NORC at the University of Chicago Alcoholism, Work, and Income Author(s): John Mullahy and Jody L. Sindelar Source: Journal of Labor Economics, Vol. 11, No. 3 (Jul., 1993), pp. 494-520 Published by: The University of Chicago Press on behalf of the Society of Labor Economists and the NORC at the University of Chicago Stable URL: http://www.jstor.org/stable/2535083 Accessed: 05-03-2018 14:39 UTC JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org. Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://about.jstor.org/terms NORC at the University of Chicago, Society of Labor Economists, The University of Chicago Press are collaborating with JSTOR to digitize, preserve and extend access to Journal of Labor Economics This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms Alcoholism, Work, and Income John Mullahy, Trinity College, Resources for the Future, and National Bureau of Economic Research Jody L. Sindelar, Yale University This article reports on an empirical analysis of the relationships between alcoholism and income and working. We show that the relationships between alcoholism and labor market success have important age or life-cycle dimensions. We present evidence that alcoholism may affect income more by restricting labor market participation than by affecting the wages of workers. Finally, we demonstrate that the effects of alcoholism on earnings depend on the extent to which one controls for other covariates associated with alcoholism; as such, we suggest that there may be important indirect as well as direct effects of alcoholism on labor market success. Individuals, besides, may sometimes ruin their fortunes by an excessive consumption of fermented liquors. [ADAM SMITH, The Wealth of Nations] An earlier version of this article was presented at the Alcohol and Public Policy session at the 1990 American Economic Association meetings. We would like to thank Phil Cook, Bill Evans, Michael Grossman, Paul Portney, Chris Ruhm, David Salkever, Paul Schultz, and seminar participants at Harvard University, Johns Hopkins University, University of North Carolina at Chapel Hill, and Vanderbilt University for helpful suggestions and comments on earlier drafts. National Institute on Alcohol Abuse and Alcoholism grant R01AA08394 to Yale University provided research support for both authors, while Mullahy's research was also supported in part by a University Fellows grant from Resources for the Future. [Journal of Labor Economics, 1993, vol. 1 1, no. 3] (C 1993 by The University of Chicago. All rights reserved. 0734-306X/93/1 103-000 1 $01.50 494 This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms Alcoholism, Work, and Income 495 I. Introduction It is widely believed that alcoholism is a major social problem with potentially important economic consequences.' Alcoholism is a prevalent disorder in the United States, affecting about one in 20 individuals at any one time and one in 10 individuals at some point during their lives. Males are three times more likely than females to suffer from alcoholism. A prevailing view on alcoholism is that it has depressant effects on earnings, income, and wages even after controlling for other important determinants of these labor market outcomes.2 Despite Adam Smith's early observation and the current popularity of this view, its veracity has been challenged by the hypothesis that increased alcohol consumption-at least within some range-might actually be productive.3 Some of this controversy is resolved immediately by distinguishing "alcohol consumption" and "alcoholism," but this distinction settles only part of the debate. The main purposes of this article are to demonstrate the complexity of the alcoholism-income issue and to provide some evidence about its structure. First, we demonstrate that the relationships between alcoholism and labor market success vary over individuals' lifetimes. Second, we present some evidence that alcoholism may affect income more by reducing employment probabilities than by reducing wages. Finally, we demonstrate how the magnitude and significance of the effects of alcoholism on earnings depend on the extent to which one controls for alcoholism-related co- variates. Before proceeding, it is useful to review some background information on alcoholism. Throughout the article we use "alcoholism" as a convenient term to summarize both the "alcohol dependence" and "alcohol abuse" disorders as defined by the American Psychiatric Association (1980, 1987). Although much about alcoholism remains unknown, most experts would agree on some general characteristics of alcoholics (see NIAAA 1990). For instance, it is now widely accepted that alcoholics come from all socioeconomic, demographic, and occupational groups. Moreover, about 10% of males and 3% of females are actively alcoholic at any point in time. For males, young adulthood is when symptoms of alcoholism are most prevalent; symptoms tend to decline after this peak.4 In addition, it is commonly observed that alcoholism tends to run in families, with the prevalent current ' U.S. Dept. of Health and Human Services, National Institute on Alcohol Abuse and Alcoholism (NIAAA 1990). 2 Berry and Boland (1977) and Rice et al. (1990) are representative of such find ings. 3 See Berger and Leigh (1988) for evidence of beneficial labor market effects o alcohol, Cook (1991) for additional discussion of such results, and Shaper et al. (1988) for a discussion of the findings of a positive health effect and some evidence explaining why the positive effects may be incorrect interpretations. 