen Homescan data. �: regular cigarettes; ♦: light
cigarettes; ×: no purchase of cigarettes but active in panel
28
E Consumption behavior
Figure A5: Coefficient of variation for monthly purchases of each household
Source: Nielsen Homescan data.
29
F Quitting attempts
Table A3: Number of quitting attempts of more than two months
# of attempts # of households Regular smokers % Light smokers %
0 6,388 23% 24%
1 6,079 22% 23%
2 7,149 27% 27%
3 3,637 14% 13%
4 1,903 7% 7%
5 937 3% 4%
6 or more 537 2% 2%
26,630 100% 100%
Source: Nielsen Homescan data
30
G Smoking restrictions
Table A4: States with smoking restrictions and year of enactment
Year States
1995 California
2002 Delaware
2003 Connecticut, Florida, and New York
2004 Indiana, Maine, and Massachusetts
2005 Georgia, Montana, North Dakota, Rhode Island, Vermont, and Washington
2006 Arkansas, Colorado, Hawaii, Nevada, New Jersey, and Ohio
2007 Arizona, District of Columbia, Louisiana, Minnesota, New Hampshire,
New Mexico, Tennessee, and Utah
2008 Illinois, Iowa, Maryland, and Pennsylvania
2009 Nebraska, Oregon, South Dakota, and Virginia
2010 Kansas, Michigan, North Carolina, and Wisconsin
2012 Indiana
Note: there are 10 states with no statewide smoking ban as of May 2015: Alabama, Alaska,
Kentucky, Mississippi, Missouri, Oklahoma, South Carolina, Texas, West Virginia, and Wyoming.
31
H Excise tax increases, 2006-2010
Table A5: Excise tax increases, rates, and date of application
Date State Increase New Rate Date State Increase New Rate
07/01/06 Alaska 0.20 1.80 04/01/09 U.S. (federal rate) 0.62 1.01
07/01/06 New Jersey 0.18 2.58 04/10/09 Rhode Island 1.00 3.46
07/01/06 North Carolina 0.05 0.35 05/15/09 Mississippi 0.50 0.68
09/30/06 Hawaii 0.20 1.60 07/01/09 Delaware 0.45 1.60
12/07/06 Arizona 0.82 2.00 07/01/09 Florida 1.00 1.33
01/01/07 South Dakota 1.00 1.53 07/01/09 Hawaii 0.60 2.60
01/01/07 Texas 1.00 1.41 07/01/09 Kentucky 0.30 0.60
03/15/07 Iowa 1.00 1.36 07/01/09 New Hampshire 0.70 1.78
07/01/07 Alaska 0.20 2.00 07/01/09 New Jersey 0.13 2.70
07/01/07 Connecticut 0.49 2.00 07/01/09 Vermont 0.45 2.24
07/01/07 Indiana 0.44 1.00 07/01/09 Wisconsin 0.75 2.52
07/01/07 New Hampshire 0.28 1.08 09/01/09 North Carolina 0.10 0.45
07/01/07 Tennessee 0.42 0.62 10/01/09 Connecticut 1.00 3.00
08/01/07 Delaware 0.60 1.15 10/01/09 District of Columbia 1.50 2.50
09/30/07 Hawaii 0.20 1.80 11/01/09 Pennsylvania 0.25 1.60
01/01/08 Maryland 1.00 2.00 05/01/10 Washington 1.00 3.03
01/01/08 Wisconsin 1.00 1.77 07/01/10 Hawaii 0.40 3.00
06/03/08 New York 1.25 2.75 07/01/10 New Mexico 0.75 1.66
07/01/08 Massachusetts 1.00 2.51 07/01/10 New York 1.60 4.35
09/30/08 Hawaii 0.20 2.00 07/01/10 South Carolina 0.50 0.57
03/01/09 Arkansas 0.56 1.15 07/01/10 Utah 1.01 1.70
Source: Federation of American Tax Administrators.
32
I Additional results
Table A6: Baseline model - First stage results
Dep. variable: price paid
All Regular Light
(1) (2) (3)
Tax rate 0.6901∗∗∗ 0.7404∗∗∗ 0.6660∗∗∗
(0.0170) (0.0295) (0.0181)
F-stat 1639.00 631.14 1355.85
χ2-stat 24.34 7.30 2.87
R2 0.0298 0.0160 0.1118
N 293947 134089 189127
Robust standard errors are in parenthesis. All the regressions include month dummies and
the lagged quantity. F-stat is the statistic for the first-stage F test of excluded instruments.
