Date of Award

Summer 8-12-2016

Degree Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Economics

First Advisor

Dr. Rusty Tchernis

Second Advisor

Dr. Tom Mroz

Third Advisor

Dr. Charles Courtemanche

Fourth Advisor

Dr. Ian McCarthy

Abstract

This dissertation consists of three chapters analyzing risky health behaviors utilizing data from the Lung Health Study (LHS), a randomized smoking cessation program. The first two chapters of this dissertation analyze the effects of smoking on alcohol consumption and BMI, respectively. The third chapter studies whether and how much the objective smoking information, which is defined by clinicians, may be misreported.

The first chapter examines the effect of smoking on alcoholic beverage consumption. The epidemiology literature suggests that both behaviors affect similar brain regions and are commonly consumed together. So far, the economics literature has presented inconclusive causal evidence on the relationship. Building on the theory of rational addiction, I estimate the relationship between smoking and alcohol consumption using several different smoking measures. I report four salient findings. First, self-reported and clinically verified smoking variables suggest that quitting smoking lowers alcoholic beverages consumption by 11.5%. Second, cigarette consumption dating back 12 months affects alcohol consumption, and those with the highest past 12 months average cigarette consumption see the largest increase in alcohol consumption. Third, I find that the length of quitting affects future alcohol consumption as well. Continuously abstaining from smoking for 12 months reduces alcoholic beverage consumption by 27.5% per week. Fourth, non-smoking for 12 months also reduces the probability of drinking any alcoholic beverages by 31%.

The second chapter aims to identify the causal effect of smoking on body mass index (BMI). Since nicotine is a metabolic stimulant and appetite suppressant, quitting or reducing smoking could lead to weight gain. Using randomized treatment assignment to instrument for smoking, we estimate that quitting smoking leads to an average long-run weight gain of 1.8-1.9 BMI units, or 11-12 pounds at the average height. These results imply that the drop in smoking in recent decades explains 14% of the concurrent rise in obesity. Semi-parametric models provide evidence of a diminishing marginal effect of smoking on BMI, while subsample regressions show that the impact is largest for younger individuals, females, those with no college degree, and those with healthy baseline BMI levels.

The third chapter analyzes and compares self-reported and clinically verified smoking information. Descriptive statistics show that about 8% of clinically verified smokers self-report that they do not smoke (under-report participation), and that smoking cessation treatment group participants misreport smoking participation 2 to 1 relative to control group participants. In our first methodological approach we regard the objectively verified smoking measure as the gold standard. We estimate linear probability models and find that being male and married increases the probability of misreporting by 10 percentage points. Additionally, older participants are more likely to misreport smoking status, while those using nicotine gum and with a higher BMI are less likely to misreport. However, all variables can only explain a small fraction of the variation that explains misreporting. Our second methodological approach takes an agnostic view on whether the clinically verified smoking information is accurate. We utilize BMI, Carbon Monoxide (CO), and Cotinine level information to inform whether a person is a smoker. We estimate a Bayesian mixture model to account for the heterogeneity in BMI, CO and Cotinine levels after a substantial decrease in post treatment smoking participation. All of our models show that smokers are more likely assigned to the low BMI, high CO and high Cotinine level distributions. Among those classified as misreporters, we find that 30% have a very high probability of being part of the non-smoking distributions. As a result, we believe that objectively- verified smoking measure may not be better than the self-reported measure.

Share

COinS