Smoking Cessation and Body Weight:Evidence from the Behavioral RiskFactor Surveillance SurveyPanagiotis Kasteridis and Steven T. YenObjective. To investigate the role of smoking cessation in body weight.Data Sources. 2004–2005 and 2009–2010 Behavioral Risk Factor Surveillance Surveys (BRFSS) (N = 349,000), Centers for Disease Control and Prevention; Tax Burdenon Tobacco (Orzechowski and Walker 2010).Study Design. The Gaussian treatment effect model is estimated for three age categories by gender. Treatment effects of quitting smoking on body mass index (BMI) by quitlength are calculated.Principal Findings. Quitting is found to be endogenous. Differentiated effects ofquitting smoking on BMI are found among quitters by gender, between age groups,and by length of time since quitting smoking, and positive association between smoking cessation and body weight confirmed. Declining smoking rates have only a modesteffect in the overweight population. The effects of quitting on BMI are considerablylower among younger men and women.Conclusion. The price that must be paid, in terms of weight gain, to enjoy the healthbenefits of smoking cessation is trivial even for the obese population.Key Words. BMI, BRFSS, smoking cessation, treatment effect modelThere have been a rapid rise in obesity and a notable decline in smoking ratesin the United States over the last few decades. The obesity epidemic spreadrapidly during the 1990s across all states, regions, and demographic groups(Mokdad et al. 1999), and the prevalence has remained high, exceeding 30%in most age and gender groups during 2007–2008 (Flegal et al. 2010).Obesity is a major public health concern because it is associated with along list of diseases such as type 2 diabetes, hypertension, dyslipidemia, certainforms of cancer, sleep apnea, and osteoarthritis. Overweight accounts for morethan 350,000 premature deaths each year in the United States, second only totobacco-related deaths (Mokdad et al. 2005). The accelerating spread of obesity has placed a tremendous burden on health care costs. The total direct and© Health Research and Educational TrustDOI: 10.1111/j.1475-6773.2012.01380.xRESEARCH ARTICLE1580Health Services Researchindirect costs attributable to overweight and obesity amounted to $117 billionin 2000 (U.S. Department of Health and Human Services [USDHHS] 2001).The Centers for Disease Control and Prevention (CDC) estimates that46 million people or 20.6% of all adults (age 18) in the United States werecurrent cigarette smokers during 2008 (CDC 2009a). Cigarette smoking is theleading cause of preventable death in the United States, accounting for approximately 443,000 deaths or 1 in every 5 deaths each year (CDC 2009b,c). Prevention of smoking initiation and cessation of smoking have become thenational objectives to reduce morbidity and mortality and lower medical costs.Indeed, over the past 30 years, there has been a substantial decline inthe proportion of adult smokers across all sociodemographic sub-populations.From 1955 to 2007, U.S. cigarette smoking rates fell from 57 to 22 percent inmen and from 28 to 17 percent in women, with an overall rate of 20 percentfor both genders in 2007 (Giovino et al. 2009). The last decade saw the smallest declines in cigarette smoking rates. CDC analyzes data from the 2008National Health Interview Survey (NHIS) which indicate that during 1998–2008, the proportion of U.S. adult smokers declined by 3.5 percentage points—from 24.1 to 20.6 percent.The declining prevalence of cigarette smoking among adults has beenattained by banning smoking in the workplace (Evans, Farrelly, and Montgomery 1999), restaurants and bars (ABC News, World News Tonight 2005),intensive anti-smoking campaigns in the media (Flynn et al. 1995), and taxincreases (USDHHS 2000). Although the overall health benefits of quittingsmoking are unquestionable, the opposite trends of obesity and smoking inthe United States have raised an important concern about an unintended effectof anti-smoking policies on obesity rates. Chou, Grossman, and Saffer (2004,p. 585) claim that rising obesity is an example of “the price that must be paidto achieve goals that are in general favored by society.” Indeed, the associationbetween smoking and body weight has become a central issue in the obesityliterature, but the accumulating evidence is conflicting.In this study, we investigate whether quitting smoking leads to increasing body weight and to what extent. We compile data for current and formersmokers (i.e., quitters) from the 2004–2005 and 2009–2010 Behavioral RiskFactor Surveillance Survey (BRFSS). The vehicle for our analysis is the treatAddress correspondence to Steven T. Yen, Ph.D., Professor, Department of Agricultural andResource Economics, The University of Tennessee, 302 Morgan Hall, Knoxville, TN 37996-4518,e-mail: syen@utk.edu