Illicit Drug Use, Employment, and Labor Force Participation.

Alexandre, Pierre Kebreau

Michael T. French [*]

M. Christopher Roebuck [+]

Pierre Kebreau Alexandre [+]

Illicit drug use has declined among the U.S. adult population, but national surveys show the majority of illicit drug users are employed. Concern about workplace productivity, absenteeism, and safety has led many employers to establish employee assistance and drug testing programs. Given the sharp interest in workplace interventions, more information is needed about the relationships between drug use and labor market status. This study estimated the probability of employment and labor force participation for different types of drug users using nationally representative data from the 1997 National Household Survey on Drug Abuse. Results strongly indicated that chronic drug use was significantly related (negative) to employment for both genders and labor force participation for males. Furthermore, nonchronic drug use was not significantly related to employment or labor force participation. These findings suggest that workplace policies for illicit drug use should consider chronic or problem drug users apart fro m light or casual users.

1. Introduction

Illicit drug abuse often leads to significant personal and social consequences, including labor market problems (Harwood, Fountain, and Livermore 1998). Casual drug use (vs. chronic drug use or drug abuse), however, has uncertain effects on job performance, especially when consumption occurs away from the workplace and a relatively lengthy period of time has elapsed between consumption and work. A growing body of literature has examined the relationships between drug use, drug abuse, and labor market measures (for recent reviews, see French 1993; Kaestner 1998). The majority of these studies have focused on wages and earnings (e.g., Kaestner 1991, 1994a; Gill and Michaels 1992; Kandel, Chen, and Gill 1995; French and Zarkin 1995; Heien 1996; Buchmueller and Zuvekas 1998; French, Zarkin, and Dunlap 1998). Researchers have also examined the relationships between substance use and labor supply measures such as employment status, labor force participation, annual weeks employed, or daily hours worked (e.g., Regis ter and Williams 1992; Kaestner 1994b; Mullahy and Sindelar 1996; Buch-mueller and Zuvekas 1998; Zarkin et al. 1998; French et al. 2001). In general, the labor supply results are inconsistent across studies.

Although the existing literature suggests that the relationships between drug use and wages are complex and perhaps ambiguous (Normand, Lempert, and O'Brien 1994; Kaestner 1998), the principal workplace impact may involve employment status and labor supply rather than earnings per se (Mullahy and Sindelar 1996; French et al. 2001). One of the most ambitious studies of substance use and employment status was by Mullahy and Sindelar (1996). These authors derived various estimates of the effects of problem drinking on employment and unemployment. [1] Specifically, they first estimated the relationships with ordinary least squares (OLS, linear probability model) and then with an instrumental variables (IV) approach. [2] Using the 1988 Alcohol Supplement of the National Health Interview Survey, they found that problem drinking was associated with lower employment and higher unemployment for males (OLS). In contrast, female problem drinkers were more likely to be both employed and unemployed (OLS). The IV results w ere consistent in sign with the OLS results for males, but the IV estimates were not statistically significant for any of the problem drinking definitions. For females, the IV findings were different from the OLS findings regarding employment, but the coefficient estimates were not significant. [3]

The research by Mullahy and Sindelar (1996) was a significant contribution to the literature because it was one of the first studies that estimated the association between substance use and employment status. The authors developed several measures of alcohol use and used an IV technique to control for the endogeneity of problem drinking. Despite these notable features, the study can be extended and modified in several ways. For example, no recent studies have used national surveys of the adult household population to examine the relationships between illicit drug use and labor market status. Second, Mullahy and Sindelar did not examine labor force participation or different measures of alcohol use in the same specification. Moreover, none of the coefficient estimates for problem drinking in the IV models used by Mullahy and Sindelar was significant, which could indicate the need to secure better instruments. As noted by the authors, "... econometric evidence presented here must be weighed carefully in light o f the reasonableness of the identifying restrictions invoked to estimate such models" (p. 432). Mullahy and Sindelar then proceeded to encourage additional research in this area with new data sets and improved measures.

The present article is an extension of earlier studies (particularly Mullahy and Sindelar 1996) and a contribution to the growing literature on the economic effects of drug use. Specifically, the analysis focused on employment and labor force participation. [4] Second, chronic drug use was examined rather than problem drinking. Third, data were obtained from a recent national survey, the 1997 National Household Survey on Drug Abuse, with excellent measures of drug use and labor market status. Finally, nonchronic drug use was examined along with chronic drug use to determine whether the labor supply effects extended to any type of drug use.

The rest of the article is organized as follows. Section 2 presents theoretical background for the study. Section 3 introduces the empirical models. The sample and data are explained in section 4. The results are presented in section 5, and the article concludes with a discussion (section 6).

2. Theoretical Background

Several theoretical or conceptual issues require careful consideration when investigating the relationships between illicit drug use and labor market behavior. For example, the direction of causality is complicated and uncertain. Does drug use lead to employment problems or does a lack of employment foster drug use? Empirically, however, the data requirements (e.g., longitudinal data with good price information and other instrumental variables) that are necessary to statistically address this issue are often beyond the reach of most investigators.

A completely different but equally important conceptual issue concerns the definition of drug use. Regarding a legal substance such as alcohol, most investigators recognize that a binary measure of alcohol use would be inappropriate because light or occasional use may actually be beneficial rather than problematic (Shaper 1990; Marmot and Brunner 1991; Coate 1993; French and Zarkin 1995; Heien 1996; Kannel and Ellison 1996). Illicit drug use is generally viewed differently than alcohol, and investigators often estimate relationships between any illicit drug use and health or labor market consequences (Kaestner 1991; Gill and Michaels 1992). Only recently have studies begun to measure the quantity/frequency dimensions of use and distinguish between the use of different types of illicit drugs (Buchmueller and Zuvekas 1998; French et al. 2000; McGeary and French 2000). The present article follows the increasing emphasis on problematic drug use by examining chronic drug users relative to nonchronic drug users and nonusers.

The measurement of employment is also important for a study of drug use and labor market status. Most survey questionnaires record employment status at the time the interview is administered. But, defined in this manner, employment status may be misrepresented for individuals who worked continuously for several months and then stopped working at the time of the interview. Alternatively, some individuals may secure a job shortly before the interview and be designated as employed even though they worked little during the preceding months. Similar issues may exist for measurement of labor force participation. Consider an individual who continuously worked for several months, became unemployed, and had not looked for work during the past 30 days prior to the interview. This person would be classified as not in the labor force. Although these measurement concerns probably only apply to a small percentage of survey respondents, it is important to recognize that changes in variable definitions and the timing of surv ey administration could have an impact on the empirical findings.