4 See Vaillant (1983), chap. 3, for an interesting discussion. This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms 496 Mullahy/Sindelar wisdom being that ther (see NIAAA 1990).5 Despite these broad agreements, there remain many lively debates on issues such as the etiology of alcoholism, the effectiveness of alternative treatment methods,6 and the life-cycle course of alcoholism problems. For instance, there have to date been insufficient longitudinal data to determine much about the lifetime course of alcoholism problems. More important for our purposes is the debate about whether alcoholism is, in a biomedical sense, a disease or, rather, is a complex set of health outcomes that arise from freely chosen behaviors. Alcoholism is widely considered in the medical literature to be a disorder with a significant genetic basis. However, such "disease models" have not always held sway and even today are not universally embraced.7 This background provides the setting for the remainder of the article. Section II discusses the data used in the empirical analysis and the characteristics of our sample. Section III presents a model and an econometric strategy for estimating the role of alcoholism in a human capital framework. Section IV provides some evidence on the age or life-cycle dimensions of the relationships between alcoholism and labor market outcomes. Section V presents a variety of econometric estimates of the role of alcoholism as 5 It might also be noted that alcoholism is generally estimated to be quite prevalent in the homeless population (Institute of Medicine 1988) as well as in some areas of the institutionalized population. The prevalence rate in the homeless population is suggested to be 20%-45% (NIAAA 1990). Since this analysis is based on a residential sample, the interesting and important issue of the effects of alcoholism in the homeless and institutionalized populations will not be treated here. To the extent that these "nonresidential" populations have greater than average propensities to be alcoholic and lower than average labor market success, our results would likely tend to underestimate the economic impacts of alcoholism in the entire population. 6 Evidence on the effectiveness of treatment is mixed (see, e.g., Hayashida et al. 1989; and NIAAA 1990). Indeed, lacking evidence on the efficacy of inpatient treatment, many third-party payers are no longer covering inpatient treatment. Many individuals seek repeated treatment and try a variety of different approaches, ranging from self-help groups, to inpatient group therapy, to drug treatment. Given remission after such sequences of treatment, it is not clear which, if any, treatment was effective, if the sequence itself mattered, or if individuals self-selecting into treatment were those relatively more likely to succeed. Moreover, according to a prevalent view, alcoholics who are not currently manifesting symptoms are "recovering" or "in remission," although they never completely recover; "once an alcoholic, always an alcoholic," even though symptoms may not be manifested currently. 7 See, e.g., Fingarette (1988) who maintains that the disease view has been perpetuated by various special interest groups that benefit from alcoholism's classification as a disorder. Perhaps not surprisingly, his perspective is the minority in the medical literature. The "rational addiction" theory of Becker and Murphy (1988) should also be noted in this context. This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms Alcoholism, Work, and Income 497 a determinant of labor market success. Section VI concludes with a discussion of the findings and of unresolved issues. II. Data and Sample Characteristics This analysis is based on wave 1 of the New Haven, Connecticut, site of the Epidemiological Catchment Area (ECA) survey conducted under the auspices of the U.S. National Institute of Mental Health (NIMH). The New Haven ECA survey is part of a larger NIMH-funded data collection effort surveying individuals 18 years old and older to assess psychiatric disorders in a population-based sample. The ECA surveys were designed primarily to assess the distribution of mental disorders in a community setting.8 Prior to the ECA surveys, there were no large U.S. samples assessing individuals' psychiatric disorders that contained reliable measures of these disorders, including alcoholism. The ECA data are particularly well suited for study of alcoholism, as they provide medically sophisticated measures of alcoholism and other mental disorders and, of particular importance for our study, information on labor market outcomes as well as socioeconomic and demographic characteristics of individuals and their households.9 Between 1980 and 1981, wave 1 of the New Haven survey was completed, yielding 5,034 observations, a 77.6% completion rate. The New Haven standard metropolitan statistical area (SMSA) was (approximately) the catchment area sampled, this area comprising 13 towns with a total adult population of 420,000. Two coordinate groups were sampled in this residential survey: all adults (18+), and individuals 65 and over. From the 5,034 observations in wave 1, we initially restrict our attention to males aged 22-64. The focus here is on males, both because they are far more likely to suffer from alcoholism than are females and because of the considerable body of accumulated research regarding the specification of earn- 8 For details on the ECA surveys, see Reiger et al. (1984), Eaton and Kessler (1985), and Robins et al. (1981). 9 We restrict attention in this analysis to the New Haven site because its data on labor market outcomes are much richer than the labor market data available from the other four survey sites: Durham, North Carolina; Baltimore, Maryland; St. Louis, Missouri; and Los Angeles, California. Prior to the ECA surveys, studies of the economic and social consequences of alcoholism had of necessity relied on data that were unsatisfactory in one way or another. Weaknesses have included selfdiagnoses of alcoholism, data obtained from individuals' visits to medical care facilities, unavailability of important covariates, and others. For instance, in an often-quoted study, Berry and Boland (1977) relied on a data set that included only household (not individual) income data and data on alcoholism for only one individual per household. Moreover, until the ECA surveys were conducted, assessments of disorders on the basis of state-of-the-art psychiatric diagnostic criteria were not available in large community data sets; mental health was typically assessed only by direct self-reporting of a diagnosis in large data sets. This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms 498 Mullahy/Sindelar ings models for males.10 As discussed further below, we select this age group in order to mitigate problems of incomplete educational spells (on the younger end) and