Employment, wages, and alcohol consumption in Russia.

Tekin, Erdal

1. Introduction

Concerns regarding the relationship between alcohol consumption and labor market productivity are well grounded. If problem drinking or alcoholism is considered a disease, then it may have depressant effects on labor market productivity, causing reductions in employment, earnings, and other labor market outcomes (Mullahy and Sindelar 1993, 1996; Kenkel and Ribar 1994). However, there is a medical literature that documents a U-shaped relationship between alcohol consumption and the risk of cardiovascular disease. According to this literature, moderate drinkers have a lower risk of cardiovascular disease than do either abstainers or heavy drinkers (Beagtehole and Jackson 1992; Doll et al. 1994). Consistent with this evidence, several economists have found a positive association between moderate drinking and earnings (Berger and Leigh 1988; French and Zarkin 1995; Heien 1996; Hamilton and Hamilton 1997; Zarkin et al. 1998; MacDonald and Shields 2001).

This article investigates the impact of alcohol consumption on employment and wages in Russia. The primary contributions of the article are twofold. First, it enhances the economic literature on the relationship between alcohol consumption and labor market behavior by using data from a longitudinal survey. The use of a longitudinal data set enables the estimation of a fixed effects model, which helps avoid the potential bias caused by unobserved individual factors not captured in cross-sectional models. The use of a fixed effects model is an important extension to the literature because the data sets used in previous studies are cross-sectional and usually lack adequate variables that could serve as identifying instruments to control for the endogeneity of alcohol consumption. (1) Furthermore, the richness of the data set used in this article allows for the application of three alternative measures of alcohol consumption in the empirical analysis. This is important because the relationship between alcohol consumption and labor market outcomes may be sensitive to the alcohol measure that is used.

The second notable contribution of this article is that it provides the first empirical evidence on the association between alcohol consumption and labor market outcomes in Russia. Excessive alcohol consumption is a widespread problem in Russia, where per capita alcohol consumption is around 14 liters per year. The World Health Organization (WHO) considers 8 liters to be a sign that a country has a critical consumption level for health problems (Quinn-Judge 1997). Furthermore, the dramatic fluctuations in mortality rates experienced in Russia over the past 15 years have generated considerable attention to the potential effect of alcohol consumption on the overall health of the Russian population. It is important to investigate how labor market outcomes would be affected by alcohol consumption in a country which demonstrates such dramatic fluctuations in alcohol consumption and mortality.

Consistent with the previous studies, cross-sectional results generally support the hypothesis of an inverse U-shaped relationship between alcohol consumption and wages and employment. Results from the fixed-effect models are found to be quite different from those of cross-sectional models. The positive effect of alcohol consumption on employment disappears for both males and females once the individual fixed effects are controlled for. The results of the wage model lend support to a small and linear return to alcohol consumption, but the effects are significant only at the 10% level. The fixed effects estimates are smaller in magnitude and estimated with less precision compared with cross-sectional estimates.

Section 2 reviews the previous literature. Section 3 discusses the conceptual issues and empirical strategy. Section 4 introduces the data set. Section 5 provides a discussion of the results and sensitivity analyses. Section 6 concludes the article.

2. Previous Literature

There is a considerable literature on the impact of alcohol consumption and alcoholism on labor market behavior. However, the vast majority of this literature concentrates on the United States, mainly due to data limitations (e.g., Berger and Leigh 1988; Mullahy and Sindelar 1993, 1996; Kenkel and Ribar 1994; French and Zarkin 1995; Heien 1996; Zarkin et al. 1998; Terza 2002). (2) Only two studies examine alcohol consumption and labor market behavior outside the United States. Hamilton and Hamilton (1997) look at the relationship between alcohol consumption and earnings for males in Canada. MacDonald and Shields (2001) estimate the impact of alcohol consumption on occupational attainment in England.

Previous research typically recognizes the potential endogeneity of alcohol consumption and accounts for it via instrumental variables. The instrumental variables approach usually generates larger estimates than the ordinary least squares (OLS). However, the coefficients are often estimated with little precision, mainly due to weak identification. Previous researchers also recognize that women and men differ systematically in their labor market behavior and alcohol consumption as well as their reaction to ethanol (Roman 1988; Wilsnack and Wilsnack 1992; Caetano 1994; Lex 1994), and therefore, conduct their analyses separately for men and women.

Previous studies vary in both the labor market outcomes analyzed and the alcohol consumption measures used. Labor market outcomes are typically measured as employment, wage rate, unemployment, and earnings. Alcohol consumption measures include a binary yes/no indicator of alcohol consumption, multiple binary indicators of alcohol consumption at different levels, frequency of consumption over a specified period (e.g., the number of drinks consumed, the amount of ethanol consumed, etc.), and intensity of consumption such as clinical measures of alcoholism and problem drinking. With the exception of Kenkel and Ribar (1994), all studies rely on cross-sectional data to estimate the impact of alcohol consumption on labor market outcomes. Limitations caused by cross-sectional data have been usually acknowledged as a shortcoming of the previous research (French and Zarkin 1995; Hamilton and Hamilton 1997; French, Roebuck, and Kebreau 2001).

The researchers have considered several mechanisms through which alcohol consumption may influence labor market outcomes. One of these mechanisms rests on the medical findings of a U-shaped association between alcohol consumption and the risk of cardiovascular disease, which suggests that alcohol consumption at moderate levels may be beneficial for health for relieving stress and reducing the incidence of heart disease. Another mechanism is based on the potentially positive effect of alcohol consumption on the sociability of individuals (Skog 1980; Montgomery 1991; Brodsky and Peele 1999; Putnam 2000). This mechanism suggests that alcohol may play a networking role if consumed during the time spent with colleagues from work by serving as a signal of the individual's commitment to the firm. Also, time spent with colleagues could help the individual derive additional information about the promotion opportunities of the workplace (MacDonald and Shields 2001). Researchers motivated by these mechanisms usually find a positive or inverse U-shaped relationship between drinking and labor market outcomes (Berger and Leigh 1988; French and Zarkin 1995; Heien 1996; Hamilton and Hamilton 1997; Zarkin et al. 1998; MacDonald and Shields 2001).

Researchers using alcoholism or problem drinking as the appropriate measure of alcohol consumption are primarily motivated by the medical literature linking alcoholism and heavy drinking to physical, psychological, and cognitive impairments, which would be detrimental to individuals' labor market productivity (Cruze et al. 1981; Farrell 1985; Fingarette 1988a, b). These researchers usually document deterrent effects of alcohol consumption on employment and earnings (Mullahy and Sindelar 1993, 1996; Kenkel and Ribar 1994). (3)

Motivated by an alarming trend in alcohol consumption and a rising death rate, researchers have concentrated exclusively on the potential link between alcohol consumption and mortality in Russia (Ryan 1995; Leon et al. 1997). (4) However, the evidence concerning the relationship between alcohol consumption and labor market behavior is limited and based primarily on descriptive evidence. For example, Bobak et al. (1999) use odds ratios calculated by unconditional logistic regression to analyze the levels and distribution of alcohol consumption in Russia. They find that factors related to reaction to economic changes and the rating of family economic situation are not strongly related to alcohol consumption. Several other studies document a link between the decline in health status and economic hardship and suggest that alcohol is largely responsible for it (Cornia 1997; Bobak et al. 1998; Walberg et at. 1998).

3. Conceptual Issues and Empirical Strategy

Employment-Alcohol Consumption

The relationship between alcohol consumption and employment can be obtained using a neoclassical framework of utility maximization in which individuals allocate their time and money among consumption of leisure, alcohol, and a composite good. The solution to this optimization would yield labor supply and alcohol demand equations as functions of all prices, wage rate, and all other observable and unobservable factors. Thus, the alcohol demand and labor supply equations can be expressed as follows:

(1) [A.sub.it] = A([P.sub.Ait], [W.sub.it], [X.sub.it], [U.sub.it])

(2) [H.sub.it] = H([P.sub.Ait], [W.sub.it], [X.sub.it], [U.sub.it]),

where [A.sub.it] is the alcohol consumption for individual i at time t, [H.sub.it] is the time spent working, [P.sub.Ait] is the price of alcohol, [W.sub.it] is the wage rate, [X.sub.it], is a vector of all observable factors, including nonwage income, and [U.sub.it] is the unobservable tastes. (5) The individual would opt for the corner solution with respect to labor force participation if her/his reservation wage exceeds [W.sub.it]. Thus, labor force participation decision is determined by the same factors that influence the labor supply decision and can be expressed as the following function:

(3) [E.sub.it] = E([P.sub.Ait], [W.sub.it], [X.sub.it], [U.sub.it]),

where [E.sub.it] is a binary indicator of employment.

