Access Type

Open Access Dissertation

Date of Award


Degree Type


Degree Name




First Advisor

Allen C. Goodman






May 2010

Advisor: Dr. Allen Goodman

Major: Economics

Degree: Doctor of Philosophy


Attempt of this research to explore the impact of alcohol use disorders on labor market outcomes is justified on the following grounds. First, alcohol use disorders (alcohol abuse and dependence) became more prevalent among the working age population over the decades (NIAAA: 2004). Second, existence of a large body of research on the nature and extent of relationship between alcohol use and labor market outcomes remained debatable, heterogeneous and counterintuitive with various explanations although it is generally agreed that harmful effects attributed by alcohol consumption may results in short-run and long-run physical and mental impairments, and it may entail enormous economic and non-economic costs to the society.


The main focus of this research is to obtain the consistent estimates of the impact of alcohol use disorders on labor market outcomes. This also examines to what extent performance of individuals with alcohol use disorders (alcohol abuse and dependence) differ from abstainers and individuals who had no alcohol use disorders (ex-drinkers, ex-abuser/dependents, new-drinkers, and ex-new-drinkers) in terms two labor market outcomes: the probability of full time work participation and annual earnings (annual personal income).


The estimation strategies are as follows. First, using the NESARC 2001-02 survey sample of 43,093, I estimate a Logit model by the Maximum Likelihood Estimation (MLE) method where the dependent variable is an indicator variable (LFPFULL) for full-time labor force participation (Chapter 3, Section 3.1). Second, using a sample of 21406 (the observations for individuals who had job), I estimate a model with the logarithm of annual earnings (LINCOME) by the Ordinary Least Square (OLS) method. In these estimations, I include a set of explanatory variables: six binary variables to indicate alcohol use status (see Tables 2A and 2B), and other binary variables for other personal and socio-demographic characteristics (see Table 4), such as individual's age, gender, race, marital status, education level, health status, location (whether individual lived in a central city or not), other source of income, and work related characteristics (industry type). Third, I perform formal tests to detect the existence of potential problems in estimations, the endogeneity and heteroscedasticity in the sample. The estimation strategy (appropriate estimation method) to address endogeneity is required to consider the potential presence of heteroscedasticity in the sample since the sample set is cross-sectional (see details in Chapter 3, section 3.2). Fourth, as the statistical tests confirmed the existence of both problems (endogeneity and heteroscedasticity) in current estimation, I re-estimate labor market outcome equations by addressing endogeneity and heteroscedasticity using GMM-IV method (proposed by Baum et al: 2003, Amemiya: 1985 and Foster: 1997).


As the statistical tests confirmed the endogeneity of alcohol use related variables and the presence of heteroscedasticity in the sample, the estimation by applying GMM-IV are expected to produce consistent and efficient estimate compared to baseline estimates (from the MLE and OLS ) which do not address the endogeneity and heteroscedasticity problems. Thus, the estimated results by MLE and OLS methods could be biased and not reliable. The followings are the key results of GMM-IV estimations.

The key result of GMM-IV regression of LFPFULL is that the marginal effects for alcohol abuse and alcohol dependent are -0.38 and -0.04 respectively (though not significant). indicating that alcohol abusers and alcohol dependents have 38 percent and 4 percent less probability of being employed full-time respectively than life-time abstainers (base category), holding all other variables constant. The marginal effect of being ex-drinker on LFPFULL is negative (statistically significant) and consistent with expectation. Ex-drinkers have 43 percent less probability of being employed full-time than life-time abstainers (base category). The positive marginal effects of being ex-abuser/dependents, new drinkers and ex-new-drinkers on LFPFULL (not significant) are 11 percent, 45 percent and 2 percent respectively though not significant.

The key result of the GMM-IV regression of LINCOME is that the marginal effect for alcohol abuse on annual income is -112,057 (though not significant). It indicates that alcohol abusers earned $112,057 less than life-time abstainers (base category), holding all other variables constant. The marginal effects of all other alcohol use status indicating binary variables (ex-drinker, ex-abuser/dependent, new-drinker, ex-new-drinker and alcohol dependents) are found to have positive impacts on annual income (statistically significant for ex-drinkers and new-ex drinkers). The positive impact on annual income for being alcohol dependent is inconsistent with expectation.

Compared to GMM-IV estimates, the MLE (without addressing endogeneity and heteroscedasticity) underestimate the effects of alcohol use variables (represent by binary variables) and overestimate the effects of other socio-demographic variables on labor market variables, and the OLS (without addressing endogeneity and heteroscedasticity) underestimate the effects of all explanatory variables on labor market outcomes.

Some observed unexpected results should be treated with cautions considering the limitations of this research: there might be measurement errors in proxy earnings variable, there were data limitations on previous drinking record of individuals who had alcohol use disorders and some labor market information such as hours of work and loss of working hours (absence from the job) due to alcohol use. Besides these limitations, overall results are largely consistent with the results that observed in parallel labor and health economics literature and can be considered representative since this research used rich and nationally representative NESARC data source. The results of this study can be useful for policy and management research to face the challenges of having and maintaining productive and healthy work force. The results imply the necessity of adopting clear and well communicated policies concerning recruitment, monitoring, early prevention and access to effective treatment, and maintaining positive work environment. The results also imply that public or private policies addressing related issues of alcohol use and employment should take into account the fact that women react differently to the amount of alcohol consumption and their work decision also different than men.

Included in

Economics Commons