retirement (on the older end). Given the oversampling of the elderly at the New Haven site, the upper-end restriction reduced our sample size considerably.11 Appendix A describes the dependent and independent variables, while table 1 displays the sample descriptive statistics. With the exception of alcoholism, the tables are largely self-explanatory."2 Alcoholism is measured in several ways, the two major definitions used here being whether or not the individual ever in his lifetime met the criteria for diagnosis of alcoholism (ALC-EVER = 1 or 0) and whether or not the individual who ever met the criteria had symptoms in the past year (ALC-YEAR = 1 or 0). Assessment of alcoholism and other mental disorders in the ECA is via a professionally designed survey instrument, the Diagnostic Interview Schedule (DIS), which conforms to the American Psychiatric Association DSM-JJJ and DSM-IIIR (Diagnostic Statistical Manual-3d ed., and 3d ed. rev., respectively) disorder criteria for diagnosis of alcohol abuse and alcohol dependence; see Appendix B for details. III. Earnings, Human Capital, and Alcoholism The tradition of including measures of individuals' health status as com ponents of their human capital in wage and earnings functions is well established.13 The basic framework posits an earnings function 10 See Willis (1986) for a good survey of earnings function estimation, and Mullahy and Sindelar (1990b, 1991) for a discussion of gender differences in the effects of alcoholism and other mental health problems. 11 The reduction in sample size from the original 5,034 observations to the 555 observations we use in much of the econometric analysis is due to the following set of restrictions: a) Nonelderly (ages 18-64): 2,458 remaining observations; b) Initial age cutoff (ages 22-64): 2,237 remaining observations; c) Restriction to estimation sample (ages 30-59): 1,420 remaining observations; d) Restriction to males: 604 remaining observations; e) Miscellaneous missing data: 555 remaining observations. 12 It might be noted, however, that the variables SCHOOLING and INCOME are created using interval midpoints. For SCHOOLING, 17 years was used for the open-ended upper interval "grad school." The 1980-81 survey asks the respondent to report income in the preceding year; income is thus expressed in 1979- 80 dollars. The variable INCOME consists of both labor income and other income "brought into" the household by the individual. For this measure, "0.5" was used for the bottom interval "less than $1,000," and "120" was used for the upper openended interval "over $100,000." While this approach is admittedly ad hoc, it greatly simplifies the econometrics. We present some evidence in table 6 that explicitly accounting for the censoring of the income measure yields results that differ little from those obtained using the "fill in the upper end" method. 13 See, e.g., Grossman (1972), Bartel and Taubman (1979, 1986), Mitchell and Butler (1986), and Frank and Gertler (1991). This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms Alcoholism, Work, and Income 499 Table 1 Sample Descriptive Statistics Mean Minimum Maximum Full Ages Full Ages Full Ages Variable Sample 30-59 Sample 30-59 Sample 30-59 Left-hand side: FULLTIME .764 .838 0 0 1 1 INCOME 20.073 23.423 .500 .500 120.0 120.0 INCOME* 22.704 25.068 .500 .500 120.0 120.0 INCOMEt 22.851 25.233 .500 .500 120.0 120.0 LOG-INCOME 2.717 2.912 -.693 -.693 4.787 4.787 LOG-INCOME* 2.959 3.067 -.693 -.693 4.787 4.787 LOG-INCOMEt 2.962 3.070 -.693 -.693 4.787 4.787 TRANSFER RECIPIENT .120 .090 0 0 1 1 FULL-NOTRANS .737 .809 0 0 1 1 Right-hand side: ALC-EVER .206 .204 0 0 1 1 ALC-YEAR .106 .101 0 0 1 1 ALC-PRE19 .080 .059 0 0 1 1 ALC- 1922 .064 .056 0 0 1 1 AGE 39.689 41.861 22 30 64 59 WHITE .861 .858 0 0 1 1 HEALTHY .887 .899 0 0 1 1 SCHOOLING 13.465 13.447 2 2 17 17 HIGH SCHOOL .465 .436 0 0 1 1 COLLEGE .359 .371 0 0 1 1 MARRIED .658 .723 0 0 1 1 OTHER INCOME 5.584 4.532 0.0 0.0 58.5 57.5 ANTISOCIAL PERSONALITY .090 .079 0 0 1 1 MENTALLY HEALTHY .891 .906 0 0 1 1 NOTE.-N = 897 for full sample; N = 555 for subsample ages 30-59. * Computed on subsample for which FULLTIME = 1, N = 685 or 465. t Computed on subsample for which FULLTIME = 1 and TRANSFER = 0; N 661 or 449. y = y(H, K, X)+c, (1) where y is some measure like log earnings; H is a vector of measures of the health components of human capital; K is a vector of nonhealth human capital measures (schooling, experience, etc.); X is a vector of other covariates (age, race, sex, etc.); and ? is a stochastic error, generally assumed to satisfy E(c I H, K, X) = 0. Following this tradition, we specify H = (A, S), where A is a vector of measures of alcoholism and S represents other health outcomes. The econometric counterpart to (1) is specified to be linear: Yt = zta + St, (2) where yt is log income; zt summarize Kt, Xt); and a is a vector of unknown This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms 500 Mullahy/Sindelar a probit equation (Heckman 1979) to describe the outcome of full-time participation in the labor market. The observed binary participation variable, It, is generated by a linear latent variable model whose error term is assumed to be homoscedastic normal, so that It = 1 (ztI3 + t > 0), where 1 (.) is the 0-1 indicator function. One standard result that will be of use below is that, in general, E (yt IIt = 1zt) zta + kt 7&zta, where Xt is the inverse Mills ratio under an assumption of normality for st, and y is an unknown scalar parameter that is a function of cov(s, q). It is assumed throughout that there are available N independent observations on (yt, It, Zt) That other components of human capital (schooling, experience, marital status, etc.) may be correlated with, and to some degree determined by, the health component(s) is usually ignored in the context of earnings function estimation. To the extent that the health components are structural determinants of the nonhealth components (e.g., if schooling attainment depends structurally on health status), then it is very easy to understate empirically the total productivity of health capital, that is, dy/dH, since indirect effects operating through the nonhealth components would not generally be captured. A