To relate alcohol consumption to employment, one can rewrite Equation 3 as an employment function conditional on alcohol demand (Mullahy and Sindelar 1996; French, Roebuck, and Kebreau 2001; DeSimone 2002). (6) Thus, the employment function can be denoted as

(4) [E.sub.it] = E([A.sub.it], [X.sub.it], [U.sub.it]).

Similarly. Equation 1 can be rewritten as an alcohol demand equation conditional on employment

(5) [A.sub.it] = A ([E.sub.it], [X.sub.it], [u.sub.it]).

Assuming linearity, the econometric counterpart to Equation 4 can be expressed as the following:

(6) [E.sub.it] = [X.sub.it][[beta].sub.1] + [A.sub.it][[beta].sub.2] + [U.sub.it],

where [beta]'s are the parameters to be estimated.

Wage Rate-Alcohol Consumption

Following Mullahy and Sindelar (1993) and French and Zarkin (1995), alcohol consumption is incorporated into the wage equation using a human capital framework. Specifically, the wage equation is formulated as

(7) [W.sub.it] = W([S.sub.it], [N.sub.it], [X.sub.it], [v.sub.it]),

where [S.sub.it], is a vector of measures of the health components of human capital, [N.sub.it] is a vector of nonhealth components of human capital such as schooling and experience, and [v.sub.it] is a vector of all unobservable determinants of wage rate.

Following Mullahy and Sindelar (1993), one can specify [S.sub.it] = ([A.sub.it], [K.sub.it]), where [K.sub.it] is a vector of other health outcomes. Substituting [S.sub.it] into Equation 7 and specifying a linear function as an econometric counterpart, Equation 7 becomes

(8) [W.sub.it] = [Q.sub.it][[alpha].sub.1] + [A.sub.it][[alpha].sub.it] + [v.sub.it]

where [Q.sub.it] is a summary of all other covariates ([X.sub.it], [K.sub.it], [N.sub.it]); and [alpha]'s are the parameters to be estimated.

It is not straightforward to establish a casual relationship between alcohol consumption and labor market outcomes for two reasons. First, alcohol consumption and labor market outcomes may be simultaneously determined. For example, if alcohol consumption is a normal good, then it follows from Equations 1, 3, and 5 that both employment and the wage rate will increase alcohol demand by increasing earnings. This would cause upward bias in the estimated effects of alcohol consumption on employment and the wage rate. Second, if there are unobserved individual factors that are correlated with both drinking and labor market outcomes, then coefficients [[beta].sub.2] and [[alpha].sub.2] will be biased. For example, individuals with a high rate of time preference may be more likely to make their consumption decisions based on the current satisfaction they derive without considering the future consequences (Becker and Murphy 1988; DeSimone 2002). Such individuals may also be more likely to select jobs with flatter age/earnings profiles. In a cross-sectional framework, a possible remedy for these problems would be to use an instrumental variables (IV) method. The success of the IV method. however, depends largely on the predictive power of the instruments used in the first-stage equations. The IV method is further complicated with the difficulty of finding instruments that can appropriately be excluded from the second-stage equations.

The use of longitudinal data provides an alternate solution to the problem of unobserved individual heterogeneity. This is implemented by including individual fixed effects in the empirical model, which is equivalent to estimating the model as deviations from the means. The fixed effects model can then be specified as

(9) [[bar.E].sub.it] = [[bar.X].sub.it][[beta].sub.1] + [[bar.A].sub.it][[beta].sub.2] + [[member of].sub.1it],

(10) [[bar.W].sub.it] = [[bar.Q].sub.it][[alpha].sub.1] + [bar.A].sub.it][[alpha].sub.2] + [[member of].sub.2it],

where the variables with bars represent deviations from their within-individual means and [epsilon]'s denote the residual disturbances that are assumed to be uncorrelated with the explanatory variables.

Fixed effects analysis, while having important advantages, is not without certain drawbacks. First, the effects of time-invariant variables cannot be estimated. Second, the fixed effects model does not eliminate bias in case of time-variant individual heterogeneity. In this case, the IV method employed together with the fixed effects is a possible remedy. The practical difficulty, however, is to find time-varying instruments that predict alcohol consumption but are uncorrelated with wages and employment. To address this issue, I use alcohol prices at the community level as identifying instruments. (7) Finally, the use of fixed effects may exacerbate the bias caused by measurement error (Griliches and Hausman 1986; Levitt 1998).

4. Data

The data used in the empirical analyses are drawn from the Russia Longitudinal Monitoring Survey (RLMS). (8) The RLMS is the first nationally representative household-based survey conducted in Russia. It comprises two phases, each conducted on a different sample. The empirical analyses in this article use data from Phase II of the survey, which includes rounds five to nine, conducted annually between November 1994 to December 2000.(9) The RLMS is an ideal data source for the purpose of the present study because its longitudinal nature allows the use of a fixed effects model. It also contains detailed information on labor market behavior and alcohol consumption of the respondents. (10)

The outcomes examined in this study are employment and the wage rate. Employment is measured by assessing whether the individual is employed at the time of the interview. The wage rate is defined as the ratio between the total monthly earnings and total number of hours of work in the last 30 days on the main job. Following Glinskaya and Mroz (2001), total monthly earnings are defined as the sum of salaries, wages, bonuses, grants, benefits, and profits plus the monetary value of the in-kind payments actually received in the last 30 days from the main place of employment. In order to adjust for the effect of inflation, the monthly Consumer Price Index (CPI), calculated by the Russian Statistical Bureau (Goskomstat) and published in Russian Economic Trends, is used (1998 base).

The survey asks respondents whether they consumed any alcohol during the past 30 days. Those who reported in the affirmative are then asked about the number of times they consumed alcohol during that period. Two sets of alcohol-consumption measures are constructed using these responses. The first measure is a single binary variable, indicating whether the respondent consumed any alcohol in the past 30 days. The second measure is a set of binary indicators that includes six different frequency responses. This constitutes the primary measure of alcohol consumption in the article and is designed to capture any nonlinear association between alcohol consumption and the outcomes analyzed. In addition to the discrete measures, a continuous measure of alcohol consumption is defined as a third measure. The RLMS also asks respondents about their alcohol consumption in grams of beer, vodka, fortified wine, table wine, and homemade liquor in grams. Using this information, a measure of composite ethanol consumption is calculated as a weighted average of the ethanol typically found in each of these beverages. The algorithm used to construct the measure of ethanol consumption assumes that total amount of ethanol is 5% in beer, 40% in vodka and homemade liquor, 20% in fortified wine, and 12% in table wine. (11)

Although the alcohol-related answers are self-reported in the RLMS, several studies have examined the validity of self-reported alcohol consumption and have found a fairly high correlation between the validity of self-reported alcohol and drug use data and alternative sources of information (Midanik 1982; Rouse, Kozel, and Richards 1985; Preston et al. 1997).

The demographic variables used in the analysis include linear and quadratic terms of age, years of education, household size; binary indicators of marital status, health, region of residence, urban residence, and occupation. (12) After eliminating the categories representing the agricultural sector, nine binary occupational indicators are constructed. (13) Individuals with missing information on the key variables are excluded from the sample. Because the analysis focuses on within-individual changes in alcohol consumption and labor market outcomes, individuals who are in the RLMS for only one round are also excluded. This leaves a total of 5565 individuals, providing 19,292 observations. One thousand nine hundred twelve of these individuals were interviewed in all five rounds, 646 in four rounds, 1134 in three rounds, and the remaining 1873 were interviewed in two rounds. Definitions of the variables used in the analysis are presented in Appendix A.

Table 1 presents the descriptive statistics for the full sample and for males and females separately. A comparison of the six binary indicators of alcohol consumption between males and females indicates that drinking intensity is substantially higher for males. For example, those who drink at least once a week make up 34% of the male sample and only 12% of the female sample. These figures suggest that the majority of females are infrequent drinkers. A similar pattern is observed when one looks at the continuous measure of alcohol consumption. The mean value of weekly ethanol consumption for males is more than twice that for females, pointing to their higher overall consumption of alcohol. Therefore, the analyses are conducted separately for males and females (Roman 1988: Mullahy and Sindelar 1991; Wilsnack and Wilsnack 1992; Caetano 1994; Lex 1994).