***Statistically different from 0 at 1% significance.
33
Table A7: Estimation of the baseline model with withdrawal costs
Dep. variable: packs purchased per month
All Regular Light
(1) (2) (3) (4) (5) (6)
OLS 2SLS OLS 2SLS OLS 2SLS
Withdrawal 0.6640∗∗∗ 0.6545∗∗∗ 0.8185∗∗∗ 0.8050∗∗∗ 0.5442∗∗∗ 0.5411∗∗∗
cost effect (0.0820) (0.0843) (0.1514) (0.1563) (0.0824) (0.0826)
Price -0.0682∗∗∗ -0.9854∗∗∗ -0.0389∗∗∗ -0.7880∗∗∗ -0.2972∗∗∗ -1.1154∗∗∗
(0.0068) (0.0842) (0.0046) (0.1322) (0.0151) (0.1080)
Packs 0.3263∗∗∗ 0.3219∗∗∗ 0.3412∗∗∗ 0.3369∗∗∗ 0.3143∗∗∗ 0.3109∗∗∗
t-1 (0.0082) (0.0083) (0.0165) (0.0168) (0.0070) (0.0070)
F-stat 5456.60 1258.19 6325.69
R2 0.6699 0.0829 0.6640 0.0628 0.6743 0.1295
N 713674 713418 293673 293543 420001 419875
Robust standard errors are in parenthesis. All the regressions include month dummies and
household fixed effects. Columns (1), (3), and (5) show the results of OLS estimation, while
Columns (2), (4), and (6) present 2SLS estimation results. The R squared reported for 2SLS
estimations is the within R squared. F-stat is the statistic for the first-stage F test of excluded
instruments. *Statistically different from 0 at 10% significance; **Statistically different from
0 at 5% significance; ***Statistically different from 0 at 1% significance.
34
Table A8: Withdrawal cost - First stage results
Dep. variable: price paid
All Regular Light
(1) (2) (3)
Tax rate 0.7699∗∗∗ 0.7593∗∗∗ 0.7784∗∗∗
(0.0104) (0.0214) (0.0098)
F-stat 5456.60 1258.19 6325.69
χ2-stat 120.75 33.09 57.89
R2 0.0234 0.0120 0.1060
N 713418 293543 419875
Robust standard errors are in parenthesis. All the regressions include month dummies, the
lagged quantity, and the withdrawal cost dummy. F-stat is the statistic for the first-stage F
test of excluded instruments. ***Statistically different from 0 at 1% significance.
35
J Impact of smoking restrictions
Table A9: Impact of state restrictions - First stage results
Dep. variable: price paid
All Regular Light
(1) (2) (3)
Tax rate 0.7265∗∗∗ 0.7998∗∗∗ 0.6733∗∗∗
(0.0272) (0.0600) (0.0190)
F-stat 712.14 177.64 1255.90
χ2-stat 19.33 1.90 10.69
R2 0.0215 0.0114 0.1188
N 213470 96016 139159
Robust standard errors are in parenthesis. All the regressions include month dummies, the
lagged quantity, and the state restrictions dummy. F-stat is the statistic for the first-stage F
test of excluded instruments. ***Statistically different from 0 at 1% significance.
36
K Household heterogeneity
Table A10: Reinforcement coefficient - split samples
Full Sample Regular smokers Light smokers
Age≤52 0.24∗∗∗ 0.24∗∗∗ 0.25∗∗∗
> 52 0.32∗∗∗ 0.44∗∗∗ 0.21∗∗∗
Income per ≤$21,250 0.32∗∗∗ 0.41∗∗∗ 0.23∗∗∗
HH member >$21,250 0.19∗∗∗ 0.15∗∗∗ 0.21∗∗∗
Note: ***Statistically different from 0 at 1% significance.
Table A11: Impact of smoking restrictions on consumption - split samples
Full Sample Regular smokers Light smokers
Age≤52 -0.55∗∗∗ -0.21 -0.83∗∗∗
> 52 -0.41 0.04 -0.28
Income per ≤$21,250 -0.20 -0.19 -0.16
HH member >$21,250 -0.88∗∗∗ -0.63 -1.03∗∗∗
Note: ***Statistically different from 0 at 1% significance.
37