fundamental point can be made in the context of the human capital framework sketched above (reinforced empirically by results presented below in table 5). Suppose for simplicity that y = y(A, K) + c with E(c I A, K) = 0, and consider how E(y I A, K) varies with A: dE(y IA, K)/dA = YA + YKdK/dA. (3) Equation (3) emphasizes what we consider to be a potentially important omission in many studies when measuring the productivity effects of alcoholism or, for that matter, of any disorder of interest. That is, the total effects of A on y are given not simply by the partial derivative YA that holds all else constant but rather by the total derivative that allows K to vary in response to variations in A. Accordingly, both direct (YA) as well as indirect (yKdK/dA) channels of influence must be admitted as possibilities if all the effects of A on y are to be evaluated and well understood in the design of policies targeted to mitigate or prevent alcoholism-related problems. For instance, see Mullahy and Sindelar (1989, 1990a, in press) and Cook and Moore (1991) for recent discussions of the relationships between alcohol use and educational attainment. It is also important to note that standard static earnings-health models like (1) typically assume that the health variables are econometrically exThis content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms Alcoholism, Work, and Income 501 ogenous. In a life-cycle context, such an assumption may be tenuous since health and labor market outcomes will be jointly determined in a health production context. Unfortunately, the data at our disposal do not suggest any reasonable instruments that would enable us to control for possible nonzero correlation between A and ?. Accordingly, mainly out of econometric necessity, we treat alcoholism as a predetermined or exogenous determinant of labor market and other sociodemographic outcomes, much like health status is typically treated in such models. This perspective on alcoholism is consonant with much of the medical literature that considers alcoholism a disease; whether this would be appropriate under a conscious choice perspective is a more tenuous matter. Our empirical results are thus reasonable to the extent that the exogeneity of alcoholism is a valid maintained assumption, but we readily admit that biased estimates of alcoholism's role in labor market success may arise should this assumption be invalid." Since whether one has ever had an alcoholism problem is less likely to be correlated with contemporaneous unobservables than whether one is currently drinking to excess, we focus primarily on ALC-EVER as the alcoholism measure.15 IV. Life-Cycle Dimensions to Alcoholism Problems This section compares how labor force participation and earnings profile vary by age for alcoholics (defined in various ways) and nonalcoholics. This life-cycle perspective provides some insights into why earlier studies have come up with conflicting results about the effect of alcohol consumption or alcoholism on earnings, income, and wages. While some studies have found negative effects (Berry and Boland 1977), others have found no significant effects (Benham and Benham 1982), and still others have found positive impacts (Berger and Leigh 1988; Cook 1991).16 In this analysis, we are particularly interested in the youngest and oldest age categories as it is these groups for whom seemingly counterintuitive 14 Even though our data do not allow us to meaningfully address the cause-effect issue in the empirical analysis, we offer the simple yet often overlooked point that a nonzero dK/dA implies that there is some correlation between K and A-causality may or may not be present. From a policy perspective, it is important to recognize that over the course of the life cycle A and K are likely to be interwoven in an intricate, complicated manner and that attributing "costs" to alcoholism without cognizance of such interrelationships-i.e., failing to account for possible indirect as well as the direct effects of alcoholism on earnings-is likely to lead to underestimates of such costs. 15 The use of the DSM-IIIR measure of alcoholism also helps justify its exogeneity in labor market equations relative to the DSM-JJJ measure, which includes symptoms of trouble at work (see App. B for additional discussion). 16 For instance, in studies of alcohol consumption (not alcoholism), Berger and Leigh (1988) and Cook (1991) find positive effects of alcohol consumption on earnings. This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms 502 Mullahy/Sindelar relationships between alcoholism and labor market success might arise. For the youngest group, alcoholism may tend to increase labor market participation and thus increase earnings. One mechanism consistent with such a finding would be that those with alcohol problems would be more likely to have trouble in school and either drop out of school or work more hours even while attending school part time. They would thus be working more hours and accumulating more labor market experience; increases in both hours and wage rates would thus tend to result in greater earnings. At some later point, however, the effects of nonalcoholics' greater educational attainment would overtake the experience advantages of the young alcoholics. The oldest age group could also exhibit seemingly inconsistent working and earning profiles. Over their lives, alcoholics may accumulate less financial capital (pensions, savings, etc.) than would nonalcoholics so that early retirement may not be a reasonable option. Instead, these aging alcoholics may continue employment later in life so that they may have higher contemporaneous labor market participation and income (but not necessarily wealth) as compared to their nonalcoholic counterparts who have begun to retire. Table 2 compares the probability that an individual is employed full time (i.e., whether the individual worked for pay all 12 months in the Table 2 FULLTIME Workers: Percentages by Age Group and Alcoholism Status ALC-EVER ALC-YEAR All Ages Observations 0 I 0 1 All ages .764 .775 719 .776 .663 N 897 712 185 802 95 Difference due to ALC = 1 relative to ALC = 0 -.056 -.113** Age subgroups: 22-29 .675 .652 .738 .678 .657 N 243 178 65 