Panagiotis Kasteridis, Ph.D., Post-doctoral Associate, is with the Department of Agricultural and Resource Economics, The University of Tennessee, Knoxville, TN.Smoking Cessation and Body Weight 1581ment effect model (Barnow, Cain, and Goldberger 1980). To quantify theeffect of smoking cessation, we calculate treatment effects for sub-samples ofindividuals. Our goal is to examine whether smoking cessation affects bodymass index (BMI) differently by gender, between age groups, and among individuals with different lengths of time since quitting smoking.LITERATUREMuch research has been conducted using the National Health and NutritionExamination Survey (NHANES) data. Albanes et al. (1987) investigate theassociation between smoking and body weight using NHANES II. They findthat cigarette smokers weigh less and are leaner than nonsmokers, andex-smokers are not heavier or fatter than nonsmokers. Flegal et al. (1995)employ data from phase 1 of NHANES III (1988–1991) and conclude thatsmoking cessation is associated with a small increase in the prevalence of overweight, although the effect was much larger among smokers who had quitwithin the past 10 years. Using more recent (1999–2002) NHANES data, Flegal (2007) finds that even substantial decreases in cigarette smoking have onlya small effect, generally less than 1 percent, on increasing the prevalence ofobesity and decreasing the prevalence of healthy weight in the population.Earlier evidence of the link between smoking cessation and weight gainincludes Coates and Li (1983), Manley and Boland (1983), Klesges et al.(1989), Shimokata, Muller, and Andres (1989), Moffatt and Owens (1991),Williamson et al. (1991), Klesges et al. (1997, 1998), Froom, Melamed, andBenbassat (1998), Mizoue et al. (1998), Froom et al. (1999), and Hudmonet al. (1999).The main limitation of the above studies is that inferences are potentiallybiased by unmeasured factors that simultaneously affect smoking or quittingand body mass. If, for example, unobserved personal traits induce individualsto smoke and consume more calories, the estimated effect of smoking on bodyweight will be biased unless endogeneity of the smoking decision is accountedfor. The absence of a mechanism in modeling endogeneity of smoking haschallenged researchers attempting to confront it in various ways.O’Hara et al. (1998) estimate weight gains associated with smoking cessation in the Lung Health Study (1986–1994), a clinical trial which randomized smokers into a control group and an intervention group who received12 weeks of behavioral intervention. Eisenberg and Quinn (2006) update theestimated weight gain upward using participation in the intervention program1582 HSR: Health Services Research 47:4 (August 2012)as an instrument. Main limitation of their analysis is lack of individual-leveldata, which would allow calculation of standard errors for the instrumentalvariable (IV) estimates. Fang, Ali, and Rizzo (2009) study the relationshipbetween cigarette smoking and obesity using data from the 2006 China Healthand Nutrition Survey and an IVestimation procedure to control for endogeneity. They find a moderately negative relationship between cigarette smokingand BMI. Their quantile regression estimates reveal a weak associationbetween smoking and BMI among subjects at the high end of the BMI distribution, but the association is considerably stronger among subjects in thehealthy weight range. Chen, Yen, and Eastwood (2007) examine the relationship between smoking and BMI employing a simultaneous-equation systemallowing for censoring and endogeneity of cigarettes smoked. They claim thatthe negative relationship between smoking and BMI reported in the literatureis attributable to simultaneity and should be interpreted with caution. A shortcoming of the study is lack of identification strategy, as no instrument is usedin the cigarette smoking equation, which may have contributed to statisticalinsignificance of the effect of smoking on BMI, despite the OLS estimates suggesting otherwise.Another line of research uses data from the BRFSS (1984–1999) to estimate how much of the trend in obesity is explained by state-specific factors,including the price/tax of cigarettes. Chou, Grossman, and Saffer (2004,2006) employ a state fixed effects model to estimate the impact of cigaretteprice on BMI. Results link the upward trend in obesity to declining smokingrates. Gruber and Frakes (2006), who control for the effects of unmeasuredtime-varying variables with time dummies instead of a time trend, criticizefindings by Chou, Grossman, and Saffer (2004, 2006) on the grounds that thestate-specific price of cigarettes is endogenous as it may be driven by marketfactors, which affect both smoking and eating. They use an alternative pricevariable, state excise tax on cigarettes, and obtain a strikingly different