Part of the reason for the growth in drug testing programs has been federal government initiatives and legislation, which has encouraged or mandated companies to implement drug-testing as a means to achieve drug-free workplaces and to improve productivity. The issue gained national attention in 1986 with President Reagan's Executive Order 12564, which required federal agencies to develop programs and policies to achieve drug free workplaces. The Drug Free Workplace Act was passed in 1986, which led to regulations by federal agencies requiring random testing of contract workers where there were concerns related to public safety or national security. According to surveys, drug testing by American companies has increased significantly from the mid-1980's to the present. For example, surveys of Fortune 500 companies have found that between 1985 and 1991, the percentage of companies conducting drug tests increased from 18 to 40 percent. Representative surveys conducted by the Bureau of Labor Statistics found close to a 50 percent increase in drug testing companies between 1988 and 1990 for work sites with more than 250 employees. (from 31.9% to 45.9 %). By 1992-93, national surveys indicated that 48 percent of work sites with 50 or more full time employees and 71 percent of work sites with 1000 or more employees conducted some type of drug tests. A 1994 survey of the American Management Association of their corporate members found a 300% increase in testing since 1987, with 87 percent of their members conducting some type of drug testing. Over half the members indicated that the decision to implement drug testing stemmed from federal government requirements. With the recent passage by the House of Representatives of the Drug Free Workplace Act of 1998, which provides incentives to small businesses to establish drug testing programs, it appears likely that the growth will continue.

The courts have provided some restrictions on the public sector regarding the implementation of test programs, generally finding that public sector employees cannot be tested without reasonable suspicion unless there is a compelling need to protect public safety. However, these restrictions do not apply to the private sector. Some states and local jurisdictions have passed laws restricting or regulating specific the types of drug testing, such as random testing of employees without reasonable cause. Recent surveys have also shown that drug testing varies according to several factors, with drug testing most widely used in transportation, mining and construction, and manufacturing. Larger firms, or firms in the South or Mid-West are more likely to test.

The economic theory providing the link between drug testing and productivity is not straightforward or unambiguous; there are reasonable arguments that can be constructed suggesting either positive and negative effects on productivity from drug testing. The arguments suggesting a positive effect are primarily as follows: drug testing reduces illicit drug use (by weeding out users or providing them with a strong incentive to stop) which, in turn, enhances productivity. Potentially positive effects could also result if highly productive workers or managers prefer to work at companies that conduct drug tests, believing it provides a safer, drug free work environment, with lower risk of accident, injury or interaction with other employees who use drugs. These companies may attract better workers, and the workers there may exhibit greater loyalty towards the company.

Implicit in the first argument suggesting a positive effect is the assumption that use of illicit drugs lowers productivity. However, there is no consensus among economists who have researched this area; in fact, some recent research suggests positive associations between drug use and productivity for at least some types of illicit drugs. In addition, drug testing does not necessarily capture impairment in the work place, and some drugs (e.g. marijuana) can be detected in the system for a long time after use. In addition, drug tests may not capture all illicit drug use because they are not 100 % reliable--false positives and false negatives, though believed to be rare, are both possible. Although the reliability of test results has improved with modern test procedures, lab error is still possible, and legal food products such as hemp seed oil and poppy seed bagels have been found to generate "false-positive" test results. Furthermore it is possible that workers who use illicit drugs may find strategies that allow them to pass a drug test, such as substitution or adulteration of urine used for one of the more common tests. It is therefore possible that drug testing will fail to achieve desired increase in productivity if 1) drug use does not lower productivity, or 2) the drug tests fail to accurately measure drug use in the workplace. In addition, according to the CDUW report, the preventive effects of drug testing on overall drug use in the work place has not been scientifically documented. Nevertheless, it is reasonable to assume that drug testing should serve to limit drug use in the work place by providing a disincentive to workers from engaging in illicit drug use, with potentially positive effects on productivity.