As displayed in Table 1, there are some differences between males and females in characteristics other than alcohol consumption. About 77% of males and 75% of females in the sample are employed. Males earn more than females. Females are slightly more educated than males, while males are slightly healthier than females. Only 3% of males and 2% of females are in excellent health. Females are more likely to be employed as professionals, technicians, clerks, and sales workers, and elementary (unskilled) workers, while males usually occupy managerial, craft-related, plant, and machine-operator jobs. Males live in larger households than females. This makes sense given that they are also more likely to be married than are females in the sample.

Table 2 presents the mean values of employment and wages by drinking status and gender subgroups. For both males and females, the wage rate is lower for nondrinkers than drinkers, and this differential seems to persist even at higher drinking levels. Similarly, the employment rate is higher for drinkers than nondrinkers, but the difference gets smaller as the level of drinking intensity increases. For example, those males and females who drink every day have a lower employment rate than those who do not drink.

5. Empirical Results

The results of the empirical analyses are presented in two stages. In the first stage, the results from the cross-sectional analyses are presented both to serve as a link to the previous studies and as a base for evaluating the relative contribution of controlling for unobserved heterogeneity. In the second stage, the fixed effects results are discussed. All regressions include year fixed effects. These control for unobserved time-variant determinants of labor market outcomes, which would affect all individuals in the same manner. Each model is estimated with each of the three alcohol-consumption measures.

Baseline Estimates

Tables 3 and 4 present the cross-sectional ordinary least squares results for the employment model for males and females, respectively. I specify a linear probability model for the employment model for ease of estimation and interpretation. (14) Because the observations are clustered within primary sampling units (PSU) in the RLMS, the standard errors are corrected to account for the effects of intracluster correlation caused by clustered data in the cross-sectional models.

Employment Models

The coefficient estimates for the demographic and human capital variables behave as one would expect and remain relatively stable across alternative measures of alcohol consumption for both males and females. Increased levels of healthiness raise the likelihood that an individual will be employed. Education is estimated to have a small positive impact. The age effect has the expected quadratic shape. Nonwage income has a statistically significant negative coefficient, indicating that leisure is a normal good. The full set of regression results is not displayed in a table for the interest of space. However, they are available on request from the author.

The primary interest is the impact of alcohol consumption on employment. Looking at the coefficients of the six binary indicators of alcohol consumption in Tables 3 and 4 reveals an inverse U-shaped relationship between drinking and employment for both males and females. However, a specification test failed to reject the hypothesis that the six coefficients are equal to each other for males. This implies that the relationship between alcohol consumption and employment may be described by a simple shift in mean employment, independent of the level of alcohol consumption (column 2). For females, the pattern of coefficient estimates in column 1 resembles an inverse U-shaped relationship. Moreover, the hypotheses of joint significance and equality are strongly rejected. As reported in the third column of Tables 3 and 4, the existence of an inverse U-shaped relationship is further evidenced by the coefficients of the continuous measure of ethanol consumption. (15) In general, the employment propensities for females are more robust and larger in magnitude than those for males in all three specifications.

Wage Models

Tables 5 and 6 present the cross-sectional OLS results for the wage models for males and females, respectively. The dependent variable is the natural logarithm of the hourly wage rate. The coefficient estimates for the demographic and human capital variables have the expected signs and are similar in magnitude across the three specifications. (16) The coefficients on variables other than alcohol consumption are not displayed in a table to economize on space. Instead, they are available on request from the author.

Coefficient estimates reported in the first and second columns of Table 5 provide no evidence of an inverse U-shaped relationship between wages and drinking for males. The coefficients in the first column do not resemble any particular pattern, and only one coefficient is statistically significant in the first two columns. By way of contrast, the coefficients of the linear and quadratic terms of the ethanol consumption lend support to an inverse U-shaped relationship.

Turning to females in Table 6, the estimates of the alcohol consumption variables are much larger than those for males and are estimated much more precisely. The point estimates of the six binary indicators of drinking in column 1 clearly resemble an inverse U-shaped relationship between alcohol consumption and wages. Furthermore, a specification test strongly rejected the hypothesis that the coefficients of the six indicators in column 1 are equal to each other. The nonlinear association between alcohol consumption and wages is further supported by the coefficient estimates of ethanol consumption in the third column.

Fixed Effects

To account for the possible endogeneity of alcohol consumption, the models are estimated with fixed effects. (17) Because the majority of the explanatory variables exhibit little or no variation over the sample period, these coefficients are either dropped from the models or are estimated with little precision. Therefore, the results for these variables are not discussed in the text.

Employment Models

Results from the fixed effects employment model are displayed in Table 7. These results differ sharply from the cross-sectional ones and reveal a great deal concerning the biases in the cross-sectional estimates. In all three specifications, the fixed effect estimates of alcohol consumption are smaller in magnitude and are estimated with much less precision than the cross-sectional estimates. For all practical purposes, the estimates from all three specifications can be considered zero. In fact, a specification test indicated that the estimated effects of the six binary indicators of drinking are jointly zero. This finding suggests that individual unobserved heterogeneity is positively correlated with alcohol consumption and, once it is controlled for, the positive impact of alcohol consumption on employment disappears.

Table 8 presents the fixed effects estimates for the employment model for females. In all three models, the estimates are much smaller than those of cross-sectional ones. The coefficients are not statistically significant and a specification test failed to reject the hypothesis that they are jointly zero. The finding of no significant effect is also supported by the second and third columns, which report the results of the specifications with the binary and continuous measures of alcohol consumption. These results suggest that the estimates without fixed effects capture not only the impact of drinking but also the positive impact on alcohol consumption of the unobserved individual factors. After accounting for unobserved heterogeneity, drinking males and females are found to be no more likely to be employed than are nondrinking males.

Wage Models

The results from the fixed effects wage models for males and females are displayed in Tables 9 and 10, respectively. Similar to the fixed effects employment results, the estimated effects of alcohol consumption on the wage rate become smaller when individual fixed effects are controlled for. The estimates in column 1 of Table 9 reveal no indication of an inverse U-shaped relationship and an F-test failed to reject the hypothesis that these coefficients are equal to each other. According to the second column, alcohol consumption is associated with a 7% increase in male wages. Referring to column 3, the linear term for the continuous measure of alcohol consumption is significant and positive while the quadratic term is insignificant, lending support to the positive but linear impact of drinking on wage rate found in the second column.

The point estimates for female wages displayed in Table 10 resemble an inverse U-shaped relationship between alcohol consumption and wages. Only two of the six coefficients are significant. However, specification tests rejected the hypothesis that they are jointly zero, but failed to reject that they are equal to each other. When a linear association is imposed, as in column 2, female drinkers are estimated to earn about 10% more than female nondrinkers. This positive association is also supported by the third column. As with males, the fixed effects results are smaller in magnitude than those in the cross-sectional results.

Problem-Drinking

The results discussed so far do not reveal any information about the potential relationship between unhealthy levels of drinking and labor market outcomes. It would be useful to provide some insights into that link to connect the present study to both the literatures of alcohol consumption-labor market outcomes and problem drinking-labor market outcomes. The continuous measure of alcohol consumption used in this article allows for such an implementation. Specifically, a binary indicator of problem drinking is formulated using the 90th percentile in the distribution of the ethanol consumption in the sample (Mullahy and Sindelar 1996; Terza 2002). The 90th percentile for males and females are 376 and 143 grams of ethanol per week, respectively. This measure selves as an indicator of unhealthy or problem drinking. All the models are estimated with the binary indicator of problem drinking. The cross-sectional results suggest that problem drinking is associated with both lower employment and wages for males as well as females. Specifically, problem drinking is associated with decreases of 3.6% and 4.7% in employment for males and females, respectively. For the wage models, problem drinking tends to be associated with a decreases in wage rate by 2.9% and 3.7% for males and females, respectively.

In the fixed-effect models, the effects of problem drinking on employment and the wage rate are much smaller in magnitude compared with its effects in the cross-sectional models for males and females. Furthermore, the standard errors are large and none of the estimates reach statistical significance at conventional levels. Measurement error is likely to be more severe for the ethanol consumption measure than for the binary indicators. Also, it is well known that the attenuation bias caused by measurement error is exacerbated in the case of fixed effects. Therefore, these results should be viewed with some caution.