resultfrom that of Chou, Grossman, and Saffer (2004, 2006)—a negative relationship between cigarette tax and BMI implying that reduced smoking lowers,rather than raises, body weight. The estimated effects in both of these studiesare, as Gruber and Frakes (2006, p. 194) suggest, “implausibly large.” Gruberand Frakes (2006) find that individuals who quit smoking are 56 percent lesslikely to be obese, while smoking one fewer pack of cigarettes per day lowersthe odds of obesity by 40 percent. The results of Chou, Grossman, and Saffer(2004, 2006) are also enormous, but in the opposite direction. These mixedresults reported in the literature call for a more in-depth analysis of the effectsof quitting smoking on body weight. In this study, we take a slightly differentSmoking Cessation and Body Weight 1583approach from that of Chou, Grossman, and Saffer (2004, 2006) and Gruberand Frakes (2006)—we investigate the role of quitting smoking directly, ratherthan by way of cigarette prices or tax which determine smoking, in weightchanges by estimating a treatment effect model commonly used in programevaluation.CONCEPTUAL FRAMEWORKOur empirical specification is motivated by a simplified consumer utility maximization theory, similar to that in Yen, Chen, and Eastwood (2009), Philipsonand Posner (2003), and Schroeter, Lusk, and Tyner (2008). Conditional onsocio-demographics, lifestyle, and environmental factors such as state regulations on smoking in public places, an individual derives utility from bodyweight and levels of food, cigarettes, and other goods consumed. Body weightis a function of food and cigarettes consumed, conditional on socio-demographic and lifestyle variables. Then, maximizing the utility function subject toan income constraint produces the equations estimated in this study: an optimum weight equation along with a cigarette smoking equation that is endogenous to the system. Instead of smoking, we estimate a binary quitting equation.Prices of food are not available, but regional and intertemporal variations infood prices are reflected in the regional and state variables used (discussednext). Price of cigarettes is an important variable and, drawing on Gruber andFrakes (2006), we use state excise tax on cigarettes as a proxy for price.METHODEconometric ModelTreatment effect models have a long history of uses in program evaluation(Barnow, Cain, and Goldberger 1980). The model features a binary endogenous (treatment) variable di for quitting smoking (henceforth, “quitting”) byindividual i, which is modeled as probit di ¼ 1 if zi0a þ ui > 0¼ 0 if zi0a þ ui 0ð1Þ and appears as a regressor in the outcome equation for BMI (yi):log yi ¼ xi0b þ d di þ vi ð2Þ1584 HSR: Health Services Research 47:4 (August 2012)In (1) and (2), zi and xi are vectors of explanatory variables, a and b arevectors of parameters, d is a scalar parameter, and the error terms (ui,vi) aredistributed as bivariate normal with zero means, variances (1,r2), and correlation ρ. The log-transformed dependent variable ameliorates potential nonnormality and heteroscedasticity of the error term (Yen and Rosiński 2008). Apartfrom the logarithmic transformation of yi, (1) and (2) represent the recursivemodel with qualitative and continuous variables considered by Maddala andLee (1976, p. 527).To investigate the differentiated effects of length of time since quittingsmoking (henceforth, “quit length”) on BMI, the treatment parameter d isparameterized as a linear function of quit-length dummy variables wi (withparameter vector c):d ¼ w0ic ð3ÞThis specification amounts to interacting the dummy endogenous variable di with wi. The model can be estimated by a two-step or maximum-likelihood (ML) procedure (Maddala and Lee 1976, pp. 527–529). We use the moreefficient ML procedure, by maximizing the logarithm of the sample likelihoodfunction for an independent sample of n observations (Maddala and Lee 1976,p. 528) L ¼ Yirrð4Þ ni¼1y11u log yi xi0b diwi0c U ð2di 1Þ zi0a þ q ðlog yi xi0b diwi0cÞ=r” # ð1 q2Þ1=2 where u(·) is the probability density function and Φ(·) is the cumulative distribution function, both of the standard normal distribution. The hypothesis ofexogenous treatment amounts to the parametric restriction of zero error correlation (ρ), under which the exogenous model can be estimated by separateprobit for (1) and OLS for (2) treating the quitting variables wi as exogenous.This nested hypothesis can be tested with a standard procedure such as likelihood-ratio (LR), Lagrange multiplier (LM), or Wald test (Engle 1984).Effects of quitting at different lengths on BMI can be calculated from themeans of BMI conditional on smoking and quitting (Yen and Rosiński 2008):Eðyijdi ¼ 0; a; b; r; qÞ ¼ expðxi0b þ r2=2Þ U½ðzi0a þ qrÞ=Uðzi0aÞ ð5ÞEðyijdi ¼ 1; a; b; r; qÞ ¼ expðxi0b þ wi0c þ r2=2Þ Uðzi0a þ qrÞ=Uðzi0aÞ ð6ÞSmoking