Do drinkers earn less?

I. Introduction

Theoretical work by Becker and others on human capital led initially to empirical research on the determinates of earnings. Interest in human capital has now spread to other areas, especially health. This movement is in conjunction with societal concerns regarding the role of diet and environmental factors on human well-being and performance. Following Becker [1] and Grossman [9], economists expect to find a significant relationship between health and earnings. Health aspects of human capital have been extensively explored since the early work of Grossman [9; 10]. Good health is both beyond the control of the individual (exogenous) as a result of genetic factors and random events, and controllable (endogenous) through the regulation of activities such as smoking, drinking, eating, exercise and other informed choices. This study examines the relation between earnings and drinking.

Curiously, little rigorous empirical work has been done on the relation between earnings and drinking. Irving Fisher's claim to the contrary,(1) there is little evidence that prohibition led to substantial, if any, productivity increases. Although it is established that excessive drinking leads to poor health, the empirical evidence regarding earnings and drinking is mixed and inconclusive. According to the U.S. Government, abusive drinkers earn less. In their biennial reports to Congress, the National Institute on Alcohol Abuse and Alcoholism (NIAAA) estimates the "economic cost to society" of alcohol abuse. Approximately half of this cost arises from the estimated loss in earnings (productivity) due to abusive drinking.(2) Their approach consists of estimating a wage equation which includes human capital variables as well as a dichotomous abusive drinking variable and quantity of alcohol consumed.(3)

Berger and Leigh [3] used a sample selection model to test whether drinkers earned more or less than nondrinkers. They divided their sample (from the 1972-73 Quality of Employment Survey) into two categories: drinkers and nondrinkers. Using a sample selection correction estimator, hedonic wage regressions for each category (and by sex) were estimated. They found that drinkers (moderate and abusive combined) earned more than nondrinkers. Comparisons were then made between wages from each regression with demographic and human capital variables set at the same levels. The drinkers earned more in all cases. Model [21] found positive and significant effects on income and wages for both moderate and heavy alcohol use. Model used the National Household Survey on Alcohol and Drug Abuse: 1984. Heien and Pittman [13] using the same data and specification as the 1984 NIAAA study, but a different econometric procedure found a positive, but not significant, effect of alcohol on earnings. Manning et al. [20] found that alcohol consumption had no effect on days lost from work or on visits to the doctor. Work by Kenkel and Ribar [16] on alcohol use by young adults shows mixed effects for men versus women. Recent work by French and Zarkin [8] and Heien [12] develop and test models along the lines discussed below.

An alternative approach is to examine the effect of psychiatric diagnoses on earnings, since alcoholism is defined in those terms. Using data containing information on DSM (Diagnostic and Statistical Manual) criterion, Benham and Benham [2] found that alcoholism was not a statistically significant factor in either earnings or employment. The effects were mixed in sign. The most recent NIAAA study(4) uses the Epidemiologic Catchment Area data base and employs the DSMALC criterion for alcoholism. By this definition, alcoholics were found to have lower earnings.(5) Mullahy and Sindelar [23] used the DSMALC criterion and found similar results. Excellent reviews of the work in this area can be found in Cook [6] and Mullahy [22].

These previous studies have either used hedonic regressions and/or sample partitioning methods to test for the effect of alcohol on earnings. As noted above, these efforts have yielded conflicting results. While there is considerable reference to the well known negative effects of alcohol abuse,(6) there is little consideration given to medical findings on the effects of moderate drinking or abstinence. Interestingly, the medical literature also has a good deal to say concerning the effects of moderate drinking and abstinence.