208 35 Decrease or increase due to ALC = 1 relative to ALC = 0 +.086 -.021 30-44 .845 .875 .733 .873 .643 N 348 273 75 306 42 Decrease or increase due to ALC = 1 relative to ALC = 0 -.142*** -.230*** 45-59 .826 .858 .684 .829 .786 N 207 169 38 193 14 Decrease or increase due to ALC 1 relative to ALC = 0 -.174** -.043 60-64 .566 .565 .571 .568 .500 N 99 92 7 95 4 Decrease or increase due to ALC = I relative to ALC = 0 +.006 -.068 ** p < .05, for two-tailed tests of differenc *** p < .01, for two-tailed tests of differen This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms Alcoholism, Work, and Income 503 previous year, vacations excepted) for alcoholics and nonalcoholics by age group. The top row in the table suggests that, for all ages combined, alcoholism (occurring either within the past 12 months [ALC-YEAR] or at any point in one's lifetime [ALC-EVER]) has a dampening effect on fulltime work propensity, although the difference is statistically significant only for alcoholism symptoms in the past year. The remainder of the table shows how these relationships vary across age categories. The second through fifth rows suggest that there are important differences in full-time work propensity by alcoholism status and also show how the effects of alcoholism vary by age. The youngest and the oldest age groups both show a positive labor force participation response to alcoholism measured by ALC-EVER but not ALC-YEAR. Note that the differences are not statistically significant and that the sample of alcoholics is quite small in the oldest group. Conversely, for what might be considered the prime-aged males, ages 30-59, the effects of alcoholism are generally negative, significant and quite large, as seen in rows 3 and 4 (the exception being for current symptoms [ALC-YEAR] for individuals aged 30-44). Interestingly, when considering current symptoms only (ALCYEAR), the differences are far less striking, suggesting the possibility that the damages associated with alcoholism are much more subtle, far-reaching, and indirect than simply whether or not an individual currently is manifesting symptoms. We take up an econometric examination of this issue in Section V. Table 3 displays the results of a similar analysis, with the focus now on income. The top row of the table demonstrates that, for our sample of males, the effects of alcoholism on personal income are negative and fairly large. This result holds whether one considers the full sample (cols. 2-5) or restricts attention to the sample of males who are full-time workers (cols. 7-10), although the effects are somewhat larger for the full sample. For full-time workers, income is likely to be a better proxy for earnings and, therefore, for productivity.17 Again, however, aggregation over the age-groups masks considerable heterogeneity in these relationships across the age-groups. Statistical significance aside for the moment, alcoholism (measured either way) appears to have little effect on the incomes of the youngest group, a positive effect for the oldest group, but important negative effects on the incomes of the prime-aged group in the middle. This simple yet revealing examination of the data suggests that the effects of alcoholism vary in important ways over the life cycle."8 Whether al- 17 In Sec. V we assess in greater detail this question about the role of alcoholism as a determinant of income when the focus is on full-time workers only. 18 Interpreting alcoholism as a "hazardous behavior," the finding that there are important age or life-cycle effects should not be surprising. See Ippolito (1981) and Ehrlich and Chuma (1990) for discussion. This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms ? 11 t14H NH mmg t1H H~8-118-1~~~~J ~ L C>A Lr Ln C rn t + - o~~~~~~0 0 0 c ceo I 6 ;~~~~~~C En n Go on Ln+(9 0 CC 11 - e m e t1 t 6 Cf~~~~~~~~)L c .0 C-~ L-t _ r i ? r t r n - ? 0 O "0 - - Ob 0~~~~~~~~~~~~~~~~~~~~~ ?0 $ l l ? e t a b l e'C K 10~ct ; o o N l i -e cc e c > u~~~~~~) C- CA r- _ l l rO, CI 'IC w 3:~~~~~~~~~~~~~~~~~~o ? C O e f'r)CIA O CIA C O CO * K1 0 11 0 11 0 11 0 11 K ? H~~ ~~~ ?' I? r? r? ccX -: 0 X O X 0 X O X O n X "0~~~~~~~~~~~~~ = O'v =-V :Dv =-V =-V X; t t) ~4 |J oft ? .18). Column 4 in table 7 considers an alcoholism interaction with physical health status. While the coefficient estimate on the interaction is not significant by usual standards, the point estimates (as they were for the HIGH SCHOOL variables) accord well with common sense: nonalcoholics in good health are best off, alcoholics in good health are next, nonalcoholics in poor health follow, while worst off are alcoholics in poor health. Wald test statistics for the joint significance of the linear and interaction ALCEVER terms and for these along with the linear HEALTHY term are 5.99 (p < .05) and 14.81 (p < .01), these corresponding to x2 variates with df = 2 and df = 3, respectively. Given our relatively small sample size, it is no small task to tease out second-order effects. Nonetheless, the results in table 7 are sufficiently strong to suggest-at least for schooling attainmentthat there is indeed some important interplay in how alcoholism and schooling jointly determine labor market outcomes. 25Cook and Moore (1991) consider how matriculation to and completion of college may be related to students' drinking behavior. 