Cessation and Body Weight 1585The average treatment effect (ATE) of quit-length category j is an estimate of the expected gain from quitting for a randomly chosen individual inthat category (for j = 1,…,J ):ATE ¼ n1 Xni¼1½Eðyijwij ¼ 1; ^ a; b^;^c; r^; q^Þ Eðyijdi ¼ 0; ^ a; b^; r^; q^Þ ð7Þwhere (^ a; b^;^c; r^; q^) are ML estimates for the corresponding parameters andwij is the jth element in wi. For statistical inference, standard errors of treatment effects are calculated by the d-method (Rao 1973, p. 388).Data and SampleOur dataset contains individual-level information on BMI, smoking andquitting, education and income levels, lifestyle, and employment statusfrom the 2004–2005 and 2009–2010 Behavioral Risk Factor SurveillanceSystem (BRFSS). The BRFSS is implemented with collaborative effort ofthe CDC and state health departments. Interviewees were randomlyselected to represent U.S. civilian, noninstitutionalized adults age 18.We use the core component of BRFSS questionnaire, which consists of astandard set of questions including queries about current health-related perceptions, conditions, and behaviors (e.g., health status, diabetes, healthinsurance, and tobacco use) as well as demographics. An important focus ofthe investigation was the effect of quitting smoking, with different quitlengths, on BMI changes. Therefore, choice of the sample years was dictated by responses to an important question: “About how long has it beensince you last smoked cigarettes regularly?,” with a response ranging fromless than 1 month to 10 years or longer. Such information was collected for2004 and 2005, suspended for 2006–2008, and available again for 2009–2010. The original survey contained 1,215,314 observations from 2004–2005, 2009, and the first half of 2010.1 We focus on current and formersmokers, so individuals who had never smoked cigarettes are excluded, asare observations with missing values in important explanatory variables.Also excluded are individuals from Guam, Puerto Rico, and Virgin Islands,pregnant women, and those who “don’t know,” “refuse to answer,” or hadBMIs exceeding six standard deviations above the mean. Restricting thesample to individuals age 18–64 leads to 349,000 observations for analysis.Note that although we use an unusually large sample from multiple years ofBRFSS, it is not possible to construct a panel. A panel sample would allow1586 HSR: Health Services Research 47:4 (August 2012)examination of the dynamics of smoking, quitting, and accompanyingweight changes.To identify the model parameters and treatment effects, our IV for quitting is a cigarette tax variable.2 The Tax Burden on Tobacco (Orzechowskiand Walker 2010) provides state-level data on excise taxes of cigarettes. Weuse the yearly data along with dates of changes in taxation rate to constructmonthly state cigarette taxes. These state tax data are merged to the BRFSSsample by state of residence and month (interview time). Over 2004–2010,average state excise tax increased from 73 to 136 cents per pack, the largestincreases observed in Wisconsin and Rhode Island (175 cents). Connecticut,Washington, and Rhode Island had the highest cigarette tax rates in 2010(300, 302.5, and 346 cents), while South Carolina, Missouri, and Virginia hadthe lowest (7, 17, and 30 cents, respectively). These dramatic differences in cigarette taxes among states and the sharp increase in average tax rate over ashort period ensure enough variation in our identification variable. In addition, dummy variables for eight regions (reference = Pacific) are included inthe quitting equation and 50 states (reference = Minnesota) in the BMI equation.3The outcome variable, BMI, is a primary measure of obesity and is calculated as weight in kilograms divided by height in meters squared (kg/m2).As weight and height in the BRFSS are self-reported, bias is likely to arise.Cawley (2004), however, finds no discernible differences between results fromusing self-reported and predicted BMIs, in a different context, viz., effects ofobesity on wages. The other endogenous variable is a binary indicator indicating whether the individual had quit smoking. Technically, quitters are individuals who had smoked at least 100 cigarettes in their entire lives and theiranswer to the question “Do you now smoke cigarettes every day, some days,or not at all?” is “not at all.”4 As noted above, this potentially endogenousdummy indicator is interacted with dummy variables indicating seven categories of quit length (see Table 2; and online tables), which allow calculation ofthe effects of quit length on BMI.We perform the analysis dividing the sample into three sub-samples foreach gender: “young females” (age 18–40), “middle-aged females” (41–55),and “old females” (56–64); and likewise for males. Average BMIs are 26.82(26.99) for young female (male) smokers and 