The medical research, discussed in more detail below, indicates that both abusive drinkers and abstainers are at significantly greater risk for heart attack than moderate drinkers. Some of the problems of the previous economic analysis can perhaps be understood in the light of this research. Assuming that heart disease is related to income (also discussed below), then both abstainers and abusers will have lower incomes than moderate drinkers. The results obtained by researchers when attempting to use sample partition methods or linear regression will depend on the sample "mix." If there are large numbers of abusers, or abstainers, or both, income will be negatively related to alcohol consumption, as some studies have shown. If there are large numbers of moderate consumers, alcohol consumption will be positively related to income, as some studies have shown. If the sample is fairly balanced between abusers and abstainers and moderates, then alcohol will have no effect on income, as some studies have shown.

This paper attempts to demonstrate that the medical literature has been largely disregarded in specifying the human capital models of the relation between alcohol consumption and earnings. Furthermore, when this literature is considered in the specification, the empirical results are in line with the medical findings.

II. Alcohol and Health: A Look at the Medical Literature

Although it is well known that excessive drinkers have health problems, what is less well known is that moderate drinkers apparently have above average health - above the average of both abusers and nondrinkers alike. In the past decade the evidence for this effect has been further refined and its basis better understood.

One of the main points of this refinement relates to the type of heart disease under consideration. Drinkers have less coronary artery disease (CAD)(7) and fewer strokes due to blocked blood vessels. Heavy drinkers suffer from cardiomyopathy (heart muscle disease), hypertension (high blood pressure), hemorrhagic stroke and rhythm disturbances. Lastly, it should be noted that CAD is by far the largest heart disease problem in terms of cause of death. The second point which recent studies have clarified is the role of other factors such as age, gender and lifestyle habits, most importantly smoking.

It is useful to point out that there are two classes of those who do not drink at all. First, there are those who have never drunk and currently do not drink, either out of religious conviction or personal preference. Second, there are also those who have drunk, often quite heavily, in the past. Not infrequently, these individuals have health problems, partly or wholly as a result of past drinking patterns, and are now abstaining. The latter group frequently has medical problems not present in the former. Hence this research distinguishes between nondrinkers who have never drunk, termed Lifetime Abstainers, and Ex-Drinkers.

The notion of the convex relation between drinking and cardiovascular mortality risk was challenged by Shaper [27]. He maintained that selective migration from drinking to ex-drinking by individuals at risk resulted in this relation. He cited as additional evidence that a subset of subjects who were free of baseline risk factors show no relation between alcohol use and mortality. Several recent studies have controlled for the effect of ex-drinkers as well as baseline risk factors. Prominent among these is the Harvard study by Rimm et al. [26], Boffetta and Garfinkel [4], DeLabry et al. [7], Klatsky, Friedman, and Siegelaub [17], Klatsky, Armstrong, and Friedman [18], and Kono et al. [19]. In a study of 51,000 male health professionals Rimm et al. [26] found the relative risk of heart disease declined significantly over a range of ethanol intake from none to 30 grams per day.

The study by Klatsky, Friedman, and Siegelaub [18] covered 129,170 male and female members of Kaiser Permanente Health plan over the 1979 to 1985 period. This study had three main findings. First, ex-drinkers had higher CAD rates due to the confounding of alcohol related traits, mainly smoking and gender. Second, the U-shaped curve is not due to selective abstinence by individuals at higher risk. Third, the findings indicate a protective effect against CAD of alcohol used in moderation.(8) All studies were prospective in that they measured the trait (i.e., alcohol consumption) before the occurrence of the health event (e.g., heart attack). As well as confirming the U-shaped effect, the DeLabry study (men only) also offered evidence that moderate drinkers have greater life expectancy. A recent study Razy et al. [24] uses a sample composed entirely of women. Previous studies had concentrated mainly on men. This study also confirmed the convex relation and greater longevity.