26 We also estimated several alternative versions of the model, where quadratics in SCHOOLING and interactions between SCHOOLING and AGE were included. In none of these specifications, however, were the SCHOOLING2 or the AGE X SCHOOLING coefficient estimates individually or jointly significant. These results are available on request. This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms Alcoholism, Work, and Income 513 Our final alternative specifications are motivated because individual income as recorded in the ECA represents income from all sourceslabor earnings, nonlabor earnings, transfer payments, and so on. Accordingly, we attempt to obtain a measure of income that is likely to be closer to a measure of earnings that, we feel, would better measure the productivity effects of alcoholism. To derive such a measure, we first take the subsample of individuals who are full-time workers. Then we identify the subsample of observations reporting that they received no transfer payments in the form of social security, disability, welfare, or unemployment compensation. The intersection of the subsamples that report full-time work and no transfer receipt (FULL-NOTRANS = 1) is a subsample for which we feel the reported individual income measure better approximates earnings.27 We model this process econometrically as a two-step Heckman selection process, with a probit model describing the determinants of FULL-NOTRANS, and a X-corrected LOG-INCOME model estimated on the subsample for which FULL-NOTRANS = 1. It should be stressed that, because all variables in the FULLTIME model (interpreted as a labor supply model) must of necessity be included in the earnings equation (interpreted as labor supplied times the wage rate), the sample selection process that drives the selection into the population of full-time workers is identified only by the nonlinear functional form of the selection-correction term, X. The results of this exercise are presented in table 8. For both the baseline (cols. 1 and 2) and the augmented (cols. 3 and 4) models, the results suggest that the direct effects of ALC-EVER on income are working more through participation effects than through wage/productivity effects. That is, the ALC-EVER effect is statistically much stronger in the equation determining FULL-NOTRANS than in the equation determining LOG-INCOME conditional on FULL-NOTRANS= 1.2s The general finding is consistent with the results in table 1, where the income differences between alcoholics and nonalcoholics are greater for all individuals as compared to the differences conditional on working full time. Although evidence from both tables 1 and 8 is consistent with this interpretation, there are clearly many alternatives to examine before such a conclusion could be confirmed. Moreover, we share the concern common in applied microeconometrics when the selection model for conditional 27 It might also be noted that when only full-time workers are sampled, earnings are more likely to proxy for wages since the variation in hours over the year is reduced considerably. 28 Interestingly, while its significance is quite low, the point estimate on ALCEVER in the LOG-INCOME model estimated conditional on FULL-NOTRANS = 1 differs little from its value when the full sample is used for estimation. This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms 514 Mullahy/Sindelar Table 8 Estimates for Fulltime Workers Receiving No Transfers: Probit FULL-NOTRANS and Heckman-Corrected LOG-INCOME for FULL-NOTRANS = 1 Reduced Form Full Model FULL-NOTRANS LOG- FULL-NOTRANS LOG= 1 INCOME = 1 INCOME (Probit) (OLS) (Probit) (OLS) Variable (1) (2) (3) (4) ALC-EVER -.402 -.151 -.286 -.181 (2.67) (.74) (1.82) (1.04) AGE .256 .104 .259 .230 (3.05) (.85) (2.99) (2.06) AGE2 -.003 -.001 -.003 -.003 (3.10) (.82) (3.06) (2.01) WHITE .497 .350 .255 .283 (2.92) (1.34) (1.37) (1.45) HEALTHY .836 .096 .764 .456 (4.36) (.19) (3.84) (1.08) SCHOOLING ... ... .056 .101 (2.34) (3.77) MARRIED ... ... .550 .542 (3.74) (2.31) OTHER INCOME ... ... -.028 -.033 (3.12) (2.55) X ... .218 ... 1.475 (.18) (1.50) CONSTANT -5.462 .410 -6.245 -4.581 (3.03) (.11) (3.26) (1.40) N 555 449 555 449 NOTE.-Subsample age LOG-INCOME is identified solely by the nonlinearity of the X term. At a minimum, however, the results are suggestive of some avenues for future research pursuits. The results that we present are in many respects not definitive. Other interpretations and confounding factors have not yet been eliminated. For example, alcoholism may itself be a symptom of deeper problems that may also result in reduced earnings, so that elimination of alcoholism per se would not necessarily imply that earnings would be of the same magnitude of otherwise similar individuals without alcoholism. Furthermore, the direction of causation cannot be determined with confidence; for example, lower earnings may certainly be a factor in the onset of alcoholism symptoms.29 Of course, the altogether separate possibility that un- 29 Vaillant (1983) quotes Enoch Gordis (now director of NIAAA): "Changes in personality or mood are now recognized to be largely the consequence of alcoholism, not its cause." This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms Alcoholism, Work, and Income 515 observed heterogeneity is ultimately driving all the outcomes must be admitted. VI. Summary and Discussion Studies of the effects of alcohol on earnings, income, and productivity have to date yielded conflicting results. The popular view that has been confirmed in several studies is that problem drinking has depressant effects on income. Recently, however, some studies have found insignificant effects or even positive effects of alcohol use. Part of the confusion owes to differences in the drinking measures used in these studies (alcohol consumption, alcoholism, etc.). That existing studies have employed different measures of labor market success (family income, individual income, individual earnings, wages) and/or have focused on different populations (e.g., workers only) only serves to compound the confusion. The results reported in this article have several important implications and, we feel, provide at least a partial resolution to some of these apparently conflicting results. One is that inferences about the effects of alcoholism on income depend critically on the age-group being studied. Moreover, income alone may not be an accurate measure of well-being: Those alcoholics who earn more in youth, withdraw from school, and work more hours are not necessarily better off. Similarly, the older alcoholic who may have greater income and less leisure time is not necessarily in a preferred position. Our results also suggest that alcoholism has a more significant impact on the likelihood of working than it does on how much earned when working (compare tables 2, 3, and 8). Studies may thus vary in their estimates of the impacts of alcoholism or alcohol consumption to the extent that their samples focus only on workers. We also have shown that the extent to which one controls for variables correlated with alcoholism (e.g., schooling and marital status) has a considerable impact