It is also possible that drug testing lowers productivity. There are several reasons why this could be the case. The first reason is that drug tests can be expensive and take time to administer. It is important to consider all of the economic costs associated with drug tests. First there are the transactions costs of implementing a drug test program and (in many cases) contracting with the company that will administer the drug tests. Second is the administrative costs associated with conducting the testing, including the explicit costs of each test and the opportunity costs of time taken by company employees to either administer or take the tests. Given the possibility of false-positive test results, it is recommended that companies that conduct drug tests also hire or contract for the services of a Medical Review Officer (MRO). Third are the costs of follow-up in the event of a negative test, which can range from firing the worker, to providing a second test (provided in some cases because of the possibility of a false-positive), to providing some form of treatment or discipline for the worker. If a worker is fired, (or not hired in the event of pre-employment tests), then the company will have additional costs of searching, hiring, or training a new worker. There may be additional costs if a grievance is filed. Because drug tests entail costs and take time away from other activities, it follows that they will either lower productivity or raise costs unless there are offsetting benefits. The administrative costs are probably small but the full economic costs of drug tests have not been comprehensively researched. The costs of the drug tests have been estimated to exceed one billion dollars per year, with over 20 million workers tested annually at a cost of approximately $50 per test. The full economic costs of drug testing are clearly larger, yet few microeconomic studies of the cost- effectiveness of drug testing programs have been conducted.

The second possible reason for a negative effect is that drug testing could undermine worker morale, motivation, loyalty, or effort towards the company. Some surveys have shown that workers have a negative attitude towards drug tests, particularly random tests, which are often viewed as unfair. For example, a survey of railroad workers found that only 16 percent of the workers believed that random testing was fair. It is not surprising that many unions and the American Civil Liberties Union have opposed drug testing for a variety of reasons, including: 1)they are inconsistent with the constitutional protection against unreasonable search and seizure, 2) they are intrusive and constitute an unnecessary invasion of privacy, 3) they do not capture impairment in the workplace but rather prior use that may have occurred outside of the workplace, or 4) they do not measure impairment from alcohol, which may be the biggest contributor to productivity losses in workplace from drugs. If drug tests contribute to a negative view towards the company, then workers may not contribute as much in return, or they may seek employment elsewhere; some workers may not seek or accept jobs from companies with drug testing programs.

A third reason why drug testing may result in lower productivity is if workers who use illicit drugs are either more productive than workers who do not use illicit drugs, or more productive than they would be if they didn't use drugs. It is generally believed that drug use lowers productivity, but the research in this area is inconclusive. Dreher (1982) applied a case study approach to analyze Jamaican farming and concluded that marijuana use raised productivity. A study by Register and Williams,(1992) which controlled for the endogeneity of drug use, found that " the net effect for all marijuana users...was positive". Kaestner (1994) used the 1984 and 1988 National Longitudinal Survey of Youth to develop both cross-sectional and longitudinal (fixed effects) estimates of the effects of illicit drug use on wages, which is considered a good proxy for productivity. He was able to estimate effects separately for both men and women from cocaine as well as marijuana use. The cross sectional estimates showed positive and significant effects of both illicit drugs for both groups; and the longitudinal estimates, which controlled for unobserved heterogeneity in the sample, found positive effects for cocaine use for women. In no cases with either the cross-sectional or longitudinal estimates were coefficients representing effects from drug use found to be negative and significant. Finally, the review of the studies conducted by the CDUW found that "low to moderate use of any illicit drug or alcohol is either positively associated with productivity or simply not related" ; negative effects are found only with heavy or problem users.

At a minimum, these studies suggest the possibility that some drugs may even enhance productivity in at least some contexts. Furthermore, recent studies by health research scientists suggest that some workers may be using some illicit drugs for medical purposes. For example, Grinspoon (1997), or Zimmer and Morgan (1997) argue that marijuana can be an effective medicine for individuals suffering from pain, cancer, AIDs, multiple sclerosis, glaucoma, arthritis, migraines, or even depression, among other possible ailments. Access to medical marijuana for some patients, or rescheduling marijuana from a schedule 1 to a schedule 2 controlled substance, (which allows doctors prescriptions) has been endorsed by several major medical organizations, including the New England Journal of Medicine, the Florida Medical Association, the American Public Health Association, and the American Academy of Family Physicians. If drug tests require workers with a variety of conditions to give up effective medical treatments, there could be adverse health consequences, with negative effects on productivity.