Discussion

The fixed effects wage results suggest a wage premium of 7% for a male drinker and 10% for a female drinker. Although it is difficult to make a perfect comparison with the previous literature because of the different measures of alcohol consumption and data sets used in different studies, it would still provide some useful insights into the reliability of the magnitude of the wage effects in this study. MacDonald and Shields (2001) find a 13.7% gain for a male who drinks 21 units of alcohol per week (the mean alcohol consumption among drinkers) compared with another male with the same characteristics who does not drink. Berger and Leigh (1988) document a wage premium of 36% for male drinkers and 59% for female drinkers. Zarkin et al. (1998) find a wage premium on the order of 50--200% when they used an IV estimation. Their OLS estimation suggests that alcohol consumption is associated with a 7% and 4% increase in wages for males and females, respectively. Heien (1996) reports a wage premium of approximately 50%. These figures are indicative of a lack of consensus on the magnitude of the effect of alcohol consumption on wages in the literature. They also do not suggest any evidence against the plausibility of the estimates found in this study.

The results of the fixed effects models differ considerably from those of cross-sectional estimates. For the employment models, the inverse-U relationship observed in the cross-sectional results disappears for both males and females once the unobserved heterogeneity is controlled for. This suggests that the positive association found in the cross-sectional estimation is ultimately driven by unobserved individual heterogeneity and that alcohol consumption has no significant impact on employment. For the wage models, the fixed-effect results suggest a positive and linear effect of alcohol consumption on wages. This finding is obtained in specifications with both the discrete and continuous measures of alcohol consumption. However, care must be taken when making inferences based on the wage effects because the coefficients are significant only at the 10% level.

The qualitative findings between males and females are similar, but quantitative differences are present. The estimated effects of alcohol consumption are larger in magnitude for females than males in all three specifications. This finding is in contrast with those of French and Zarkin (1995) and MacDonald and Shields (2001), who document lower effects for females than males. Although it is common to find differential impacts of alcohol consumption on labor market outcomes for men and women in the literature, it is surprising that little explanation is provided for these differences. The unmeasured variation in the intensity of drinking behavior may be responsible for much of the noted difference in the estimated effects between men and women. It is clear from the descriptive statistics that women in the sample drink less frequently than men. It is also likely that, on those days men drink, they consume more alcohol or drink more intensely than females (which explains why the proportion of men considered binge drinkers is much higher than of women in Russia [Bobak et al. 1999; Malyutina et al. 2002]). Therefore, drinking once a week may not have the same meaning for men and women and consequently, may have different effects on labor market outcomes.

Several researchers have stressed that controlling for covariates that might he correlated with alcohol consumption (e.g., health status, education, and marital status) could have a large impact on the estimated coefficients of alcohol consumption on labor market outcomes (Mullahy and Sindelar 1993; French and Zarkin 1995; MacDonald and Shields 2001). If this is true, then it implies that alcohol consumption influences labor market outcomes both directly and indirectly through human capital and family formations. To investigate this issue, all cross-sectional models are reestimated with education, health, and marital status variables (and experience and occupation dummies in the wage models) excluded from the analyses. (18) This exercise produced little change in the estimates of the alcohol consumption coefficients, suggesting no evidence for a possible indirect association. A similar finding was obtained in Hamilton and Hamilton (1997).

The concern of a possible simultaneity between alcohol consumption and labor market outcomes has been raised in several studies (Kenkel and Ribar 1994; MacDonald and Shields 2001). In order to deal with potential simultaneity and time-variant heterogeneity, the IV method is used together with the fixed effects. Variation in the prices of various alcoholic beverages in the 160 geographic units provided in the data set is used to identify the impact of alcohol consumption. (19) The point estimates on alcohol consumption variables from two-stage least squares (2SLS) are much larger than the estimates from both the cross-sectional and fixed effects models. However, none of the estimates of alcohol consumption are statistically significant by conventional standards. Furthermore, the identifying instruments perform poorly in the first stage, casting doubt on the validity of the 2SLS results. On the grounds of weak identification, little value is placed on the estimates from 2SLS, and so they are not presented here. The same problem has been faced by several other researchers, who then estimated their models using OLS with wages as the only endogenous variable (e.g., Mullahy and Sindelar 1993; French and Zarkin 1995; Zarkin et al. 1998).

Another issue of concern is the potential endogeneity of the pregnancy decision for females. The plausibility of the results for females hinges on the assumption that pregnancy is not decided based on decisions endogenous to drinking and employment. (20) Otherwise, the estimates would be biased. To address this issue, I repeated the analyses for females, restricting the sample to those aged 40 or older, the rationale being that women who are 40 or older are likely to have already completed their pregnancy decision. Subsequently, the results using this subsample are unlikely to be subject to any endogeneity bias. This exercise did not change the implications of the findings in any significant manner.

The practice of late or nonpayment of wages is a common problem in Russia (Earle and Sabirianova 2002). The proportion of the sample facing wage arrears is about 29%, accounting for the small sample size in the wage regressions. Fortunately, the RLMS provides information on the number of months and the amount of money owed to an individual. Following Glinskaya and Mroz (2001), this information is used to construct a contractual wage rate and the analyses are repeated using this contractual wage. (21) Cross-sectional estimates of the impact of alcohol consumption on contractual wage rate appear to follow a pattern similar to those reported in Tables 5 and 6. Fixed effect regressions of contractual wage rate continue to provide estimates that are smaller in magnitude than the cross-sectional estimates. The standard errors again become larger, and although still positive, the coefficient of the single binary indicator for males (column 2) is no longer significant.

As is the case in any panel data, the RLMS is vulnerable to complications created by sample attrition. If the sample attrition--whether due to moving out of the household, nonresponse, or some other factor--were nonrandom, the estimated impact of alcohol consumption on labor market outcomes would be biased. To better assess the extent and impact of sample attrition, I looked at the characteristics of the individuals for the balanced (individuals present in all five rounds) and individual rounds. A comparison revealed that the characteristics of two samples are quite similar. Therefore, although I cannot rule out the possibility of nonrandom attrition, the estimates are unlikely to be affected significantly. In a related manner, if certain types of drinkers, for example, heavy drinkers, are more likely to have their employment or wage rate missing in the data, the sample will be biased because these observations are excluded from the analysis. To investigate this issue, I compared the percentage of observations with missing employment and wage rate data among the subsamples of drinkers, nondrinkers, and heavy drinkers (the highest two drinking categories). The percentages were similar in all three cases, suggesting that the sample is not biased toward nondrinkers or moderate drinkers.

It is important to consider whether the relationship between drinking and labor market remained stable over the sample period. To test this, I added interactions of time dummies with the binary and continuous alcohol measure into the models. This exercise did not change the implications of the results in any significant way.

As suggested previously, the fixed effects results could partly be driven by measurement error in the alcohol consumption variables, which would bias coefficient estimates toward zero. As a result, the measurement or reporting error in alcohol consumption could contribute to the smaller coefficient estimates obtained in fixed effects models. Also, the attenuation bias caused by measured error is exacerbated in the case of fixed effects. Therefore, rather than attributing the entire reduction in the estimated effects to unobserved individual heterogeneity, the fixed effects estimates must be considered as a lower bound of the impact of alcohol consumption on employment and wages.

6. Conclusions

The primary objective of this article is to provide evidence concerning the potential link between alcohol consumption and labor market outcomes in Russia and to use a fixed effects model to uncover this relationship. Using data from the Russia Longitudinal Monitoring Survey, this article examines the effects of alcohol consumption on employment and wages. Because the relationship between alcohol consumption and labor market behavior may be sensitive to the way alcohol consumption is measured, the models are estimated using three different sets of alcohol-consumption measures. Furthermore, models are estimated with and without fixed effects in order to identify the extent of bias arising from failing to control for unobserved heterogeneity.

The results highlight the importance of controlling for unobserved heterogeneity when estimating the relationship between alcohol consumption and labor market outcomes. The cross-sectional findings support the view that moderate alcohol consumption is positively associated with employment and wages for females. The relationship appears to follow an inverse U-shape in both employment and wage models for females in all three models. For males, the continuous measure supports the inverse U-shaped relationship, but the evidence is less clear when the binary indicators are used. Overall, these findings are generally consistent with the studies for the United States, Canada, and Great Britain. Results from the fixed-effect models are found to be quite different from those of cross-sectional models. The positive effect of alcohol consumption on employment disappears for both males and females once the individual fixed effects are controlled for. The results of the wage model lend support to a small and linear return to alcohol consumption, but the effects are significant only at the 10% level. The fixed effects estimates are smaller in magnitude and estimated with less precision compared with cross-sectional estimates.