on the estimated effects of alcoholism. For instance, based on the results in table 5, the inclusion of covariates correlated with alcoholism reduces the estimated effects of alcoholism on income from 31% to 17%. Estimates of the magnitude of the effect of alcoholism on earnings may thus differ across studies depending on the extent to which one controls for such covariates. The full effect of alcoholism may be estimated by omitting such correlated variables. However, one could equally well be interested in estimating the effect of alcoholism on earnings after controlling for the indirect effects. The different estimates correspond to different lines of inquiry. To summarize, while this article has not solved all margins of the alcoholincome puzzle, we feel it has suggested some important directions for future research. These would include-but not be limited to-further examination of alcoholism's life-cycle course, direct versus indirect effects This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms 516 Mullahy/Sindelar on labor market outcom and differential impacts on wages and participation. Appendix A Variable Definitions FULLTIME = 1 if individual worked 12 months for pay in previous year (including paid vacations), = 0 else INCOME = how much of household's total income before taxes for past year, including salaries, wages, social security, welfare, and any other income, was earned or brought in by individual ( .1,000) LOG-INCOME = natural log of INCOME TRANSFER RECIPIENT = 1 if individual reported receiving transfer payments in the form of social security, disability, welfare, or unemployment compensation, = 0 else FULL-NOTRANS = 1 if FULLTIME = 1 and TRANSFER RECIPIENT = 0, = 0 else ALC-EVER = 1 if symptoms of alcoholism present in past year if ever met the cr rion, = 0 else ALC-PRE19 = 1 if earliest symptoms of alcoholism present at age 18 or earlier, = 0 if earliest symptoms later or never ALC-1922 = 1 if earliest symptoms of alcoholism present between ages 19 and 22, - 0 if earliest symptoms at other time or never AGE = age in years AGE SQUARED = AGE squared WHITE = 1 if race is white, = 0 if race is nonwhite HEALTHY = 1 if individual reports physical health excellent or good, = 0 if reports reports fair or poor SCHOOLING = years of completed schooling HIGH SCHOOL = 1 if 12 < SCHOOLING < 15, = 0 else COLLEGE = 1 if SCHOOLING > 16, = 0 else MARRIED = 1 if currently married, = 0 else OTHER INCOME = other household income, measured as the greater of zero or household income minus personal income (in thousands) This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms Alcoholism, Work, and Income 517 ANTISOCIAL PERSONALITY= 1 if ever met criterion for antisocial personality, = 0 else MENTALLY HEALTHY = 1 if individual reports emotional/ mental health excellent or good, = 0 if reports fair or poor Appendix B Additional Details on the Definition of Alcoholism Assessment of alcoholism in the ECA is via a professionally designed survey instrument, the Diagnostic Interview Schedule (DIS). The DIS consists of a battery of questions on symptoms. These symptoms are used to obtain diagnoses consistent with the American Psychiatric Association's (APA) Diagnostic Statistical Manual (DSM) criteria for diagnoses of alcohol abuse and dependence. Alcohol abuse and dependence are two separate but related disorders according to the APA criteria. Diagnosis of alcohol dependence requires an individual to have symptoms in at least three of nine diagnostic criteria and diagnosis of alcohol abuse requires that they have symptoms in at least one of two categories (APA 1987). We focus on whether the individual met the criteria for either dependence or abuse or both, and refer to this throughout the article as "alcoholism." Questions in the DIS refer to symptoms such as had blackouts when drinking, heard things that weren't really there, had fits or seizures after cutting down on drinking, had the shakes, wanted to stop drinking but could not, continued to drink despite serious physical illness, needed a drink as soon as woke up, gone on benders for a couple of days, school or job troubles due to drinking, family objected to your drinking, fights while drinking, and arrested while drinking. If an individual ever met the criteria for alcohol abuse or dependence, then the age at which he or she first had any symptom and age at which he or she last had a symptom (e.g., in the last year) were recorded. From this information we formulated two variables that are used in much of the analysis and two variables that we use in sensitivity analysis; all four are binary variables. The two main variables are ALC-EVER and ALC-YEAR. The variable ALC-EVER indicates whether the individual ever had the cluster of symptoms that met the criteria for alcohol abuse and dependence. The variable ALC-YEAR indicates for those who ever met the criteria for diagnoses whether they had suffered from any symptoms in the past year. The other two variables, ALC-PRE19 and ALC-1922, are indicators of when (i.e., before age 19 or between ages 19 and 22) the first symptoms occurred for only those who ever met the criteria. The APA criteria have typically been used in a clinical setting by psychiatrists. Only recently with availability of the ECA data have diagnoses of mental health disorders been made in a survey of the general population. Using general population data avoids the selfselection problem in which only individuals who seek treatment are This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms 518 Mullahy/Sindelar observed. Using the DIS diagnostic approach reduces the potential for self-reporting biases with regard to alcoholism. That is, individuals are not merely asked "Are you an alcoholic?" or "Do you suffer from alcoholism?" as has been the case in some other surveys. Nonetheless, the input information for the DIS is provided by the respondent and thus may involve response bias. Fortunately, the ECA's DIS-based diagnosis of alcoholism has been found to have good correspondence with alternative diagnostic approaches (Anthony et al. 1985). The DSM-III standards (APA 1980) for alcohol abuse and dependence were used in coding the ECA data. However, we revised the diagnosis of alcoholism using the symptom list in conjunction with the more recent DSM-IIIR criteria (APA 1987). This revision could have been important since the DSM-IIIR definition drops some of the labor market behavior symptoms used in diagnosis (see above), clearly a problem when using the diagnosis as an explanatory variable in labor market outcome models. However, for our category of alcohol abuse and/or dependence, the designation of "alcoholic" did not change for any observation in our sample for any of the four definitions discussed above. References American Psychiatric Association. Diagnostic and Statistical Manual of Mental Disorders. 