III. Theoretical Considerations

The following model, which is a modification of the one found in Grossman [9], is used to provide a formal statement of the theory. Individuals are assumed to possess a multiperiod utility function of the form,

U = U([[Phi].sub.0][H.sub.0],..., [[Phi].sub.n][H.sub.n], [Z.sub.0],..., [Z.sub.n]),(1)

where [H.sub.0] is the "inherited" stock of health capital, [H.sub.i] is the stock of health capital in the ith period, [[Phi].sub.i] is the coefficient which converts capital stock into a flow, and [Z.sub.i] is a vector of commodities, and services in the ith period. The health stock is governed by the standard identity

[H.sub.i] = [H.sub.i-1] + [I.sub.i] - [Delta][H.sub.i], (2)

where [I.sub.i] is gross investment in health and [Delta] is the depreciation rate. Health investment is given by the household production function

[I.sub.i] = [I.sub.i]([M.sub.i], T[H.sub.i] : [E.sub.i]) (3a)

where [M.sub.i] is medical care, T[H.sub.i] is time spent investing in health care, and [E.sub.i] is human capital. The production of [Z.sub.i] is given by,

[Z.sub.i] = [Z.sub.i]([X.sub.i], [T.sub.i]: [E.sub.i]) (4)

where [X.sub.i] is a vector of purchased inputs, including alcohol, and [T.sub.i] is the time spent producing [Z.sub.i]. The consumer faces a wage rate, W, and as a result has two constraints to deal with. The first is the lifetime budget constraint 

[summation of] [(1 + r).sup.-i][P.sub.i][X.sub.i] where [infinity] to i = 1 = [summation of] [(1 + r).sup.-i] T[W.sub.i][W.sub.i] + N[W.sub.0] where [infinity] to i = 1 (5)

where [P.sub.i] is the price of [X.sub.i], TW is hours worked and N[W.sub.0] is net worth in the base period. The second is the simple time constraint 

T[W.sub.i] + T[L.sub.i] + T[H.sub.i] + [T.sub.i] = [Omega] (6)

where TL is time lost due to ill health, and [Omega] is total time available. The consumer now maximizes (1) subject to the various constraints.

Rather than face an externally given wage rate, the representative earners are instead assumed to face a wage reaction function or a hedonic wage relation where they can, to some extent, endogenize their wage rate by their choice of education, occupation, and health. This function specifies the wage rate as a function of age, education, region, occupation, marital status, ethnic background, family size, union membership, and various measures of health. These health-influencing behaviors include whether or not the individual smokes, the amount of alcohol consumed, and whether or not the individual has a major health problem. This function permits a unique wage determination for the individual under consideration. The specification of this wage function is,

[W.sub.i] = [W.sub.i]([H.sub.i], [E.sub.i], [D.sub.i]) (7)

where W, H and E are as above and D represents predetermined variables such as region, occupation and marital status. The production of health investment relation is now modified to allow for specific health effects, or

[I.sub.i] = [I.sub.i]([M.sub.i], T[H.sub.i], [A.sub.i], [S.sub.i], [O.sub.i] : [E.sub.i]) (3b)

where A is the amount of alcohol consumed, S is the amount of smoking activity, and O is other health related problems. Combining (2) and (3b) and substituting into (7) yields,

[W.sub.i] = [W.sub.i]([M.sub.i], T[H.sub.i], [A.sub.i], [S.sub.i], [O.sub.i], [D.sub.i], [E.sub.i]). (8)

Equation (8) now gives the wage rate as a function of medical care, hours spent investing in health care, alcohol consumed(9), whether or not the individual smokes, other health problems, various predetermined choice variables such as occupation, marital status, etc. and human capital, mainly as measured by education.

It is well established that good health is an important determinant of earnings. One important distinction made by Grossman was that while increases in human capital increase worker productivity and hence wages, good health not only increases productivity, but also the amount of time one can work. It is also clear that CAD will diminish many dimensions of health as they relate to earnings. There will be more time lost due to recuperation, longevity will be shorter, and disability, resulting in both time lost and less productivity while on the job, will be greater.