The analyses are repeated using a binary indicator of unhealthy or problem drinking, which is calculated from the distribution of the ethanol consumption. Based on this indicator, an individual drinking at more than the 90th percentile is considered to be a problem drinker. The results from the cross-sectional analyses suggest that problem drinking has adverse effects for employment and wages for both males and females. However, these adverse effects disappear once the fixed effects are controlled for.

While this article extends the literature on the relationship between alcohol consumption and labor market outcomes by using a longitudinal data set from Russia, more research is clearly needed to see whether these results would be supported by longitudinal data from other countries. Also, an examination of the link between binge drinking and labor market outcomes would further enhance our understanding of the subject.

Appendix A
Variable Definitions

Wage Hourly wage rate in rubles
Work Dummy variable (=1) if employed, 0 otherwise
Alcohol Dummy variable (=1) if used alcoholic
 beverages in the last 30 days, 0 otherwise
Ethanol Ethanol consumed (in grams) in the last week
Drinks every day Dummy variable (=1 if used alcoholic
 beverages every day in the last 30 days,
 0 otherwise
Drinks 4-6 times a week Dummy variable (=1) if used alcoholic
 beverages 4-6 times a week in the last 30
 days, 0 otherwise
Drinks 2-3 times a week Dummy variable (=1) if used alcoholic
 beverages 2-3 times a week in the last 30
 days, 0 otherwise
Drinks once a week Dummy variable (=1) if used alcoholic
 beverages once a week in the last 30 days,
 0 otherwise
Drinks 2-3 times a month Dummy variable (=1) if used alcoholic
 beverages 2-3 times in the last 30 days,
 0 otherwise
Drinks once a month Dummy variable (=1) if used alcoholic
 beverages once in the last 30 days, 0
 otherwise
Age Age in years
Education Years of completed schooling
Never married Dummy variable (=I) if never married, 0
 otherwise
Married Dummy variable (=1) if married, 0 otherwise
Divorced (a) Dummy variable (=1) 1 if divorced or
 widowed, 0 otherwise
Nonwage income Total monthly income from all sources net of
 labor income
Excellent health Dummy variable (=1) if health status
 reported very good, 0 otherwise
Good health Dummy variable (=1) if health status
 reported good, 0 otherwise
Average health Dummy variable (=1) if health status
 reported average, 0 otherwise
Household size Number of household members
Bad health (b) Dummy variable (=1) if health status
 reported bad, 0 otherwise
Urban Dummy variable (=1) if resides in an urban
 settlement, 0 otherwise
Occupation 1 Dummy variable (=1) if senior official or
 manager, 0 otherwise
Occupation 2 Dummy variable (=1) if professional, 0
 otherwise
Occupation 3 Dummy variable (=1) if technician or
 associate professional, 0 otherwise
Occupation 4 Dummy variable (=1) if clerk, 0 otherwise
Occupation 5 Dummy variable (=1) if service or sales
 worker, 0 otherwise
Occupation 6 Dummy variable (=1) if craft and related
 trades, 0 otherwise
Occupation 7 Dummy variable (=1) if plant and machine
 operators, 0 otherwise
Occupation 8 (a) Dummy variable (=1) if elementary
 occupations, 0 otherwise
Occupation 9 Dummy variable (=1) if army, 0 otherwise
Region 1 Dummy variable (=1) if resides in Moscow and
 St. Petersburg, 0 otherwise
Region 2 Dummy variable (=1) if resides in north and
 northwest, 0 otherwise
Region 3 Dummy variable (=1) if resides in Central
 and Central Black-Earth, 0 otherwise
Region 4 Dummy variable (=1) if resides in Volga-
 Vaytski and Volga Basin, 0 otherwise
Region 5 Dummy variable (=1) if resides in North
 Caucasian, 0 otherwise
Region 6 Dummy variable (=1) if resides in Ural, 0
 otherwise
Region 7 Dummy variable (=1) if resides in Western
 Siberia, 0 otherwise
Region 8 (a) Dummy variable (=1) if resides in Eastern
 Siberia and Far East, 0 otherwise

(a) Omitted category.

Table 1. Descriptive Statistics

Variable Full Sample

Work 0.759 (0.428)
Wage 8.651 (12.746)
Alcohol 0.655 (0.475)
Drinks every day 0.011 (0.106)
Drinks 4-6 times/week 0.017 (0.131)
Drinks 2-3 times/week 0.081 (0.273)
Drinks once/week 0.143 (0.350)
Drinks 2-3 times/month 0.235 (0.424)
Drinks once/month 0.167 (0.373)
Ethanol/week 138.703 (127.601)
Age 37.786 (9.239)
Education 9.509 (1.203)
Never married 0.076 (0.238)
Married 0.781 (0.402)
Divorced 0.143 (0.216)
Nonwage income 259.539 (702.1)
Excellent health 0.021 (0.144)
Good health 0.299 (0.458)
Average health 0.589 (0.492)
Bad health 0.091 (0.288)
Household size 3.346 (1.432)
Urban 0.758 (0.436)
Occupation 1 0.032 (0.176)
Occupation 2 0.168 (0.374)
Occupation 3 0.155 (0.362)
Occupation 4 0.063 (0.243)
Occupation 5 0.083 (0.275)
Occupation 6 0.170 (0.375)
Occupation 7 0.203 (0.403)
Occupation 8 0.115 (0.319)
Occupation 9 0.012 (0.109)
Region 1 0.074 (0.262)
Region 2 0.077 (0.266)
Region 3 0.187 (0.390)
Region 4 0.181 (0.385)
Region 5 0.135 (0.341)
Region 6 0.156 (0.363)
Region 7 0.101 (0.301)
Region 8 0.090 (0.286)
Round 5 18.0%
Round 6 21.0%
Round 7 20.7%
Round 8 21.5%
Round 9 18.8%

Number of observations 19,292

Variable Males

Work 0.771*** (0.420)
Wage 9.771*** (13.088)
Alcohol 0.744*** (0.436)
Drinks every day 0.021*** (0.142)
Drinks 4-6 times/week 0.031*** (0.174)
Drinks 2-3 times/week 0.137*** (0.344)
Drinks once/week 0.198*** (0.398)
Drinks 2-3 times/month 0.246*** (0.431)
Drinks once/month 0.112*** (0.315)
Ethanol/week 192.449*** (145.130)
Age 38.017*** (9.898)
Education 9.405** (1.307)
Never married 0.085*** (0.251)
Married 0.803*** (0.439)
Divorced 0.112*** (0.208)
Nonwage income 301.690*** (741.2)
Excellent health 0.031*** (0.174)
Good health 0.373*** (0.484)
Average health 0.522*** (0.500)
Bad health 0.074*** (0.262)
Household size 3.482** (1.519)
Urban 0.745* (0.413)
Occupation 1 0.039*** (0.193)
Occupation 2 0.107*** (0.309)
Occupation 3 0.067*** (0.250)
Occupation 4 0.011*** (0.106)
Occupation 5 0.050*** (0.219)
Occupation 6 0.279*** (0.448)
Occupation 7 0.333*** (0.471)
Occupation 8 0.093*** (0.290)
Occupation 9 0.021*** (0.143)
Region 1 0.071 (0.256)
Region 2 0.077 (0.266)
Region 3 0.186 (0.389)
Region 4 0.186* (0.388)
Region 5 0.139* (0.346)
Region 6 0.153 (0.360)
Region 7 0.100 (0.300)
Region 8 0.089 (0.284)
Round 5 18.2%
Round 6 21.5%
Round 7 20.9%
Round 8 21.1%
Round 9 18.3%