A fourth reason why drug tests may result in lower productivity is if workers (rather than give up drug use altogether because of the drug tests) substitute other drugs that are more harmful to performance in the workplace. For example, most of the positive test results are for marijuana, which can be detected up to one month after use. Yet, according to many experts, marijuana use outside of the workplace will not adversely effect performance at work, because any intoxicating effects or impairments of reasoning or motor skills are short-lived. Because of the drug tests, workers may switch to "harder drugs", like heroin, cocaine, or amphetamines, which do not remain in the system as long. Or they might switch to alcohol, or drugs that are not tested for, which could have more significant adverse effects on performance and health. Some evidence of substitution effects have been found by other researchers.

In summary, theory and prior evidence suggests that positive or negative effects on productivity are possible. The issue should ultimately be resolved on the basis of scientific evidence--the findings from carefully constructed statistical models based on some underlying theory, and detailed case studies. To our knowledge, no one has applied an economic production function model using firm-level data to measure or test for effects from either drug use or drug testing. Since workers are not likely to reveal their illicit drug use, it is not possible to apply a production function model to directly measure effects on productivity using microeconomic firm-level data. However, data on drug tests by individual companies are now becoming available which allows for application of such a model to investigate productivity effects from drug tests. The goal of this paper is to develop such a model and then apply it to a test industry. The next section presents the Cobb-Douglas production model that is used for these purposes, followed by statistical estimates of the effect of drug testing on productivity using data from the computer and communications equipment industries.

The Cobb-Douglas (CD) production function is the most common form used in applied studies because it is simple to estimate and is consistent with the economic theory of production. It is commonly used in empirical studies to analyze effects of varying workplace characteristics on productivity (for example, unionization, profit sharing, flexible work schedules). The mathematical derivation of the estimating equation is presented in the appendix to this paper. Applying the CD model, it is possible to estimate the effect of drug testing programs on productivity.

The estimating equation used in this study represents the intensive form of the Cobb-Douglas model; a measure of output per worker is used as the dependent variable (representing average productivity) in a modified regression equation. The independent variables include the capital-labor ratio, the level of labor, and a dummy variable for whether the firm has a drug testing program. Econometric methods commonly applied in other production function studies are used to estimate the production parameters. Control variables for capital quality, rates of capacity utilization, and other possible variables are readily incorporated into the model.

To estimate the model, data on drug testing from a sample of companies in several related 3-digit SIC code industries comprising the computer and communications equipment industries were obtained. The data on drug policies used for this study was collected at an internet site where employees reported their employer’s drug policy. The accuracy of the data was checked by 1) comparing with other sites with comparable data, 2) checking the internet site of the individual companies and 3) telephoning companies not previously verified. Some companies refused to provide information on drug testing programs, however, no discrepancies were found in the policies that were reported. The results of our check suggested that no significant biases were present with the employee-reported data. The data were then merged with financial data from a sample of the same companies obtained through COMPUSTAT, which provides standardized information on variables needed for production function estimation (from company annual reports, 10-K reports and other financial documents). A final data set with 63 organizations, all from SIC (standard industrial classifications) codes of either 357 (Computer and Office Equipment) or 737 (Computer and Data Processing Services) was developed. The drug policy for each organization was assumed to have been intact between 1994 and 1996. (Our COMPUSTAT data covers the years 1994 to 1996 --56 companies have three years of data and 6 have two years of data).

The dependent variable uses the log of net sales divided by number of employees, as a proxy for productivity. The measure for capital is the log of gross plant, property and equipment divided by the number of employees. The log of employees is used to measure labor's input. To measure potential differences between industries in their production functions, a variable was coded one if the company is in SIC 737 and zero if in SIC 357 and was used in interaction with the labor and capital variables. The drug testing variables are equal to one if the testing is used and zero otherwise. Capital quality is the net plant, property and equipment divided by the gross plant property and equipment. Finally, a time trend variable is produced to control for variations of production over time.