The present analysis considers two measures of illicit drug use (based on consumption over the prior 12 months) and two measures of labor market status (pertaining to the time of survey administration). To guide the empirical work that follows, the conceptual framework advanced by Mullahy and Sindelar (1996) is adapted. Specifically, assuming that product and input markets are competitive and individuals are price and wage takers, the implicit functions for drug use (D) and labor market status (L) can be specified as

D = D(p, w, [X.sub.D], [[alpha].sub.D]) (1)

L = L(p, w, [X.sub.L], [[alpha].sub.L]), (2)

where p is a vector of all prices, w is a vector of all wages, [X.sub.D] and [X.sub.L] are vectors of all observable factors (e.g., demographics, human capital, health status, geographic controls) that influence D and L, and [[alpha].sub.D] and [[alpha].sub.L] are vectors of unobservable characteristics that are related to D and L. In addition to p and w, other common variables are likely present in the X's and [alpha]'s across Equations 1 and 2.

Structural equations for D and L can be derived through a static utility maximization model of drug use and leisure subject to a full budget constraint. A flexible form of the utility function is assumed whereby preferences for drug use and leisure are implicitly separable from other prices (Browning, Deaton, and Irish 1985; Blundell and Meghir 1986; Kaestner 1994a). Such a process could also lead to a set of reduced-form equations similar to the following functions:

D = D(L, X, [alpha]) (3)

L = L(D, X, [alpha]), (4)

where X includes all variables in [X.sub.D] and [X.sub.L], and [alpha] includes all factors in [[alpha].sub.D] and [[alpha].sub.L]. The effect of D on L in Equation 4 is the relationship of interest for this article. However, as suggested by Mullahy and Sindelar (1996), estimation of Equation 4 with standard techniques such as OLS or probit would generate a biased result because D is correlated with [alpha]. Specifically, the effect of D on L can be written as

dL/dD = [L.sub.D] + [L.sub.[alpha]]d[alpha]/dD. (5)

An efficient or consistent estimate of dL/dD can be obtained by addressing the endogeneity of D in the labor market equation through variations of a two-stage IV technique (Register and Williams 1992; Kaestner 1994b; Mullahy and Sindelar 1996; Greene 1997; Norton, Lindrooth, and Ennett 1998; Evans, Farrelly, and Montgomery 1999; Alexandre and French 2000). The following section outlines the empirical approach.

3. Empirical Models

The empirical approach involved estimation of several specifications of employed and labor force participation. Each specification included a measure of chronic drug use or nonchronic drug use during the past year. [5] All analyses were gender specific to account for the differences in substance use and labor market behavior between men and women (Mullahy and Sindelar 1991). The research was designed to test the following two hypotheses:

H1: Chronic drug users are less likely to be employed or in the labor force relative to nonchronic drug users and nondrug users.

H2: Nonchronic drug users have statistically similar employment and labor force participation probabilities relative to nondrug users.

To test these hypotheses, the analysis began with a standard probit specification that can be expressed as follows:

Pr([L.sub.i] = 1) = [phi]([X.sub.i][beta]), (6)

where [L.sub.i] is an indicator variable for either of the two labor market variables (employment and labor force participation); [X.sub.i] is a vector of variables that influence labor market status, including demographic characteristics (e.g., age, race, education, marital status), health status, geographic controls, and a dichotomous measure of illicit drug use; [beta] is a vector of parameters to estimate; and [phi] is the cumulative normal distribution. Univariate probit models were estimated and marginal effects were calculated for chronic drug use and nonchronic drug use.

As described in the previous section, using the univariate probit technique to estimate Equation 4 could lead to biased results if drug use is correlated with the unobservable characteristics that influence [L.sub.i]. The Hausman--Wu test (Wu 1973; Hausman 1983) is often used to test for the potential endogeneity of substance use, but because the models are nonlinear, the analysis used the method suggested by Smith and Blundell (1986). The null hypothesis that drug use was exogenous involved a [[chi].sub.2]-test of the explanatory power of the residuals from the first-stage equation for drug use when added to Equation 6. Similar to the Hausman--Wu test, the Smith--Blundell test is influenced to a large degree by the reliability of the instruments (Nelson and Startz 1990; Bollen, Guilkey, and Mroz 1995; Bound, Jaeger, and Baker 1995; Staiger and Stock 1997; Norton, Lindrooth, and Ennett 1998). Thus, [[chi].sub.2]-tests for the significance of the instrument in explaining drug use were used to test instrument reliabili ty (Bollen, Guilkey, and Mroz 1995; Ettner 1996; Staiger and Stock 1997; Norton, Lindrooth, and Ennett 1998). [6]

The initial step to implement the IV technique was to estimate a first-stage probit for drug use ([D.sub.i]) with the observable explanatory variables noted earlier ([Z.sub.i]) and a composite measure of three additional variables ([A.sub.i]) as

Pr([D.sub.i] = 1) = [phi]([Z.sub.i][gamma] + [A.sub.i][delta]), (7)

where [gamma] and [delta]are vectors of parameters to estimate and [A.sub.i] is a dichotomous variable equal to one if the respondent agreed or strongly agreed with each of the following statements: "Religious beliefs are important to me," "Religious beliefs influence my decisions," and "It is important that my friends share my religious beliefs." This composite measure for religiosity is intended to identify the strongly religious individuals in the sample. The expectation is that religiosity is inversely related to drug use because many organized religions advocate a wholesome lifestyle that is free of unhealthy substances (Cochran, Beeghley, and Bock 1988; Fetzer Institute 1999). Illicit drug use is often discouraged because it is an illegal activity and it interferes with spiritual pursuits. Moreover, Schwartz and Huismans (1995) found that religious people place less value on self-indulgence and pleasure seeking, which are often motivations for drug use. Religious belief measures were used as instruments in pre vious studies, and results consistently showed a negative and significant relationship with drug use (Register and Williams 1992; Kaestner 1994a; Alexandre and French 2000).

The second-stage probit estimated the probability of being employed and the probability of being in the labor force with the corresponding explanatory variables ([Z.sub.i]) and the predicted values of drug use (D) from the first-stage probit (Maddala 1983) as

Pr([L.sub.i] = 1) = [phi]([Z.sub.i][beta] + [D.sub.i][[beta].sub.D]). (8)

All variables are the same as defined previously. The use of D in Equation 8 introduced a potential measurement error in the estimated standard errors (Maddala 1983; Murphy and Topel 1985). Since Monte Carlo evidence has not revealed any significant gain in estimating the more complex adjusted standard errors (Bollen, Guilkey, and Mroz 1995), the present analysis reports the unadjusted standard errors. [7] Finally, marginal effects and corresponding standard errors were calculated for D and D, and coefficient estimates were compared across the two specifications (univariate probit and IV).

4. Sample and Data

The 1997 National Household Survey on Drug Abuse (NHSDA) was the 17th survey in a series that was begun in 1971. The sample design was a nationally stratified multistage area probability sample of the noninstitutionalized household population in the 50 United States who were 12 years of age and older. Various segments of the population were oversampled, including youth, minorities, and current smokers aged 18-34. A major change in questionnaire design was initiated in 1994, which placed more emphasis on health status and health care, access to care, and mental health (SAMHSA 1999a, b). Thus, some of the measures in 1997 are not comparable with the measures from 1994-A and earlier surveys.