Number of observations 9525

Variable Females

Work 0.743 (0.434)
Wage 7.557 (11.869)
Alcohol 0.568 (0.495)
Drinks every day 0.002 (0.048)
Drinks 4-6 times/week 0.004 (0.063)
Drinks 2-3 times/week 0.027 (0.161)
Drinks once/week 0.089 (0.285)
Drinks 2-3 times/month 0.225 (0.418)
Drinks once/month 0.221 (0.415)
Ethanol/week 86.289 (77.537)
Age 37.561 (8.542)
Education 9.610 (1.083)
Never married 0.067 (0.213)
Married 0.760 (0.411)
Divorced 0.173 (0.234)
Nonwage income 218.432 (622.5)
Excellent health 0.011 (0.106)
Good health 0.227 (0.419)
Average health 0.654 (0.476)
Bad health 0.108 (0.310)
Household size 3.273 (1.334)
Urban 0.771 (0432)
Occupation 1 0.025 (0.156)
Occupation 2 0.229 (0420)
Occupation 3 0.244 (0.429)
Occupation 4 0.115 (0.319)
Occupation 5 0.115 (0.318)
Occupation 6 0.060 (0.237)
Occupation 7 0.073 (0.261)
Occupation 8 0.137 (0.343)
Occupation 9 0.003 (0.056)
Region 1 0.078 (0.268)
Region 2 0.077 (0.267)
Region 3 0.187 (0.390)
Region 4 0.177 (0.381)
Region 5 0.130 (0.336)
Region 6 0.159 (0.365)
Region 7 0.102 (0.303)
Region 8 0.091 (0.287)
Round 5 17.7%
Round 6 20.6%
Round 7 20.6%
Round 8 21.8%
Round 9 19.3%

Number of observations 9767

Standard deviations are in parentheses. The definitions of the
variables are presented in Appendix A.

* Statistically different from female mean at p < 0.10.

** Statistically different from female mean at p < 0.05.

*** Statistically different from female mean at p < 0.01.

Table 2. Distribution of Employment and Wages

 Full Sample Males

 Work Wage Work Wage

Drink in the last 30 days

Yes 0.787 9.472 0.788 10.177
 (0.439) (13.719) (0.406) (13.604)

No 0.706 7.089 0.720 8.590
 (0.416) (10.761) (0.449) (10.123)

Frequency of drinking

Every day 0.603 11.752 0.628 11.997
 (0.383) (14.129) (0.415) (14.451)

4-6 times a week 0.724 12.882 0.758 13.202
 (0.403) (17.403) (0.422) (18.141)

2-3 times a week 0.758 10.321 0.771 10.555
 (0.413) (13.336) (0.411) (15.432)

Once a week 0.804 10.072 0.811 10.412
 (0.457) (13.103) (0.438) (13.390)

2-3 times a month 0.798 9.315 0.797 10.085
 (0.422) (16.228) (0.421) (13.138)

Once a month 0.790 8.257 0.787 8.316
 (0.444) (12.230) (0.443) (7.898)

 Females

 Work Wage

Drink in the last 30 days

Yes 0.785 8.571
 (0.410) (12.796)

No 0.698 6.222
 (0.381) (7.100)

Frequency of drinking

Every day 0.391 9.664
 (0.375) (13.886)

4-6 times a week 0.462 10.437
 (0.488) (14.134)

2-3 times a week 0.691 9.139
 (0.383) (13.121)

Once a week 0.788 9.336
 (0.452) (12.689)

2-3 times a month 0.799 8.495
 (0.455) (11.332)

Once a month 0.792 8.228
 (0.451) (8.301)

Standard errors are in parentheses.

Table 3. Cross-Sectional Estimates of Employment Model for Males

Variable Model One Model Two

Alcohol -- 0.050 *** (0.013)
Drinks every day -0.053 (0.035) --
Drinks 4-6 times/week 0.032 (0.026) --
Drinks 2-3 times/week 0.040 *** (0.014) --
Drinks once/week 0.062 *** (0.013) --
Drinks 2-3 times/month 0.057 *** (0.012) --
Drinks once/month 0.050 *** (0.015) --
Ln(ethanol) -- --
Ln[(ethanol).sup.2] -- --
[R.sup.2] 0.058 0.055
Number of observations 9525 9525

Variable Model Three

Alcohol --
Drinks every day --
Drinks 4-6 times/week --
Drinks 2-3 times/week --
Drinks once/week --
Drinks 2-3 times/month --
Drinks once/month --
Ln(ethanol) 0.069 *** (0.011)
Ln[(ethanol).sup.2] -0.011 *** (0.002)
[R.sup.2] 0.059
Number of observations 9525

Robust standard errors are in parentheses.

*, **, and *** Estimated coefficients are statistically different from
zero at the 10%, 5%, and 1% levels respectively.

Table 4. Cross-Sectional Estimates of Employment Model for Females

Variable Model One Model Two

Alcohol -- 0.072 *** (0.009)
Drinks every day -0.314 *** (0.100) --
Drinks 4-6 times/week -0.267 *** (0.078) --
Drinks 2-3 times/week -0.002 (0.30) --
Drinks once/week 0.079 *** (0.015) --
Drinks 2-3 times/month 0.086 *** (0.011) --
Drinks once/month 0.073 *** (0.011) --
Ln(ethanol) -- --
Ln[(ethanol).sup.2] -- --
[R.sup.2] 0.078 0.071
Number of observations 9767 9767

Variable Model Three

Alcohol --
Drinks every day --
Drinks 4-6 times/week --
Drinks 2-3 times/week --
Drinks once/week --
Drinks 2-3 times/month --
Drinks once/month --
Ln(ethanol) 0.089 *** (0.014)
Ln[(ethanol).sup.2] -0.022 *** (0.004)
[R.sup.2] 0.079
Number of observations 9767

Robust standard errors are in parentheses.

*, **, and *** Estimated coefficients are statistically different from
zero at the 10%, 5%,, and 1% levels respectively.

Table 5. Cross-Sectional Estimates of Log(Wage) Model for Males

Variable Model One Model Two

Alcohol -- 0.105 (0.094)
Drinks every day 0.288 (0.303) --
Drinks 4-6 times/week 0.090 (0.277) --
Drinks 2-3 times/week 0.205 (0.164) --
Drinks once/week 0.214 * (0.130) --
Drinks 2-3 times/month -0.019 (0.128) --
Drinks once/month 0.068 (0.160) --
Ln(ethanol) -- --
Ln[(ethanol).sup.2] -- --
[R.sup.2] 0.056 0.051
Number of observations 4985 4985

Variable Model Three

Alcohol --
Drinks every day --
Drinks 4-6 times/week --
Drinks 2-3 times/week --
Drinks once/week --
Drinks 2-3 times/month --
Drinks once/month --
Ln(ethanol) 0.075 * (0.042)
Ln[(ethanol).sup.2] -0.021 * (0.014)
[R.sup.2] 0.055
Number of observations 4985

Robust standard errors are in parentheses.

*, **, and *** Estimated coefficients are statistically different from
zero al the 10%, 5%, and 1% levels, respectively.

Table 6. Cross-Sectional Estimates of Log(Wage) Model for Females

Variable Model One Model Two

Alcohol -- 0.164 ** (0.075)
Drinks every day -0.414 (0.694) --
Drinks 4-6 times/week -0.180 (0.806) --
Drinks 2-3 times/week 0.162 (0.274) --
Drinks once/week 0.218 * (0.115) --
Drinks 2-3 times/month 0.175 ** (0.097) --
Drinks once/month 0.136 *** (0.046) --
Ln(ethanol) -- --
Ln[(ethanol).sup.2] -- --
R2 0.060 0.059
Number of observations 5483 5483

Variable Model Three

Alcohol --
Drinks every day --
Drinks 4-6 times/week --
Drinks 2-3 times/week --
Drinks once/week --
Drinks 2-3 times/month --
Drinks once/month --
Ln(ethanol) 0.081 * (0.048)
Ln[(ethanol).sup.2] -0.010 ** (0.005)
[R.sup.2] 0.059
Number of observations 5483

Robust standard errors are in parentheses.

*, **, and *** Estimated coefficients are statistically different from
zero at the 10%. 5%, and 1% levels, respectively.

Table 7. Fixed Effects Estimates of Employment Model for Males

Variable Model One Model Two Model Three

Alcohol -- 0.007 (0.021) --
Drinks every day -0.109 (0.072) -- --
Drinks 4-6 times/week 0.064 (0.044) -- --
Drinks 2-3 times/week 0.010 (0.008) -- --
Drinks once/week 0.003 (0.015) -- --
Drinks 2-3 times/month 0.012 (0.022) -- --
Drinks once/month 0.009 (0.010) -- --
Ln(ethanol) -- -- 0.011 (0.008)
Ln[(ethanol).sup.2] -- -- 0.003 (0.004)
[R.sup.2] 0.145 0.122 0.143
Number of observations 2787 2787 2787

Standard errors are in parentheses.