26.90 (28.00) for young female(male) quitters. Average BMIs are only slightly different among old female(26.78) and male (27.23) smokers, but notably higher among the old female(28.60) and male (29.07) quitters. About 39.2 percent (40.8 percent) of theSmoking Cessation and Body Weight 1587young females (males) are quitters. The percentage of old females (males) whoquit is 63.8 percent (69.7 percent).The definitions and sample statistics of all variables are presented intables in the online appendix and we summarize key figures. There are notable differences in the explanatory variables between smokers and quittersacross the six sub-samples. For example, among young males, average household incomes are $47,866 for smokers and $65,531 for quitters.5 Among oldmales, 9 percent of smokers reported excellent health compared to 16 percentof quitters. In the old female sample, the percentage of college graduates is 18percent among smokers and 35 percent among quitters. Among youngfemales, 42 percent of smokers and 63 percent of quitters and are married.Remarkable differences are also seen on employment, race, age, and healthcoverage.State cigarette taxes for quitters are higher than that for smokers. Forinstance, young female smokers reside in states that tax cigarettes at an average of $0.97 per pack and young female quitters in states with average tax rateof $1.10. Importantly, with a coefficient of variation of over 65 percent for allsamples, the data provide ample variations in the tax rate variable to explainthe quitting decision.RESULTSModel EstimationWe estimate the treatment effect model for the six samples by ML method.We estimate the log-transformed model and find it preferable to the untransformed alternative (not reported) for each sample using a nonnested specification test.One important empirical issue relates to uses of valid IVs (viz., properexclusion restriction[s]) to identify the model parameters and treatmenteffects. For IV estimation, parameter identification requires at least one variable that is correlated with the endogenous variable, uncorrelated with errorterm of the outcome equation, and does not affect the outcome conditional onits regressors (Angrist, Imbens, and Rubin 1996). For ML estimation of thecurrent model, however, nonlinear identification criteria are met withoutexclusion restrictions owing to distributional assumption of the error terms.Nonlinear functional form inherent in the distributional assumption, however,often fails to generate sufficient variation to identify the model parameters soit is capricious to rely solely on distributional assumptions for identification.1588 HSR: Health Services Research 47:4 (August 2012)To avoid over-burdening the nonlinear functional form for parameter identification, we impose exclusion restrictions. Specifically, a state cigarette tax variable, discussed above, is included solely in the quitting equation since there isno a priori expectation that it would directly affect BMI. French and Popovici(2011, p. 137) note that cigarette taxes are “by far the most popular IV used inthe literature to estimate the effects of smoking on … health” (Mullahy andPortney 1990; Leigh and Schembri 2004) and adult obesity (Rashad 2006),and that they “prove to be excludable from the structural equation for a variety of dependent variables.” Also included uniquely in the quitting equationare dummy variables indicating regions.Validity of these IVs is supported by testing for their joint significance inthe quitting equation, estimated separately by ML method for the purpose ofthese tests. The hypothesis of weak instruments, for cigarette tax separatelyand jointly with the regional dummies, is rejected by likelihood-ratio (LR)tests, at the 1 percent significance level for all but the old males sample(Table 1). For this old males sample, the tax rate variable turns significantwhen regional variables are excluded from the quitting equation. This resultprovides a guidance for our estimation of the treatment effect model for oldmales, viz., for which regional variables are excluded from the treatment equation.6 The above test for weak IVs is similar in spirit to that of Staiger andStock (1997) for a more conventional model; also see French and Popovici(2011). We then test for exogeneity of quitting, and the results are alsoTable 1: Likelihood-Ratio Tests for Weak Instrument(s) and ExogenousTreatmentSampleWeak Instrument: TaxRatesWeak Instruments: TaxRates and Regions Exogenous TreatmentLR (df = 1) p-value LR (df = 8) p-value LR (df = 1) p-valueFemalesAge 18–40 6.77 0.009 453.21

- Assignment status: Already Solved By Our Experts
*(USA, AUS, UK & CA PhD. Writers)***CLICK HERE TO GET A PROFESSIONAL WRITER TO WORK ON THIS PAPER AND OTHER SIMILAR PAPERS, GET A NON PLAGIARIZED PAPER FROM OUR EXPERTS**

**NO PLAGIARISM**– CUSTOM PAPER