3d ed. Washington, D.C.: American Psychiatric Association, 1980. . Diagnostic and Statistical Manual of Mental Disorders. 3d ed., rev. Washington, D.C.: American Psychiatric Association, 1987. Anthony, James C., et al. "Comparisons of the DIS and a Standardized Psychiatric Diagnosis." Archives of General Psychiatry 42 (July 1985): 667-75. Bartel, Ann P., and Taubman, Paul. "Health and Labor Market Success: The Role of Various Diseases." Review of Economics and Statistics 61 (February 1979): 1-8. "Some Economic and Demographic Consequences of Mental IIIness." Journal of Labor Economics 4 (April 1986): 243-56. Becker, Gary S., and Murphy, Kevin M. "A Theory of Rational Addiction." Journal of Political Economy 96 (August 1988): 675-700. Benham, Lee, and Benham, Alexandra. "Employment, Earnings, and Psychiatric Diagnosis." In Economic Aspects of Health, edited by V. Fuchs. Chicago: University of Chicago Press, 1982. Berger, Mark C., and Leigh, J. Paul. "The Effect of Alcohol Use on Wages." Applied Economics 20 (October 1988): 1343-51. Berry, Ralph E., and Boland, James P. The Economic Costs ofAlcohol Abuse. New York: Free Press, 1977. Cook, Philip J. "The Social Costs of Drinking." In Expert Meeting on Negative Social Consequences of Alcohol Use. Oslo: Norwegian Ministry of Health and Social Affairs, 1991. This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms Alcoholism, Work, and Income 519 Cook, Philip J., and Moore, Michael J. "Drinking and Schooling." Paper presented at the Third Annual Health Economics Workshop, Johns Hopkins University, Baltimore, May 1991. Eaton, William W., and Kessler, Larry. Epidemiological Methods in Psychiatry: The NIMH Epidemiologic Catchment Area Program. New York: Academic Press, 1985. Ehrlich, Isaac, and Chuma, Hiroyuki. "A Model of the Demand for Longevity and the Value of Life Extension." Journal of Political Economy 98 (August 1990): 761-82. Fingarette, Herbert. Heavy Drinking. Berkeley and Los Angeles: University of California Press, 1988. Frank, Richard G., and Gertler, Paul J. "Assessment of the Measurement Error Bias for Estimating the Effect of Mental Distress on Income." Journal of Human Resources 26 (Winter 1991): 154-64. Grossman, Michael. "On the Concept of Health Capital and the Demand for Health." Journal of Political Economy 80 (April 1972): 223-55. Hayashida, Motoi, et al. "Comparative Effectiveness and Costs of Inpatient and Outpatient Detoxification of Patients with Mild-to-Moderate Alcohol Withdrawal Syndrome." New England Journal of Medicine 320 (February 9,1989): 358-65. Heckman, James J. "Sample Selection Bias as Specification Error." Econometrica 47 (January 1979): 153-61. Institute of Medicine. Homelessness, Health, and Human Needs. Washington, D.C.: National Academy Press, 1988. Ippolito, Pauline M. "Information and the Life Cycle Consumption of Hazardous Goods." Economic Inquiry 19 (October 1981): 529-58. Mitchell, Jean M., and Butler, J. S. "Arthritis and the Earnings of Men: An Analysis Incorporating Selection Bias." Journal of Health Economics 5 (March 1986): 81-98. Mullahy, John, and Sindelar, Jody L. "Lifecycle Effects of Alcoholism on Education, Earnings, and Occupation." Inquiry 26 (Summer 1989): 272- 82. ."An Ounce of Prevention: Productive Remedies for Alcoholism." Journal of Policy Analysis and Management 9 (Spring 1990): 249-53. (a) "Gender Differences in the Effects of Mental Health on Labor Force Participation." In Research in Human Capital and Development: Female Labor Force Participation, edited by I. Sirageldin. Greenwich, Conn.: JAI, 1990. (b) "Gender Differences in Labor Market Effects of Alcoholism." American Economic Review 81 (Papers and Proceedings) (May 1991): 161-65. "Direct and Indirect Effects of Alcoholism on Human Capital." Milbank Quarterly (in press). Murphy, Kevin M., and Welch, Finis. "Empirical Age-Earnings Profiles." Journal of Labor Economics 8 (April 1990): 202-29. Powell, James L. "Symmetrically Trimmed Least Squares Estimation for Tobit Models." Econometrica 54 (November 1986): 1435-60. This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms 520 Mullahy/Sindelar Reiger, Darrel A., et al. "The NIMH Epidemiologic Catchment Area (ECA) Program: Historical Context, Major Objectives, and Study Population Characteristics." Archives of General Psychiatry 41 (October 1984): 934-41. Rice, Dorothy P., et al. The Economic Costs of Alcohol and Drug Abuse and Mental Illness: 1985. PHS/ADAMHA DHHS Publication No. (ADM) 90-1694. Washington, D.C.: U.S. Government Printing Office, 1990. Robins, Lee N., et al. "National Institute of Mental Health Diagnostic Interview Schedule: Its History, Characteristics, and Validity." Archives of General Psychiatry 38 (April 1981): 381-89. Shaper, A. G., et al. "Alcohol and Mortality in British Men: Explaining the U-Shaped Curve." Lancet (December 3,1988), pp. 1267-73. U.S. Department of Health and Human Services, National Institute on Alcohol Abuse and Alcoholism. Seventh Special Report to the U.S. Congress on Alcohol and Health. PHS/ADAMHA/NIAAA DHHS Publication No. (ADM) 90-1656, Washington, D.C.: U.S. Government Printing Office, 1990. Vaillant, George E. The Natural History of Alcoholism: Causes, Patterns, and Paths to Recovery. Cambridge, Mass.: Harvard University Press, 1983. Willis, Robert J. "Wage Determinants: A Survey and Reinterpretation of Human Capital Earnings Functions." In Handbook of Labor Economics, vol 1, edited by 0. Ashenfelter and R. Layard. Amsterdam: NorthHolland, 1986. This content downloaded from 147.251.185.127 on Mon, 05 Mar 2018 14:39:30 UTC All use subject to http://about.jstor.org/terms