Although the NHSDA was one of the largest surveys of drug use ever undertaken in the United States, it had certain limitations that affected the analyses (SAMHSA 1999a). Most importantly, the data were self-reported, which raises questions regarding validity and reliability. NHSDA procedures were designed to maximize honesty and recall, but ultimately the value of the data depends on respondents' truthfulness and memory. A few studies have examined the validity of self-reported drug use information in this context and have found the measures to be quite good (Rouse, Kozel, and Richards 1985; Turner, Lessler, and Devore 1992; Harrison and Hughes 1997; Preston et al. 1997).

Another limitation of the NHSDA was the cross-sectional design. It would be useful to analyze and report longitudinal changes in drug use patterns and how these changes were related to labor market status for a panel of individuals. However, the NHSDA cannot examine these issues because a new cohort was sampled every year.

Furthermore, a small segment of the U.S. population (slightly less than 2%) was excluded from the sampling frame because they were not part of the target population. The excluded subpopulations were members of the active duty military and persons in institutional group quarters (e.g., hospitals, prisons, nursing homes, treatment centers). For additional details on sample design, response rates, major findings, and other technical details of the 1997 NHSDA, refer to SAMHSA (1999a).

For the present study, drug users were divided into two categories based on criteria defined by the Office of National Drug Control Policy (ONDCP 1996). Chronic drug users included all individuals who used one or more illicit drugs weekly or more often during the past year. Nonchronic drug users included individuals who used any illicit drug during the past year but not in a chronic nature. French et al. (in press) determined that the criteria for chronic drug use was significantly correlated with clinically based criteria for problematic drug use, and both measures produced similar results in health services demand models. As described earlier, information on labor market status was consistent with Bureau of Labor Statistics definitions for employed and labor force participation (BLS 2000).

Table 1 (males) and Table 2 (females) report sample means for all of the variables that were used in the empirical models. The statistics were segmented between drug-using status (chronic drug users, nonchronic drug users, and nondrug users) and the full sample. Kruskall--Wallis rank tests were performed to determine significant differences in mean values across the drug-using categories. Individuals under the age of 25 or over the age of 59 were excluded from the analysis given their unique employment choices (e.g., full-time students, new labor market participants, occasional workers, and retirees). [8] The following section reports the findings of the statistical analysis when employment and labor force participation were estimated with univariate probit and IV procedures.

5. Results

Results of the multivariate analyses are presented in Tables 3-6. The first-stage probit results for chronic drug use are reported in Table 3. The [[chi].sup.2]-statistics for the significance of the composite religiosity instrument in the first-stage equations were significant (p [less than] 0.05), indicating that the instrument was a good predictor of chronic drug use for both males and females. [9]

The full estimation output (probit and IV) is reported for the chronic drug use specification (Table 4 for males and Table 5 for females), and selected results are reported in Table 6 for both genders. Starting with the employment models, the univariate probit results are followed by those from the IV estimation technique. The same format is used for the labor force participation models. The important estimates correspond to the chronic-drug-use variable, the marginal effects of chronic drug use, and the [[chi].sup.2]-statistics for the test of potential endogeneity of chronic drug use.

Looking first at Table 4 for males for both the employment and the labor force participation specifications, the analysis failed to reject the null hypothesis that chronic drug use was exogenous. These specification tests suggested that the univariate probit specifications were appropriate for the male group. The probit findings indicated that chronic drug use was associated with a 0.089 (p [less than] 0.01) decrease in the probability of being employed and a 0.037 (p [less than] 0.05) decrease in the probability of participating in the labor force. [10]

Numerous other coefficients were significant in the employment and labor force participation equations. Age was nonlinear in both employment and labor force participation, with optimum values at 34.0 years for employment and 36.2 years for labor force participation. White men were more likely to be employed and to participate in the labor force. Marriage was positively related to employment and labor force participation, and education was positively related to employment. The number of people living in the household was significantly related to labor force participation but not to employment. Finally, men who were in fair health or better were more likely to be employed and to be in the labor force.

The results for women are reported in Table 5, and with few exceptions, they were qualitatively similar to the results for men. Since the exogeneity test failed to reject the null hypothesis that chronic drug use was an exogenous variable in both the employment and labor force participation models, the analysis focused on the estimates from the univariate probit. These results suggested that chronic drug use was negatively related to employment (p [less than] 0.05) but not significantly related to labor force participation. [11] Specifically, being a female chronic drug user was associated with a 0.091 (p [less than] 0.05) decrease in the probability of employment.

Several other coefficient estimates were significant for the female sample. Age was nonlinear for both employment and labor force participation, with optimum values at 41.1 years for employment and 40.3 years for labor force participation. As expected, married women were less likely to be employed and in the labor force, and education was positively associated with both labor market variables. The number of moves in the past year was negatively related to employment, and the number of people in the household was negatively associated with both employment and labor force participation. As with men, health status was strongly associated with employment and labor force participation.

To determine whether light or casual drug use had similar adverse effects in the labor market relative to chronic drug use, the univariate probit specifications described earlier were reestimated with chronic drug use and nonchronic drug use in the same equations. Table 6 provides regression results for employment and labor force participation (men and women). The primary results can be summarized as follows. [12] First, the significant results for chronic drug use in the employment (men and women) and labor force participation (men) specifications continued to prevail when nonchronic drug use was included in the model. [13] In addition, the estimated marginal effects of chronic drug use were almost identical to the previous estimates.

Second, the coefficient estimates for nonchronic drug use were small and insignificant in each of the four specifications. This suggests that the adverse labor supply effects of illicit drug use can be mainly attributed to chronic drug users rather than all illicit drug users.

Third, illicit drug use (chronic and nonchronic) was unrelated to labor force participation for females. Although this result was counter to the labor force participation findings for males, it suggests that females have less stable and continuous labor force patterns than males, and illicit drug use is not a strong mediator in this process. This result is not unique, however, as other substance use and labor market studies have found opposite results for men and women (Mullahy and Sindelar 1991; Kaestner 1994a).

Finally, in addition to univariate probit and IV specifications, the analysis also estimated bivariate probit models of chronic drug use and employment status to account for the joint decision to use drugs and work. The estimated correlation (p) among equations ranged from 0.4920 to -0.4557 but was statistically significant (p [less than] 0.05) in only one of the four specifications-chronic drug use and employed for males. Thus, the bivariate probit results largely confirmed the Smith and Blundell (1986) tests for exogeneity of chronic drug use, suggesting that the univariate probit results were unbiased, consistent, and preferable (smaller mean squared error) to the IV results (Davidson and MacKinnon 1993; Norton, Lindrooth, and Ennett 1998). Nevertheless, the IV results are presented in Tables 4 and 5 for comparison purposes. All of the results discussed above are available upon request from the corresponding author.

6. Discussion

The primary objective of this study was to estimate the relationship between illicit drug use and labor market status while addressing the possibility of drug use as an endogenous variable. Nationally representative data were used to analyze the association between two measures of drug use and two measures of labor supply. The reliability of the religiosity instrument was assessed through a significance test in the first-stage probit for the drug use variables. Exogeneity tests were executed for chronic drug use, and results were presented for both a univariate probit specification and an IV estimation technique. The overall analysis was similar to the approach in Mullahy and Sindelar (1996) to establish comparability between their results for problem drinking and the present findings for problem drug use.