Equation (8) is, of course, only one of several relations yielded by the consumer optimization process. Maximization of (1) subject to the appropriate time constraints will yield demand relations for the various inputs (the [X.sub.i]'s) for each time period. For the problem at hand one important relation will be the demand for alcoholic beverages, or

[A.sub.i] = [A.sub.i]([W.sub.i], [M.sub.i], T[H.sub.i], [P.sub.i], [S.sub.i], [O.sub.i], [D.sub.i], [E.sub.i]). (9)

IV. Empirical Analysis

Since equation (9) is stochastic and [A.sub.i] appears on the RHS of the wage determination relation (8), there will be simultaneous equation bias. One solution to the problem would be to include the price of alcoholic beverages in (8) and (9). There would then be no simultaneous equation estimation problem. However, the data sets, discussed below, which contain the necessary information to test this hypothesis, are medical data sets and do not contain information of prices paid for inputs such as alcohol. While it might be possible to use BLS area prices indices, they are only computed for large geographical areas and do not adequately reflect micro price behavior. Also, they tend to have the effect of regional dummy variables. Since the crux of the simultaneous problem is embedded in (8) and (9) it is necessary to estimate both relations by an appropriate simultaneous equations estimator.

This section presents the results of estimating the earnings function and the demand for alcohol by Non-Linear Three Stage Least Squares (NL3SLS). The hypothesis to be tested is that the relation between earnings and alcohol consumption will be a concave quadratic function with both classes of abstainers and heavy drinkers earning less than moderate drinkers. This hypothesis is based on two considerations. First, the well known convex curve for CAD discussed earlier shows that moderate drinkers will have less heart disease and as a result will live longer and more productive lives. Second, the other positive health effects of moderate drinking, also discussed above, will tend through the same mechanism to produce higher earnings than those of abstainers or abusive drinkers. These considerations in conjunction with the knowledge that heavy drinking has severe negative health impacts will also produce an earnings curve which is quadratic in alcohol consumption.

In order to test this hypothesis a data base which contains health information, alcohol consumption, demographic data, as well as economic information is required. Health information is needed to correct for compounding factors such as smoking and medical problems. Information on alcohol consumption is needed so that moderate consumption can be differentiated from heavy consumption and both classes of abstinence. Data must be available on education and various demographic variables. Last, the data base must also contain information on the respondents income and hours worked.

One survey which satisfies this criterion is the National Household Survey on Alcohol Use (NHSA). This survey is sponsored by the NIAAA and is conducted by the Alcohol Research Group in Berkeley, California. This survey is conducted every five years and the surveys from 1979 and 1984 are used in this study. The 1990 survey is not yet available. This survey contains data on various socioeconomic factors such as marital status, age, occupation, region, education, number of children, religious preference, health, drinking practices (frequencies and amounts), and household income (but not hours worked).(10) The 1979 survey contains 1772 observations, 243 of which were dropped due to missing information on income. The 1984 survey contains 3828 observations. Both surveys differentiated between lifetime abstainers and ex-drinkers.

As noted above it can be argued that most of the variables appearing on the right hand side of (8) are exogenous. Variables such as age, sex, etc. are clearly so. Other choice variables such as education, occupation, and marital status were selected in the past, so that they were choice variables at one time but are now also given. However, the variables representing the amount of alcohol consumed are endogenous and are jointly determined along with income and other consumption decisions. Hence, estimation of (8) by OLS will introduce simultaneous equation bias due to the presence of the demand for alcohol relation.

Table I presents the results of applying NL3SLS to each of the two data sets mentioned above. Table I gives the results for equation (8).(11) The model is identified by the presence of a large number of religious preference variables in the demand for alcohol relation. Many of these variables were significant in the demand for alcohol relation. Significant positive variables were Presbyterian, Episcopalian, 'no religious preference,' and Orthodox Russian. Significant negative ones were Buddhist, Jewish, and Muslim.