*, **, and *** Estimated coefficients are statistically different
from zero at the 10%, 5%, and 1% levels, respectively.

Table 8. Fixed Effects Estimates of Employment Model for Females

Variable Model One Model Two Model Three

Alcohol -- 0.025 (0.019) --
Drinks every day -0.035 (0.068) -- --
Drinks 4-6 times/week -0.054 (0.059) -- --
Drinks 2-3 times/week 0.038 (0.037) -- --
Drinks once/week 0.044 (0.035) -- --
Drinks 2-3 times/month 0.013 (0.021) -- --
Drinks once/month 0.009 (0.012) -- --
Ln(ethanol) -- -- 0.012 (0.009)
Ln[(ethanol).sup.2] -- 0.007 (0.005) --
[R.sup.2] 0.158 0.147 0.153
Number of observations 2778 2778 2778

Standard errors are in parentheses.

*, **, and *** Estimated coefficients are statistically different from
zero at the 10%, 5%, and 1% levels, respectively.

Table 9. Fixed Effects Estimates of Log(Wage) Model for Males

Variable Model One Model Two

Alcohol -- 0.071 * (0.038)
Drinks every day 0.218 (0.214) --
Drinks 4-6 times/week 0.041 (0.032) --
Drinks 2-3 times/week 0.069 * (0.034) --
Drinks once/week 0.102 * (0.054) --
Drinks 2-3 times/month 0.066 ** (0.027) --
Drinks once/month 0.032 (0.021) --
Ln(ethanol) -- --
Ln[(ethanol).sup.2] -- --
[R.sup.2] 0.125 0.118
Number of observations 1454 1454

Variable Model Three

Alcohol --
Drinks every day --
Drinks 4-6 times/week --
Drinks 2-3 times/week --
Drinks once/week --
Drinks 2-3 times/month --
Drinks once/month --
Ln(ethanol) 0.012 * (0.007)
Ln[(ethanol).sup.2] 0.001 (0.002)
[R.sup.2] 0.121
Number of observations 1454

Standard errors are in parentheses.

*, **, and *** Estimated coefficients are statistically different from
zero at the 10%, 5%, and 1% levels, respectively.

Table 10. Fixed Effects Estimates of Log(Wage) Model for Females

Variable Model One Model Two

Alcohol -- 0.101 * (0.(156)
Drinks every day -0.243 (0.853) --
Drinks 4-6 times/week 0.086 * (0.044) --
Drinks 2-3 times/week 0.124 ** (0.058) --
Drinks once/week 0.181 (0.103) --
Drinks 2-3 times/month 0.113 * (0.056) --
Drinks once/month 0.094 (0.073) --
Ln(ethanol) -- --
Ln[(ethanol).sup.2] -- --
[R.sup.2] -- --
Number of observations 1507 1507

Variable Model Three

Alcohol --
Drinks every day --
Drinks 4-6 times/week --
Drinks 2-3 times/week --
Drinks once/week --
Drinks 2-3 times/month --
Drinks once/month --
Ln(ethanol) --
Ln[(ethanol).sup.2] 0.023 * (0.014)
[R.sup.2] -0.011 (0.008)
Number of observations 1507

Standard errors are in parentheses.

*, **, and *** Estimated coefficients are statistically different from
zero at the 10%, 5%, and 1% levels, respectively.

(1) One exception is Kenkel and Ribar (1994), who exploit the longitudinal nature of the National Longitudinal Survey of Youth (NLSY). However, their measure of alcohol consumption is heavy drinking, which is different from the one used in this article.

(2) See MacDonald and Shields (2001) for an excellent review of the literature.

(3) A number of studies also provide evidence of a positive effect of illicit drug use on earnings (Kaestner 1991; Gill and Michaels 1992). Most recently. Dave and Kaestner (2001) found a weak and indeterminate relationship between alcohol taxes and labor market outcomes, which implies that alcohol use does not adversely affect labor market outcomes. French, Roebuck, and Kebreau (2001) find that nonchronic drug use is significantly related to employment.

(4) In the period from 1985 to 1988, mortality in Russia and other parts of the Soviet Union fell sharply. Between 1984 and 1987, age-standardized deaths fell by 12% among males and by 7% among females (Leon et al. 1997). Deaths from alcohol-related causes were the most affected. The trend then reversed and, in the period from 1990 to 1995, mortality dramatically increased. Since 1995, the rates appear to have stabilized, although life expectancy in Russia remains worse than it had been at any time in the past half century (Rehm, Room, and Edwards 2001). The decline in mortality coincided with the period of the Gorbachev antialcohol campaign, which started in 1985 and collapsed in 1987. The increase in mortality, on the other hand. corresponds to the period of the dissolution of the former Soviet Union and the state's subsequent loss of control of the alcohol market. This temporal association suggests that the mortality fluctuations may be related to alcohol consumption. However. the scientific support for a link between alcohol and mortality in Russia is weak, and so far. all the evidence has been indirect and based on the presumed temporal association.

(5) Price of composite good is normalized to one.

(6) This function can be justified under the separability of alcohol demand from leisure and composite consumption (French, Roebuck. and Kebrean 2001; DeSimone 2002).

(7) The Russia Longitudinal Monitoring Survey (RLMS) data set has other potentially valid instruments, such as the religious preference variables. However, these variables exhibit very little variation across rounds.

(8) The RLMS is a collaborative effort of the University of North Carolina at Chapel Hill, the Russian Central Statistical Bureau (Goskomstat), and the All-Russia Center of Preventive Medicine. Detailed information about the survey can be found at http://www.cpc.unc.edu/projects/rlms.

(9) Data are collected for 160 survey sites (communities). These 160 sites are allocated into 38 primary sampling units based largely on geographical factors and level of urbanization. These sampling units are then collapsed into eight regions (See Table 1 for a list of these regions).

(10) The alcohol data from the RLMS have been shown to match well with estimates from other sources (Treml 1997: Jensen 2001).

(11) A similar algorithm is also used in Mullahy and Sindelar (1996).

(12) Occupations are coded according to the four-digit International Standard Classification of Occupations (ILO 1988). These codes are then collapsed into a single digit title, using the guidelines provided in the survey description.

(13) Individuals working in the agricultural sector are excluded because these jobs are likely to be temporary or seasonal jobs.

(14) In similar contexts, Mullahy and Sindelar (1996) used OLS to examine the impact of problem drinking on binary employment and unemployment outcomes, and Kenkel and Ribar (1994) used OLS to estimate the impact of alcohol consumption on binary marriage outcome.

(15) In order to minimize the influence of skewed data, the natural logarithm of the ethanol consumption is used. However, a pure logarithmic transformation cannot be implemented due to the presence of nondrinkers. To overcome this difficulty, the variable is redefined as Log(ethanol + 1) (French and Zarkin 1995).

(16) The occupation and the region of residence are estimated to be significant determinants of the wage rate. The occupation indicators are included in the wage model in order to capture any productivity differences across different occupations. It can be argued that the effect of occupation on earnings operates, in part. via education. The models that excluded occupation dummies did not change the implications of the alcohol consumption coefficients. I thank Michael Grossman for bringing this to my attention.

(17) One can consider random effects as art alternative to the fixed effects. A Hausman test yielded a p-value of less than 0.0001 for both employment and wage models for both genders, which suggests that differences between fixed effects and random effects are systematic. Therefore, I reject the use of random effects.

(18) Fixed effects models are not reestimated because there is not enough variation in these covariates to have any effect on wages or employment.

(19) The Russian Consumer Price Index for food and beverages is used to deflate the prices from different years to 1998 December rubles. The price information is missing for several sites in some of the rounds. These missing prices are replaced by the inflation-adjusted average prices from the rounds in which they are available.

(20) I thank an anonymous referee for this point.

(21) Glinskaya and Mroz (2001) defined the contractual wage as (total salary + [total amount owed/number of months owed]/hours worked).

References

Beaglehole, R., and R. Jackson. 1992. Alcohol, cardiovascular diseases and all causes of mortality: A review of the epidemiological evidence. Drug and Alcohol Review II:275-90.

Becker, S. Gary, and Kevin M. Murphy. 1988. A theory of rational addiction. Journal of Political Economy 96:675-700.

Berger, Mark C., and J. Paul Leigh. 1988. The effect of alcohol use on wages. Applied Economics 20:1343-51.