The key findings can be summarized as follows. First, the exogeneity of chronic drug use was not rejected in any specification. Second, chronic drug use was significantly related to employment (negative) for both genders and to labor force participation (negative) for males. Third, nonchronic drug use was not statistically related to either of the labor supply measures, indicating that light or casual drug use did not lead to negative effects on labor supply. Finally, quantitative differences were present across genders, but the qualitative findings were similar, except for the chronic drug use measure in the labor force participation model.

The present research with chronic drug users was consistent with the problem drinking research by Mullahy and Sindelar (1996) in the sense that both studies found that problematic substance use had the predicted negative effect on employment. However, unlike the results for chronic drug use, Mullahy and Sindelar found that problem drinking was endogenous and stronger gender differences were present with problematic alcohol consumption than for chronic drug use. Furthermore, the marginal effects for chronic drug use were generally larger than the marginal effects for problem drinking. Since Mullahy and Sindelar did not examine the effects of illicit drug use on labor force participation, a direct comparison of these results was not possible.

Some qualifications of the findings should be noted that offer opportunities for future analyses. Most of the research shortcomings pertain to data limitations. For example, it would have been interesting to analyze panel data in addition to cross-sectional data to control for personal and environmental fixed effects. Panel data would have also allowed the investigation of longer term effects from heavy and prolonged drug use. Second, many illicit drug users were also heavy alcohol users. However, it was difficult to determine the relative contribution of these substances (and others such as prescription drugs and tobacco) to labor market problems due to multicollinearity and other concerns (see footnote 5). Third, precision of the exogeneity tests and the IV estimates could have been improved if other instruments were available in addition to the religiosity variable that was used in the first-stage probit (Eqn. 7). Illicit drug prices and drug policies are conceptually appealing instruments, but these measu res were not available on the NHSDA and obtaining authorization to merge the data proved unsuccessful. Recent studies, however, suggest that currently available data on illicit drug prices may be unreliable (e.g., Chaloupka et al. 1999; Farrelly et al. 1999).

In conclusion, this study found that chronic drug use was significantly related to employment status for men and women. On the other hand, male chronic drug users were less likely to participate in the labor force, but no significant relationship existed between chronic drug use and labor force participation for females. [14] Perhaps the most important finding of this study, however, was the lack of any significant relationships between nonchronic drug use, employment, and labor force participation. An implication of this finding is that employers and policy makers should focus on problematic drug users in the same way that they focus on problematic alcohol users. Additional research is encouraged to determine whether this important finding endures with other data sets and different labor market variables.

(*.) University of Miami (D93), Department of Epidemiology and Public Health, 1801 NW 9th Avenue, Third Floor, Miami, FL 33136, USA; E-mail mfrench@miami.edu; corresponding author.

(+.) Health Services Research Center and Department of Epidemiology and Public Health (D93), University of Miami, 1801 NW 9th Avenue, Third Floor, Miami, FL 33136.

Financial assistance for this study was provided by the Robert Wood Johnson Foundation (grant 035152). This research was completed while Michael French was a visiting professor at the Research Center for Health and Economies, Department of Economics, Pompeu Fabra University, Barcelona, Spain. Jay Bhatttacharya, Chee-Ruey Hsieh, Kathryn McCollister, Kerry Anne MeGeary, Helena Salome, Silvana Zavala, two anonymous referees, and participants at the 1999 Western Economic Association Conference, the 2000 Pacific Rim Allied Economics Association Conference, and the labor seminar series at Pompeu Fabra University provided helpful suggestions on earlier versions of the article. Carmen Martinez offered excellent administrative support.

Received March 2000; accepted January 2001.

(1.) The Bureau of Labor Statistics (2000) established the following standardized definitions for employed, unemployed, and not in the labor force. Employed comprises (a) all persons, who, during the reference week, did any work for pay or profit (minimum of an hour's work) or worked 15 hours or more as unpaid workers in a family enterprise and (b) all persons who were not working but who had jobs or businesses from which they were temporarily absent for noneconomic reasons whether they were paid for the time off or were seeking other jobs. Unemployed comprises those persons who had no employment during the reference week, who made specific efforts to find a job within the previous 4 weeks, and who were available for work during that week except for temporary illness. Persons on layoff from a job and expecting recall also are classified as unemployed. All other persons 16 years old and over who do not fit into one of the categories above are considered not in the labor force.

(2.) If unmeasured or unobserved factors are correlated with both substance use and labor supply (or wages), then the error term in the labor supply equation will be jointly distributed. Consequently, the coefficient estimate for substance use in a labor supply equation will be biased if the estimator does not correct for the endogeneity of substance use. Instrumental variables (IV) techniques are the most commonly exercised endogeneity correction methods, and several studies have employed these techniques in substance use research (Kaestner 1991; Register and Williams 1992; Mullahy and Sindelar 1996; Norton, Lindrooth, and Ennett 1998; Alexandre and French 2000).

(3.) Although Mullahy and Sindelar (1996) did not explicitly examine age effects of problem drinking, some of their earlier research established that problem drinking and associated consequences varied over the life cycle (Mullahy and Sindelar 1993).

(4.) It is hypothesized that ding use will be differentially related to employment and labor force participation. Namely, employment represents an agreement between an employer and employee. Thus, an individual cannot independently decide to become employed unless he or she is self-employed. Alternatively, labor force participation is largely a personal choice. An individual who is unable to find an agreeable employment contract can still be part of the labor force if he or she chooses to actively seek work. For these reasons, it is advisable to examine employment and labor force participation as separate conditions in the empirical models.

(5.) Information on alcohol consumption was also available from the survey questionnaire. However, a measure for alcohol consumption was not included in the final specifications because drug use and alcohol use were highly correlated (p [less than] 0.0001) and including both variables in the model led to concerns about multicollinearity. In addition, alcohol use is also potentially endogenous, and controlling for the endogeneity of both drug use and alcohol use would add considerable complexity to the analysis. Finally, the present analysis was intended to provide results that could be directly compared with those from Mullahy and Sindelar (1996). These authors did not include measures of both alcohol use and illicit drug use in their models. Despite these qualifications, a measure of alcohol use (i.e., total number of drinks during the previous 30 days) was added to the univariate probit specifications for both males and females. The impact on the ding use coefficients was very small compared with the presen t results and statistical significance was never altered. These findings are available on request.

(6.) Since the number of excluded instruments (religiosity) is equal to the number of potentially endogenous variables (chronic drug use) in the present study, the model is exactly identified. Thus, tests for overidentifying restrictions of the instruments do not apply (Davidson and MacKinnon 1993).

(7.) In similar studies, Evans, Farrelly, and Montgomery (1999), Norton, Lindrooth, and Ennett (1998), Bollen, Guilkey, and Mroz (1995), and Kaestner (1994a) followed the same approach as the present analysis and did not adjust the estimated standard errors.

(8.) This age criterion is identical with the approach in Mullahy and Sindelar (1996).