Table I. Regression Results: NL3SLS Quadratic Functional Form

Data Base                       NHSA: 1979        NHSA: 1984
Dependent Variable            Annual Earnings   Annual Earnings

Intercept                         -20311.          -11288.
                                     4.4               6.2

Ex-drinker                        -1493.1          -388.4
                                       .7              .3

Number of Drinks                    332.9           128.2
                                      2.2             2.4

Number of Drinks Squared             -3.1            -.61
                                      2.3             2.6

Suburban                           5300.9          2423.9
                                       5.             4.8

Age                                 581.3           483.8
                                      5.8             6.4

Age x Age                           -5.78           -5.54
                                      6.5             6.8

Married                             760.4           689.7
                                       .5            10.8

Divorced                           -5410.          -1651.
                                      3.1             2.5

Widowed                            -3433.            779.
                                      1.6              .9

Education                           1175.           772.5
                                     12.5            10.4

Children                             406.              NA
                                       .4

Medical Problem                    -276.1              NA
                                       .5

Sex: 1 if male                       3124.         1880.8
                                       1.3            2.5

Mean-Dependent Variable            18014.4        17345.8

Sample Size                           1529           3838

Drinking Impact Test                   9.8            4.6

Notes:

Numbers below regression coefficients are t-ratios.

Not shown are coefficients for occupations and ethnicity (seven
categories for NHSA).

The second column of Table I gives the results for (8) for the 1979 National Survey. Both age and education are highly significant determinants of earnings. This is consistent with general human capital findings. Other significant variables are marital status, suburban and sex. In addition to the variables shown in Table I there were four regional variables, ten occupational classes, and seven ethnicity classes. The occupational variables tended to be significant, as did some of the ethnicity ones. The relations presented in Table I were estimated with the significant regional and ethnicity variables retained.

The third column of Table I gives the results for the National Household Survey on Alcohol for 1984 for relation (8). These results are similar to those from the 1979 survey. The effects of alcohol consumption are similar to the 1979 results with the exception that excessive drinkers do not show a significant drop in income. Ex-drinkers earn less than lifetime abstainers in both surveys. Human capital variables, age and education, are again highly significant.

In both of the regressions the linear and quadratic terms are significant. The impact of the drinking variables was tested with the standard F-test for a restricted subset using the 2SLS estimates. This test consisted in running each regression from Table I with none of the drinking variables (linear, squared, and ex-drinker terms). This is the restricted regression. This model was then compared to the full model. The results of this test, given under Drinking Impact Test in Table I, show that the drinking variables add significantly to the regression in every case.

V. Conclusions

This paper attempts to show how knowledge of medical research can assist in specifying the effect of health related variables in a human capital model. Specifically, the paper uses medical findings to specify the relationship between alcohol consumption and earnings. This relationship is quadratic and recognizes the distinction between two classes of abstainers: ex-drinkers and lifetime abstainers. The empirical estimates lend support to the hypothesis that moderate drinkers earn more than either abstainers or abusive drinkers. The paper presents estimates of the quadratic relation using data from two separate surveys. Both are large micro based surveys with considerable economic and epidemiological detail.

The effect of moderate alcohol consumption on earnings is statistically significant, but not as significant as the effect of traditional human capital variables such as education or age. The effect of drinking on earnings is statistically significant in both models. The results for weekly earnings imply that total hours worked is also an inverted U-shaped function of alcohol consumption.

Knowledge of the medical literature not only aided in the specification of the model, it also suggests an explanation for reconciling contradictory results obtained by other researchers. Some previous studies have found alcohol to have negative effects on earnings while other studies have found the opposite. These studies have not allowed for curvilinear, or inverted U shaped, effects. Hence it is conjectured here that the effects found, positive or negative, depended on the "sample mix" of heavy drinkers and abstainers relative to moderate drinkers in conjunction with the linear specification. Previous studies have, for example, used a dichotomous variable to differentiate between moderate consumers and abusers. They also failed to differentiate between abstainers and moderate drinkers or distinguish between classes of abstainers.