Bobak, Martin. Martin McKee, Richard Rose, and Michael Marmot. 1999. Alcohol consumption in a national sample of the Russian population. Addiction 94:857-66.

Bobak, Martin, Hynek Pikhart, Clyde Hertzman, Richard Rose, and Michael Marmot. 1998. Socioeconomic factors, perceived control and self-reported health in Russia. A cross-sectional survey. Social Science Medicine 47:269-79.

Brodsky, Archie, and Stanley Peele. 1999. Psychological benefits of moderate alcohol consumption: Alcohol's role in a broader conception of health and well-being. In Alcohol and pleasure: A health perspective, edited by S. Peele, and M. Grant. Philadelphia, PA: Brunner/Mazel, pp. 187-207.

Caetano, R. 1994. Drinking and alcohol-related problems among minority women. Alcohol Health and Research World 18: 233-42.

Cornia, G. A. 1997. Labor market shocks, psychological stress and the transition's mortality crisis. Working Paper, United Nations University, World Institute for Development Economics Research, Helsinki.

Cruze, A., H. Haywood, P. Kristiansen, J. Collins, and D. Jones. 1981. Economic costs of alcohol and drug abuse and mental illness--1977. Research Triangle Park, NC: Research Triangle Institute.

Dave, Dhaval, and Robert Kaestner. 2001. Alcohol taxes and labor market outcomes. NBER Working Paper No. 8562.

DeSimone, Jeff. 2002. Illegal drug use and employment. Journal of Labor Economics 4:924-52.

Doll, R., R. Peto, E. Hall. K. Wheatley, and R. Gray. 1994. Mortality in relation to consumption of alcohol: 13 years' observations on male British doctors. British Medical Journal 309:911-8.

Earle, John S., and Klara Z. Sabirianova. 2002. How late to pay? Understanding wage arrears in Russia. Journal of Labor Economics 20:661-707.

Farrel, Sarah. 1985. Review of national policy measures to prevent alcohol-related problems. Geneva: World Health Organization.

Fingarette, Herbert. 1988a. Heavy drinking. Berkeley, CA: University of California Press.

Fingarette, Herbert. 1988b. Alcoholism: The mythical disease. The Public Interest 91:3-22.

French, Michael T., and Gary A. Zarkin. 1995. Is moderate alcohol use related to wages? Evidence from four worksites. Journal of Health Economics 14:319-44.

French, Michael T., M. Christopher Roebuck, and Alexandre Pierre Kebreau. 2001. Illicit drug use, employment, and labor force participation. Southern Economic Journal 68:349-68.

Gill, Andrew, and Robert J. Michaels. 1992. Does drug use lower wages? Industrial and Labor Relations Review 45:419-34.

Glinskaya, Elena, and Thomas A. Mroz. 2001. The gender gap in wages in Russia from 1992 to 1995. Journal of Population Economies 13:353-86.

Griliches, Zvi, and Jerry Hausman. 1986. Errors in variables in panel data. Journal of Econometrics 31:93-118.

Hamilton, Vivian, and Barton H. Hamilton. 1997. Alcohol and earnings: Does drinking yield a wage premium? Canadian Journal of Economics 30:135-51.

Heien, Dale M. 1996. Do drinkers earn less? Southern Economic Journal 63:60-8.

International Labor Office (ILO). 1988. International standard classification of occupations, edited. Geneva: International Labor Office.

Jensen, Robert. 2001. Job security, stress, and health: Evidence from the Russian privatization experience. Paper presented at the Population Association of America 2002 Annual Meetings.

Kaestner, Robert. 1991. The effect of illicit drug use on the wages of young adults. Journal of Labor Economics 9:381-412.

Kenkel, D. S., and David C. Ribar. 1994. Alcohol consumption and young adults' socioeconomic status. Brookings Papers on Economics Activity: Microeconomics:119-61.

Leon, D. A., L. Chenet, V. M. Shkolnikov, S, Zakharov, J. Shapiro, G. Rakhmanova, S. Vassin, and M. McKee. 1997. Huge variation in Russian mortality rates 1984-1994: Artefact, alcohol, or what? Lancet 350:383-8.

Levitt, Steven D. 1998. Why do increased crime arrest rates appear to reduce crime: Deterrence, incapacitation, or measurement error? Economic Inquiry 36:353-72.

Lex, Barbara W. 1994. Alcohol and other drug abuse among women. Alcohol Health and Research World 3:212-9.

MacDonald, Ziggy, and Michael A. Shields. 2001. The impact of alcohol consumption on occupational attainment in England. Economica 68:427-53.

Malyutina, Sofia, Martin Bobak, Kurilovitch Svetlana, Valery Gafarov, Galina Simonova, Yuri Nikitin, and Michael Marmot. 2002. Relation between heavy and binge drinking and all-cause and cardiovascular mortality in Novosibirsk, Russia: A prospective cohort study. Lancet 360:1448-54.

Midanik, Lorraine T. 1982. The validity of self-reported alcohol consumption and alcohol problems: A literature review. British Journal of Addiction 77:357-82.

Montgomery, J. D. 1991. Social networks and labor-market outcomes: Toward an economic analysis. American Economic Review 81:1408-18.

Mullahy, John, and Jody L. Sindelar. 1991. Gender differences in labor market effects of alcoholism. American Economic Review (Papers and Proceedings) 81:161-5.

Mullahy, John, and Jody L. Sindelar. 1993. Alcoholism, work, and income. Journal of Labor Economics 11:494-520.

Mullahy, John, and Jody L. Sindelar. 1996. Employment, unemployment, and problem drinking. Journal of Health Economics 15:409-34.

Preston, Kenzie L., Kenneth Silverman, Charles R. Schuster, and Edward J. Cone. 1997. Comparison of self-reported drug use with quantitative and qualitative urinalysis for assessment of drug use in treatment studies. NIDA Research Monograph No. 167.

Putnam, R. D. 2000. Bowling alone: The collapse and revival of American community. New York: Simon and Schuster.

Quinn-Judge, Paul. 1997. Russian Roulette. Time Europe (online). 11 August. Accessed 19 May, 2004, Available http:// www.time.com/time/magazine/1997/int/970811/europe.rusian_roulet.html.

Rehm, Nina, Robin Room, and Griffith Edwards. 2001. Alcohol in the European region: Consumption, harm, and policies. Copenhagen: World Health Organization Regional Office for Europe.

Roman, P. M. 1988. Biological features of women's alcohol use: A review. Public Health Reports 103:628-37.

Rouse, B., N. Kozel, and L. Richards. 1985. Self-report methods of estimating drug use: Meeting current challenges to validity. Rockville, MD: National Institute on Drug Abuse.

Ryan, Michael. 1995. Alcoholism and rising mortality in the Russian federation. British Medical Journal 310:646-8.

Skog, Ole-Jurgen J. 1980. Social interaction and the distribution of alcohol consumption. Journal of Drug Issues 10:71-92.

Terza, Joseph V. 2002. Alcohol abuse and employment: A second look. Journal of Applied Econometrics 17:393-404.

Treml, Vladimir. 1997. Soviet and Russian statistics on alcohol consumption and abuse. In Premature death in the new independent states, edited by J. L. Bobadilla, C. A. Costello, and F. Mitchell. Washington, DC: National Academy Press, pp. 220-38.

Walberg, P., M. McKee, V. Shkolnikov, L. Chenet, and D. A. Leon. 1998. Economic change, crime, and the Russian mortality crisis: A regional analysis. British Medical Journal 317:312-8.

Wilsnack, Richard W., and Sharon C. Wilsnack. 1992. Women, work, and alcohol: Failures of simple theories. Alcoholism: Clinical and Experimental Research 16:172-9.

Zarkin, Gary, T. Mroz, J. Bray, and Michael French. 1998. Alcohol use and wages: New results from the national household survey on drug abuse. Journal of Health Economics 17:53-68.

Erdal Tekin, Department of Economics, Andrew Young School of Policy Studies, Georgia State University, University Plaza. Atlanta, GA 30303-3083 and the National Bureau of Economic Research: E-mail: [email protected].

I thank David Blau, Michael Grossman, Julie Hotchkiss, Naci Mocan, Paula Stephan, Volkan Topalli, and two anonymous referees for valuable comments. Djesika Amendah and Roy Wada provided excellent research assistance. All remaining errors are mine.

Received April 2003: accepted January 2004.