(9.) Analyses were also conducted with the three dichotomous measures for religious beliefs that were described earlier instead of the composite measure for strongly religious. Since the results were very similar in both cases, we only report the estimates that apply to the composite measure of religiosity. However, the additional analyses are available from the corresponding author.

(10.) All estimates for marginal effects were calculated using the mean values for the independent variables as outlined in Greene (1997). The estimated marginal effects of the binary variable chronic drug use (CDU) on labor market outcome (L) were approximated using the formula

[partial]E(L/[x.sub.*])/[partial]CDU = exp([beta]'x/[x.sub.*], CDU = 1) - exp([beta]'x/[x.sub.*], CDU = 0)

= exp([beta]'[x.sub.*] + [[beta].sub.CDU]) - exp([beta]'[x.sub.*]),

where x is the vector of explanatory variables, [x.sup.*] is the vector of mean values of x, excluding CDU, and [beta] is the vector of coefficient estimates.

(11.) It may be interesting to note that these significance tests produced the same qualitative result whether one used the probit or IV specification.

(12.) The first-stage probit results arc not reported for nonchronic drug use because the religiosity instrument was significant in every specification (p [less than] 0.05) and the quantitative results were similar to the estimates reported for chronic drug use in Table 3.

(13.) The earlier specification with chronic drug use were actually more conservative tests for the significance of chronic drug use in the labor supply equations because the analysis was relative to the combined group of nonchronic drug users and nondrug users rather than nondrug users only.

(14.) One possible explanation for this result for females could involve the way in which labor force participation and employment decisions are made. As noted earlier, whether or not to participate in the labor force is a choice largely made by the individual. Conversely, employment is a bilateral agreement between the individual and the employer. Familial obligations and/or workplace policies (e.g., family leave policies, workplace drug policies and testing programs) may be effectively screening nut female drug users, thereby negatively impacting their employment status. Future analyses are planned to explore the validity of this hypothesis.

References

Alexandre, Pierre K., and Michael T. French. 2000. Community-based relationships between illicit drug use and labor supply. Unpublished paper, University of Miami.

Blundell, Richard, and Costas Meghir. 1986. Selection criteria for a microeconometric model of labour supply. Journal of Applied Econometrics 1:55-80.

Bollen, Kenneth A., David K. Guilkey, and Thomas A. Mroz. 1995. Binary outcomes and endogenous explanatory variables: Tests and solutions with an application to the demand for contraceptive use in Tunisia. Demography 32:111-31.

Bound, John, David A. Jaeger, and Regina M. Baker. 1995. Problems with instrumental variables estimation when the correlation between the instruments and the endogenous explanatory variable is weak. Journal of the American Statistical Association 90:443-50.

Browning, Martin, Angus Deaton, and M. Irish. 1985. A profitable approach to labor supply and commodity demands over the life-cycle. Econometrica 53:503-43.

Buchmueller, Thomas C., and Samuel H. Zuvekas. 1998. Drug use, drug abuse, and labour market Outcomes. Health Economics 7:229-45.

Bureau of Labor Statistics (BLS). 2000. Employment and earnings. Washington, DC: Bureau of Labor Statistics.

Chaloupka, Frank J., Rosalie L. Pacula, Matthew C. Farrelly, Lloyd D. Johnston, and Patrick M. O'Malley. 1999. Do higher cigarette prices encourage youth to use marijuana? NBER Working Paper No. W6939.

Coate, Douglas. 1993. Moderate drinking and coronary heart disease mortality: Evidence from NHANES I and the NHANES I follow-up. American Journal of Public Health 83:888-90.

Cochran, J. K., L. Beeghley, and E. W. Bock. 1988. Religiosity and alcohol behavior: An exploration of reference group theory. Sociology Forum 3:256-76.

Davidson, Russel, and James G. MacKinnon. 1993. Estimation and inference in econometrics. New York: Oxford University Press.

Ettner, Susan L. 1996. New evidence on the relationship between income and health. Journal of Health Economics 15:67-85.

Evans, William N., Matthew C. Farrelly, and Edward Montgomery. 1999. Do workplace smoking bans reduce smoking? American Economic Review 89:728-47.

Farrelly, Matthew C., Jeremy W. Bray, Gary A. Zarkin, B. W. Wendling, and Rosalie L. Pacula. 1999. The effects of prices and policies on the demand for marijuana: Evidence from the National Household Surveys on Drug Abuse. NBER Working Paper No. W6940.

Fetzer Institute. 1999. Multidimensional measurement of religiousness/spirituality for use in health research. Rockville, MD: U.S. Department of Health and Human Services, National Institute on Aging.

French, Michael T. 1993. The effects of alcohol and illicit drug use in the workplace: A review. Journal of Employee Assistance Research 2:1-22.

French, Michael T., Alphonse G. Holtmann, Kerry Anne McGeary, and Gary A. Zarkin. 2001, Substance use and workplace attendance. In Economic analysis of substance use and abuse: The experience of developed countries and lessons for developing countries, edited by Michael Grossman and C. R. Hsieh. Cheltenham, UK: Edward Elgar Publishing.

French, Michael T., Kerry Anne McGeary, Dale D. Chitwood, and Clyde B. McCoy. 2000. Chronic illicit drug use, health services utilization, and the cost of medical care. Social Science and Medicine 50:1703-13.

French, Michael T., M. Christopher Roebuck, Kerry Anne McGeary, Dale D. Chitwood, and Clyde B. McCoy. In press. Using the Drug Abuse Screening Test (DAST-10) to analyze health services utilization for substance abusers in a community-based setting. Substance Use and Misuse.

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., Gary A. Zarkin, and Laura J. Dunlap. 1998. Illicit drug use, absenteeism, and earnings at six U.S. worksites. Contemporary Economic Policy 16:334-46.

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

Greene, William H. 1997. Econometric analysis. 3rd edition. Upper Saddle River, NJ: Prentice Hall.

Harrison, Lana, and Arthur Hughes (editors) 1997. The validity of self-reported drug use: Improving the accuracy of survey estimates. National Institute on Drug Abuse Research Monograph 167, NIH Publication No. 97-4147. Rockville, MD: U.S. Department of Health and Human Services, National Institute on Drug Abuse.

Harwood, Henrick, Douglas Fountain, and G. Livermore. 1998. The economic costs of alcohol and drug abuse in the United States, 1992. National Institute on Drug Abuse/National Institute on Alcohol Abuse and Alcoholism, NIH Publication No. 98-4327. Rockville, MD: U.S. Department of Health and Human Services, National Institute on Drug Abuse/National Institute on Alcohol Abuse and Alcoholism.

Hausman, Jerry A. 1983. Specification and estimation of simultaneous equation models. In Handbook of econometrics. Volume I. Amsterdam: North Holland.

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

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

Kaestner, Robert. 1994a. New estimates of the effect of marijuana and cocaine use on wages. industrial and Labor Relations Review 47:454-67.

Kaestner, Robert. 1994b. The effect of illicit drug use on the labor supply of young adults. Journal of Human Resources 29:126-55.

Kaestner, Robert. 1998. Illicit drug use and labor market outcomes: A review of economic theory and its empirical implications. Journal of Drug issues 28:663-80.