1. In Prohibition at Its Worst (1926), Fisher claimed that prohibition would increase national productivity by 5%.

2. These lost earnings are not to be confused with the income losses suffered by individuals involved in auto accidents with drunken drivers. Those costs, for both drunken drivers and their victims, are another category of the "economic cost to society." The government does not distinguish between external and internal costs in their methodology.

3. The methodology described here was used in the 1980 estimates. See Harwood et al. [11]. This methodology is criticized in Helen and Pittman [13].

4. See Rice et al. [25].

5. The DSMALC criterion (for alcoholism) and the NIAAA criterion (for alcohol abuse) are very different. According to the NIAAA, 2 drinks or greater per day is alcohol abuse.

6. In some cases, such as cirrhosis of the liver, there is a clear physiological relationship. In other cases, there are difficulties disentangling the effects of alcohol from chronic smoking and other lifestyle effects.

7. The main causes of CAD are smoking, high blood pressure, diabetes, and a fatty diet. Death usually results from heart attack due to blood vessel blockage.

8. The study also found that the protective effect of alcohol was more pronounced for women.

9. A more rigorous formulation would replace the quantity of alcohol consumed with input prices, especially those for alcoholic beverages. This problem is discussed below.

10. This survey is described in detail in Clark and Midanik [5].

11. The estimates of (9), which are not of primary interest here, are available on request. The results did show that income is not a significant determinate of alcohol consumption. Previous studies substantiate this finding. Heien and Pompelli [14] found small negative income elasticities for alcoholic beverages. Johnson and Oksanen [15] found extremely small positive and negative income elasticities for alcoholic beverages. Both studies employed cross section data so that trends and habit were not confounded with income effects. The demand for ethanol is even less income elastic as it is widely recognized that consumers shift up to higher quality beverages as income increases. In this sense, alcohol is much like other food products.

References

1. Becker, Gary. Human Capital. Chicago: University of Chicago Press, 1964.

2. Benham, Lee and Alexandra Benham. "Employment, Earnings, and Psychiatric Diagnosis," in Economic Aspects of Health, edited by V. Fuchs. Chicago: University of Chicago Press, 1982.

3. Berger, Mark C. and J. Paul Leigh, "The Effect of Alcohol Use on Wages." Applied Economics, 1988, 1343-51.

4. Boffeta, Paolo and Lawrence Garfinkel, "Alcohol Drinking and Mortality among Men Enrolled in an American Cancer Society Prospective Study." Epidemiology, September 1990, 342-48.

5. Clark, William B. and Lorraine Midanik. "Alcohol Use and Alcohol Problems among U.S. Adults: Results of the 1979 National Survey," in Alcohol Consumption and Related Problems, Alcohol and Health Monograph No. 1, DHHS Publication No. (ADM) 82-1190. Washington: U.S. Government Printing Office, 1982.

6. Cook, Philip. "The Social Costs of Drinking," Chapter 2 in The Negative Social Consequences of Alcohol Use. Oslo, Norway: Norwegian Ministry of Health and Social Affairs, 1990.

7. DeLabry, Lorraine O., Robert J. Glynn, Michael R. Levenson, John A. Hermos, Joseph S. LoCastro, and Pantel S. Vokonas, "Alcohol Consumption and Mortality in an American Male Population: Recovering the U-Shaped Curve-Findings for a Normative Aging Study." Journal of Studies on Alcohol, 1992, 25-32.

8. French, Michael T. and Gary A. Zarkin, "Is Moderate Alcohol Use Related to Wages? Evidence from Four Worksites." Journal of Health Economics, 1995, 319-44.

9. Grossman, Michael, "On the Concept of Human Capital and the Demand for Health." Journal of Political Economy, 1972, 223-55.

10. -----. The Demand for Health: A Theoretical and Empirical Investigation. Occasional Paper no. 119. New York: Columbia University Press (NBER), 1972.