Kandel, Denise B., Kevin Chen, and Andrew Gill. 1995. The impact of drug use and earnings: A life-span perspective. Social Forces 74:243-70.

Kannel, W. B., and R. C. Ellison. 1996. Alcohol and coronary heart disease: The evidence for a protective effect. Clinica Chimica Acta 246:59-76.

Maddala, G. S. 1983. Limited dependent and qualitative variables in econometrics. Cambridge, MA: Harvard University Press.

Marmot, Michael, and Eric Brunner. 1991. Alcohol and cardiovascular disease: The status of the U-shaped curve. British Medical Journal 303:565-8.

McGeary, Kerry Anne, and Michael T. French. 2000. Illicit drug use and emergency room utilization. Health Services Research 35:153-69.

Mullahy, John, and Jody L. Sindelar. 1991. Gender differences in labor market effects of alcoholism. American Economic Review 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.

Murphy, Kevin M., and Robert Topel. 1985. Estimation and inference in two-step econometric models. Journal of Business and Econometric Statistics 4:370-9.

Nelson, Charles R., and Richard Startz. 1990. The distribution of the instrumental variables estimator and its t-ratio when the instrument is a poor one. Journal of Business 58:S125-40.

Normand, Jacques, R. Lempert, and C. O'Brien (editors). 1994. Under the influence? Drugs and the American work force. Washington, DC: National Academy Press.

Norton, Edward C., Richard C. Lindrooth, and Susan T Ennett. 1998. Controlling for the endogeneity of peer substance use on adolescent alcohol and tobacco use. Health Economics 7:439-53.

Office of National Drug Control Policy (ONDCP). 1996. The national drug control strategy. Washington, DC: The White House.

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.

Register, Charles A., and Donald R. Williams. 1992. Labor market effects of marijuana and cocaine use among young males. Industrial and Labor Relations Review 45:435-48.

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

Schwartz, S. H., and S. Huismans. 1995. Value priorities and religiosity in four Western religions. Social Psychology Quarterly 58:88-107.

Shaper, A. G. 1990. Alcohol and mortality: A review of prospective studies. British Journal of Addiction 85:837-47.

Smith, R., and R. Blundell. 1986. An exogeneity test for a simultaneous equation Tobit model with an application to labor supply. Econometrica 54:679-85.

Staiger, Douglas, and James H. Stock. 1997. Instrumental variables regression with weak instruments. Econometrica 65:557-86.

Substance Abuse and Mental Health Services Administration (SAMHSA). 1999a. National household survey on drug abuse: Public release codebook 1994-B. Rockville, MD: U.S. Department of Health and Human Services.

Substance Abuse and Mental Health Services Administration (SAMHSA). 1999b. National household survey on drug abuse: Main findings 1994. DHHS Publication No. (SMA)95-3063. Rockville, MD: U.S. Department of Health and Human Services.

Turner, C., Judith Lessler, and J. Devore. 1992. Effects of mode of administration and wording on reporting of drug use. In Survey measurement of drug use: Methodological studies. Rockville, MD: National Institute on Drug Abuse, pp. 221-44.

Wu, De-Min. 1973. Alternative tests of independence between stochastic regressors and disturbances. Econometrica 41:733-50.

Zarkin, Gary A., Thomas A. Mroz, Jeremy W. Bray, and Michael T. French. 1998. The relationship between substance use and labor supply for young men. Labour Economics 5:385-409.

Table 1.

Variable Means, by Drug Using Status: Males [a]

 CDU NCDU NDU
Variable (n = 222) (n = 382) (n = 3305)

Age [**] 33.194 32.948 37.256
White 0.680 0.764 0.754
Black 0.279 0.212 0.195
Hispanic [**] 0.185 0.191 0.291
Married [**] 0.329 0.361 0.630
Highest grade completed [*] 12.428 12.882 12.678
Number of moves past year [**] 0.553 0.542 0.294
Number of people in household [**] 3.279 2.942 3.394
Excellent health [**] 0.243 0.344 0.385
Very good health 0.365 0.354 0.310
Good health 0.279 0.236 0.216
Fair health 0.095 0.050 0.067
Poor health 0.018 0.016 0.023
New England Census Division 0.023 0.016 0.027
Middle Atlantic Census Division 0.090 0.113 0.110
East North Central Census Division 0.095 0.102 0.091
West North Central Census Division 0.059 0.055 0.040
South Atlantic Census Division 0.162 0.105 0.159
East South Central Census Division 0.050 0.071 0.053
West South Central Census Division 0.095 0.068 0.131
Mountain Census Division 0.194 0.196 0.162
Pacific Census Division 0.234 0.275 0.229
Strongly religious [**] 0.195 0.197 0.292
Employed [**] 0.748 0.853 0.880
Labor force participation 0.869 0.935 0.923

 Total
Variable (n = 3909)

Age [**] 36.604
White 0.751
Black 0.201
Hispanic [**] 0.275
Married [**] 0.586
Highest grade completed [*] 12.684
Number of moves past year [**] 0.333
Number of people in household [**] 3.343
Excellent health [**] 0.373
Very good health 0.317
Good health 0.221
Fair health 0.067
Poor health 0.022
New England Census Division 0.026
Middle Atlantic Census Division 0.109
East North Central Census Division 0.092
West North Central Census Division 0.042
South Atlantic Census Division 0.153
East South Central Census Division 0.054
West South Central Census Division 0.123
Mountain Census Division 0.167
Pacific Census Division 0.234
Strongly religious [**] 0.277
Employed [**] 0.870
Labor force participation 0.921

Kruskal-Wallis rank test for statistically significant differences in
variable mean across the drug-using categories:

(*)= p [less than or equal to] 0.05;

(**)= p [less than or equal to] 0.01.

(a)CDU = chronic drug user; NCDU = nonchronic drug user, NDU = nondrug
user. Strongly religious is a dichotomous variable equal to one if the
respondent agreed or strongly agreed with each of the following
statements: "Religious beliefs are important to me," "Religious belief
influence my decisions," and "It is important that my friends share my
religious beliefs."
Table 2.

Variable Means, by Drug Using Status: Females [a]

 CDU NCDU NDU
Variable (n = 157) (n = 350) (n = 5205)

Age [**] 34.427 33.480 36.442
White [*] 0.592 0.666 0.704
Black [*] 0.363 0.297 0.248
Hispanic [**] 0.140 0.117 0.243
Married [**] 0.318 0.386 0.577
Highest grade completed [**] 12.318 12.857 12.643
Number of moves past year [**] 0.813 0.418 0.282
Number of people in household [**] 3.389 3.320 3.595
Excellent health [**] 0.172 0.255 0.338
Very good health 0.312 0.347 0.317
Good health 0.331 0.284 0.240
Fair health 0.121 0.100 0.082
Poor health 0.064 0.014 0.023
New England Census Division 0.013 0.009 0.034
Middle Atlantic Census Division 0.096 0.083 0.115
East North Central Census Division [*] 0.159 0.186 0.107
West North Central Census Division 0.025 0.029 0.044
South Atlantic Census Division 0.217 0.191 0.168
East South Central Census Division 0.057 0.057 0.065
West South Central Census Division 0.115 0.083 0.131
Mountain Census Division 0.115 0.140 0.140
Pacific Census Division 0.204 0.223 0.198
Strongly religious [**] 0.129 0.239 0.324
Employed [*] 0.567 0.706 0.688
Labor force participation 0.745 0.763 0.725

 Total
Variable (n = 5712)

Age [**] 36.205
White [*] 0.699
Black [*] 0.255
Hispanic [**] 0.233
Married [**] 0.558
Highest grade completed [**] 12.648
Number of moves past year [**] 0.305
Number of people in household [**] 3.572
Excellent health [**] 0.329
Very good health 0.318
Good health 0.245
Fair health 0.084
Poor health 0.024
New England Census Division 0.032
Middle Atlantic Census Division 0.113
East North Central Census Division [*] 0.113
West North Central Census Division 0.042
South Atlantic Census Division 0.170
East South Central Census Division 0.064
West South Central Census Division 0.127
Mountain Census Division 0.139
Pacific Census Division 0.199
Strongly religious [**] 0.314
Employed [*] 0.686
Labor force participation 0.728

Kruskal--Wallis rank test for statistically significant differences in
variable means across the drug-using categories:

(*)= p [less than or equal to] 0.05;

(**)= P [less than or equal to] 0.01.

(a)CDU = chronic drug user; NCDU = nonchronic drug user; NDU = nondrug
user. Strongly religious is a dichotomous variable equal to one if the
respondent agreed or strongly agreed with each of the following
statements; "Religious beliefs are importment to me," "Religious beliefs
influence my decisions," and "It is important that my friends share my
religious beliefs."
Table 3.

First-Stage Probit Results for Chronic Drug Use (CDU) [a]

Variable Males Females

Age 0.002 0.083 [*]
 (0.036) (0.039)
Age squared 0.0002 -0.001 [*]
 (0.0004) (0.001)
White 0.053 0.040
 (0.174) (0.190)
Black 0.234 0.094
 (0.188) (0.204)
Hispanic -0.318 [**] -0.379 [**]
 (0.101) (0.119)
Married -0.455 [**] -0.340 [**]
 (0.079) (0.084)
Highest grade completed -0.029 [*] -0.039 [*]
 (0.014) (0.016)
Number of moves past year 0.125 [**] 0.250 [**]
 (0.044) (0.039)
Number of people in household 0.040 -0.010
 (0.022) (0.026)
Excellent health -0.250 -0.686 [**]
 (0.257) (0.205)
Very good health 0.047 -0.462 [*]
 (0.255) (0.198)
Good health 0.113 -0.338
 (0.256) (0.196)
Fair health 0.193 -0.302
 (0.271) (0.215)
Strongly religious -0.191 [*] -0.584 [**]
 (0.088) (0.107)
Constant -0.821 -2.092 [**]
 (0.742) (0.797)

(a)Standard errors reported in parentheses. Coefficient estimates for
eight geographical controls (see Table 1 and 2) not reported. Strongly
religious is a dichotomous variable equal to one if the respondent
agreed or strongly agreed with each of the following statements:
"Religious beliefs are important to me," "Religious beliefs influence my
decisions," and "It is important that my friends share my religious
beliefs."

(*)Statistically significant, p [less than or equal to] 0.05;

(**)statistically significant, p [less than or equal to] 0.01.
Table 4

Estimation Results for Employed and Labour Force Participation: Males

 Employed
Variable Probit IV Estimator

Chronic drug use (CDU) -0.413 [**] -2.833 [*]
 (0.105) (1.377)
Marginal effect of CDU -0.089 [**] 0.495 [*]
 (0.027) (0.241)
Smith-Blundell test ([x.sup.2],
 d.f. = 1) [H.sub.0]: CDU exogenous -- 2.08
Age 0.102 [**] 0.096 [**]
 0.026) (0.027)
Age squared -0.002 [**] -0.001 [**]
 (0.0003) (0.0003)
White 0.312 [*] 0.340 [**]
 (0.125) (0.126)
Black 0.037 0.109
 (0.139) (0.145)
Hispanic 0.062 -0.033
 (0.080) (0.095)
Married 0.516 [**] 0.390 [**]
 (0.065) (0.097)
Highest grade completed 0.035 [**] 0.027 [*]
 (0.010) (0.011)
Number of moves past year -0.018 0.029
 (0.041) (0.050)
Number of people in household 0.005 0.014
 (0.018) (0.020)
Excellent health 1.733 [**] 1.667 [**]
 (0.166) (0.168)
Very good health 1.740 [**] 1.746 [**]
 (0.166) (0.167)

Good health 1.388 [**] 1.426 [**]
 (0.165) (0.166)
Fair health 0.786 [**] 0.842 [**]
 (0.176) (0.179)
Constant -2.784 [**] -2.320 [**]
 (0.562) (0.627)

 Labor Force Participation
Variable Probit

Chronic drug use (CDU) -0.304 [*]
 (0.129)
Marginal effect of CDU 0.037 [*]
 (0.019)
Smith-Blundell test ([x.sup.2],
 d.f. = 1) [H.sub.0]: CDU exogenous --
Age 0.186 [**]
 (0.032)
Age squared -0.003 [**]
 (0.0004)
White 0.368 [*]
 (0.147)
Black 0.183
 (0.164)
Hispanic 0.050
 (0.098)
Married 0.309 [**]
 (0.080)
Highest grade completed 0.013
 (0.013)
Number of moves past year 0.001
 (0.051)
Number of people in household 0.084 [**]
 (0.024)
Excellent health 2.171 [**]
 (0.174)
Very good health 2.166 [**]
 (0.175)

Good health 1.803 [**]
 (0.171)
Fair health 1.114 [**]
 (0.180)
Constant -4.257 [**]
 (0.676)

 Labor Force
 Participation
Variable IV Estimator

Chronic drug use (CDU) -0.184
 (1.672)
Marginal effect of CDU -0.018
 (0.165)
Smith-Blundell test ([x.sup.2],
 d.f. = 1) [H.sub.0]: CDU exogenous 0.46
Age 0.187 [**]
 (0.032)
Age squared -0.003 [**]
 (0.0004)
White 0.362 [*]
 (0.148)
Black 0.184
 (0.171)
Hispanic 0.056
 (0.117)
Married 0.317 [**]
 (0.117)
Highest grade completed 0.013
 (0.014)
Number of moves past year -0.004
 (0.061)
Number of people in household 0.082 [**]
 (0.026)
Excellent health 2.168 [**]
 (0.178)
Very good health 2.153 [**]
 (0.175)

Good health 1.792 [**]
 (0.173)
Fair health 1.108 [**]
 (0.185)
Constant -4.289 [**]
 (0.754)

Standard errors reported in parentheses. All estimates for marginal
effects were calculated using the mean values for the independent
variables. Coefficient estimates for eight geographical controls not
reported.