11. Harwood, Hendrick J., Diane M. Napolitano, Patricia L. Kristansen, and James J. Collins. Economic Costs to Society of Alcohol and Drug Abuse and Mental Illness. Prepared for the Alcohol Drug Abuse, and Mental Health Administration, RTI/2734/00-01FR. Research Triangle Park, N.C.: Research Triangle Institute, 1984.

12. Heien, Dale M., "The Relationship between Alcohol Consumption and Earnings." Forthcoming in Journal of Studies on Alcohol.

13. ----- and David Pittman, "The Economic Costs of Alcohol Abuse: An Assessment of Current Methods and Estimates." Journal of Studies on Alcohol, 1989, 567-79.

14. Heien, Dale M. and Greg Pompelli, "The Demand for Alcoholic Beverages: Economic and Demographic Effects." Southern Economic Journal, January 1989, 759-70.

15. Johnson, James A. and Ernest H. Oksanen, "Estimation of Demand for Alcoholic Beverages in Canada from Pooled Time Series and Cross Sections." Review of Economics and Statistics, February 1977, 113-18.

16. Kenkel, Donald S. and David C. Ribar. "Alcohol Consumption and Young Adults' Socioeconomic Status." Brooking Papers on Economic Activity: Microeconomics 1994. Washington: The Brookings Institution, 1994.

17. Klatsky, Arthur L., Gary D. Friedman and A. B. Siegelaub, "Alcohol and Mortality: A Ten Year Kaiser Permanente Experience." Annals of Internal Medicine, 1981, 139-45.

18. Klatsky, Arthur L., Mary Anne Armstrong, and Gary D. Friedman, "Risk of Cardiovascular Mortality in Alcohol Drinkers, Ex-Drinkers and Nondrinkers." The American Journal of Cardiology, November, 1990, 1237-42.

19. Kono, Suminori, Koichi Handa, Tomoki Kawano, Tadavuki Hiroki, Yasushi Ishihara, and Kikuo Arakawa, "Alcohol Intake and Nonfatal Acute Myocardial Infarction in Japan." The American Journal of Cardiology, October 1991, 1011-14.

20. Manning, Willard G., Emmett B. Keeler, Joseph P. Newhouse, Elizabeth M. Sloss, and Jeffrey Wassermann. The Costs of Poor Health Habits. Cambridge, Mass.: Harvard University Press, 1991.

21. Model, Karen E. "The Relationship between Drug and Alcohol Use and Income: An Econometric Model." Unpublished manuscript, Harvard University, 1991.

22. Mullahy, John. "Alcohol and the Labor Market." Mimeo, May 1992.

23. ----- and Jody Sindelar, "Lifecycle Effects of Alcoholism on Education, Earnings, and Occupation." Inquiry, Summer 1989, 272-82.

24. Razy, G., K. W. Heaton, C. H. Bolton, and A. O. Hughes, "Alcohol Consumption and Its Relation to Cardiovascular Risk Factors in British Women." British Medical Journal, January 11, 1992, 80-83.

25. Rice, Dorothy P., Sander Kelman, Leonard S. Miller, and Sarah Dunmeyer. The Economic Costs of Alcohol and Drug Abuse and Mental Illness: 1985. San Francisco: Institute for Health and Aging, University of California: 1990.

26. Rimm, Eric B., Edward L. Giovannucci, Walter C. Willett, Graham A. Colditz, Alberto Ascherio, Bernard Rosner and Meir J. Stampfer, "Prospective Study of Alcohol Consumption and Risk of Coronary Disease in Men." The Lancet, August 24, 1991, 464-68.

27. Shaper, A. Gerald, Goya Wannamethee, and Mary Walker, "Alcohol and Mortality in British Men: Explaining the U-Shaped Curve." The Lancet, December 3, 1988, 1267-73.