(*)Statistically significant, p [less than or equal to] 0.05;

(**)statistically significant, p [less than or equal to] 0.01.
Table 5

Estimation Results for Employed and Labor Force Participation: Females

 Employed
Variable Probit IV Estimator

Chronic drug use (CDU) -0.248 [*] -0.249
 (0.109) (0.821)
Marginal effect of CDU -0.091 [*] -0.087
 (0.042) (0.286)
Smith-Blundell test ([[chi].sup.2],
 d.f. = 1)
 [H.sub.0]: CDU exogenous -- 1.32
Age 0.101 [**] 0.101 [**]
 (0.018) (0.018)
Age squared -0.001 [**] -0.001 [**]
 (0.0002) (0.0002)
White 0.067 0.070
 (0.086) (0.087)
Black 0.057 0.064
 (0.095) (0.095)
Hispanic 0.050 0.057
 (0.052) (0.055)
Married -0.153 [**] -0.157 [**]
 (0.040) (0.044)
Highest grade completed 0.076 [**] 0.077 [**]
 (0.007) (0.007)
Number of moves past year -0.054 [*] -0.054
 (0.027) (0.037)
Number of people in household -0.102 [*] -0.102 [**]
 (0.012) (0.013)
Excellent health 1.404 [**] 1.392 [**]
 (0.133) (0.140)
Very good health 1.372 [**] 1.356 [**]
 (0.133) (0.137)
Good health 1.219 [**] l.205 [**]
 (0.133) (0.136)
Fair health 0.857 [**] 0.836 [**]
 (0.140) (0.143)
Constant -3.305 [**] -3.289 [**]
 (0.383) (0.384)

 Labor Force
 Participation
Variable Probit

Chronic drug use (CDU) 0.109
 (0.119)
Marginal effect of CDU 0.034
 (0.035)
Smith-Blundell test ([[chi].sup.2],
 d.f. = 1)
 [H.sub.0]: CDU exogenous --
Age 0.091 [**]
 (0.018)
Age squared -0.001 [**]
 (0.0002)
White 0.060
 (0.088)
Black 0.153
 (0.097)
Hispanic 0.005
 (0.053)
Married -0.283 [**]
 (0.042)
Highest grade completed 0.070 [**]
 (0.007)
Number of moves past year -0.030
 (0.029)
Number of people in household -0.105 [**]
 (0.013)
Excellent health 1.545 [**]
 (0.133)
Very good health 1.521 [**]
 (0.132)
Good health 1.387 [**]
 (0.132)
Fair health 0.961 [**]
 (0.138)
Constant -2.913 [**]
 (0.393)

 Labor Force
 Participation
Variable IV Estimator

Chronic drug use (CDU) 1.378
 (0.869)
Marginal effect of CDU 0.441
 (0.278)
Smith-Blundell test ([[chi].sup.2],
 d.f. = 1)
 [H.sub.0]: CDU exogenous 3.32
Age 0.085 [**]
 (0.019)
Age squared -0.001 [**]
 (0.0002)
White 0.055
 (0.089)
Black 0.148
 (0.098)
Hispanic 0.033
 (0.057)
Married -0.265 [**]
 (0.045)
Highest grade completed 0.073 [**]
 (0.007)
Number of moves past year -0.066
 (0.038)
Number of people in household 0.104 [**]
 (0.013)
Excellent health 1.600 [**]
 (0.141)
Very good health 1.557 [**]
 (0.138)
Good health 1.416 [**]
 (0.137)
Fair health 0.985 [**]
 (0.143)
Constant -2.923 [**]
 (0.395)

Standard errors reported in parentheses. All estimates for marginal
effects were calculated using the mean values for the independent
variables. Coefficient estimates for eight geographical controls not
reported.

(*)Statistically significant, p [less than or equal to] 0.05;

(**)statistically significant, p [less than or equal] to 0.01.
Table 6

Probit Estimation Results for Employed and Labor Force Participation:
Males and Females

 Males Females
 Labor Force
Variable Employed Participation Employed

Chronic drug use (CDU) -0.437 [***] -0.306 [*] -0.247 [*]
 (0.106) (0.130) (0.109)
Marginal effect of CDU -0.095 [**] -0.037 [*] -0.091 [*]
 (0.028) (0.019) (0.042)
Nonchronic drug use (NCDU) -0.138 -0.010 0.015
 (0.094) (0.119) (0.078)
Marginal effect of NCDU -0.026 -0.001 0.005
 (0.019) (0.012) (0.027)
Age 0.102 [**] 0.186 [**] 0.101 [**]
 (0.026) (0.032) (0.018)
Age squared -0.002 [**] -0.003 [**] -0.001 [**]
 (0.0003) (0.0004) (0.0002)
White 0.323 [**] 0.368 [*] 0.066
 (0.125) (0.147) (0.086)
Black 0.047 0.183 0.057
 (0.139) (0.164) (0.095)
Hispanic 0.051 0.049 0.051
 (0.080) (0.098) (0.052)
Married 0.508 [**] 0.308 [**] -0.152 [**]
 (0.065) (0.080) (0.040)
Highest grade completed 0.035 [**] 0.013 0.076 [**]
 (0.010) (0.013) (0.007)
Number of moves past year -0.014 0.001 -0.054 [*]
 (0.041) (0.052) (0.027)
Number of people in household 0.005 0.084 [**] -0.102 [**]
 (0.018) (0.024) (0.012)
Excellent health 1.730 [**] 2.171 [**] 1.404 [**]
 (0.165) (0.174) (0.133)
Very good health 1.741 [**] 2.167 [**] 1.372 [**]
 (0.166) (0.175) (0.133)
Good health 1.390 [**] 1.804 [**] 1.219 [**]
 (0.165) (0.171) (0.133)
Fair health 0.785 [**] 1.114 [**] 0.856 [**]
 (0.176) (0.180) (0.140)
Constant -2.748 [**] -4.255 [**] -3.307 [**]
 (0.563) (0.676) (0.383)

 Females
 Labor Force
Variable Participation

Chronic drug use (CDU) 0.114
 (0.119)
Marginal effect of CDU 0.035
 (0.035)
Nonchronic drug use (NCDU) 0.049
 (0.081)
Marginal effect of NCDU 0.016
 (0.025)
Age 0.091 [**]
 (0.018)
Age squared -0.001 [**]
 (0.0002)
White 0.058
 (0.088)
Black 0.153
 (0.097)
Hispanic 0.008
 (0.053)
Married -0.281 [**]
 (0.042)
Highest grade completed 0.071 [**]
 (0.007)
Number of moves past year -0.031
 (0.029)
Number of people in household -0.105 [**]
 (0.013)
Excellent health 1.545 [**]
 (0.133)
Very good health 1.521 [**]
 (0.132)
Good health 1.386 [**]
 (0.132)
Fair health 0.959 [**]
 (0.139)
Constant -2.917 [**]
 (0.393)

Standard errors reported in parentheses. All estimates for marginal
effects were calculated using the mean values for the independent
variables. Coefficient estimates for eitht geographical controls not
reported

(*)Statistically significant, p [less than or equal] 0.05;

(**)Statistically significant, p [less than or equal] 0.01.