The basic production function estimated is a Cobb-Douglas with an additional interaction term for the SIC classification, a separate intercept variable for the SIC classification, a measure of capital quality, a time trend variable and a variable(s) for drug testing. Two types of drug testing variables are used, the first is coded one for any type of drug testing, zero otherwise, and the second categorizes them into two groups: 1) pre-employment screening testing and 2) random testing of current employees and pre-employment screening. Ordinary least squares regression was used instead of a fixed effect or random effects for these estimates. This is due to both the lack of variance in the drug testing variables over the time period and also the short time period.

The results presented in Table I show both estimated regression results. Both regressions exhibit problems with heteroscedasticity and the standard errors were corrected following White (1978). The results of the Cobb-Douglas model for both regressions find only a weakly significant difference for the labor input variable. However, there is a highly significant difference in the effect of capital per employee on productivity with a greater effect for the computers and data processing industry.

The industry variable for SIC 737 is also significant but there is no significant time effect. The estimates for the industries suggest constant returns to scale for SIC 357 and increasing returns to scale for SIC 737. The first column of results contains the variable representing any type of drug testing and it is surprisingly negative and significant. The magnitude implies that a change from not drug testing to using drug testing would reduce productivity by 19 percent. Similarly, the regression estimates in column two also suggest a large and significant decline in productivity with pre-testing use associated with 16 percent drop and random testing with 29 percent. Possible explanations for the magnitude and the direction of the estimates are explored below. Although the random test variable suggest a greater difference in productivity effects than the pre-test variable, a test that the two coefficients are equal could not be rejected.

Overall the results suggest that drug testing has served to lower rather than enhance productivity. The signs of the relevant coefficients are both negative and significant. One surprise is the large magnitude of the significant results, because they suggest that drug testing results in about a 20 percent lower level of productivity. This negative effect may appear unbelievably large, but there are several possible explanations which need to be investigated as part of future research. The first is that the non-representativeness of the sample may be biasing the results. Nevertheless, as is often the case, we were forced to deal with the data that were available subject to project resources, and there were no obvious biases inherent in the sample. The second possible reason is that the estimate of the mean effect is rather imprecise, given the relatively small sample size. The 95 % confidence interval ranges from around a negative 3 percent to a negative 31 percent, and it is possible that the true effect is closer to the smaller end of the scale. Further research on additional samples will be required to identify the true effect with greater precision.

The third possible reason is that there are omitted variables which are correlated with drug testing that are associated with companies of lower productivity. One possibility is that companies with low levels of productivity are more likely to adopt productivity enhancing programs, such as drug testing, in the hopes of improving performance. Another is that companies with inferior management are more likely to adopt drug testing. It is possible that companies that relate to employees positively with a high degree of trust are able to obtain more effort and loyalty in return. Drug testing, particularly without probable cause, seems to imply a lack of trust, and presumably could backfire if it leads to negative perceptions about the company. A good approach for assessing this hypothesis would be to apply a fixed effects model to control for unmeasured characteristics. This approach is planned as part of future research when a larger sample of longitudinal (before and after) data become available.

At the very least, the results contained in this paper cast serious doubt about claims that drug testing can significantly boost productivity. Considerable uncertainty remains concerning the economic effects of drug testing, and our evidence suggests that negative effects on productivity are possible. Despite the lack of strong scientific evidence that it is effective, drug testing has become an accepted industry practice, and the federal government continues to encourage companies in the private sector to develop drug testing programs. In recent years, the frequency of "test-positive" test results has fallen significantly, making it even less likely that drug testing programs are cost effective. Further research will be required to see if the surprising results contained in this paper hold up with other samples or in other industries.

The discussion has also highlighted possible ways in which drug testing might adversely affect productivity. If drug testing creates a negative work environment, or causes substitutions of more dangerous drugs or alcohol, then worker effort or employee selection may be diminished. Overall, productivity could be adversely affected even if there are some positive outcomes such as reduced absenteeism. Drug testing may generate economic benefits at some work sites, however, there may be more efficient, less costly, and less intrusive ways for companies to identify workers who are impaired on the job. Drug tests do not measure impairment, and employees have reported ingenious ways to get around or beat the drug tests. Companies and test laboratories must then refine the test methods in response. Eventually, more perfect test and verification methods might be developed that greatly reduce chances of "false-positive" or "false-negative" test results. But there is no evidence that productivity would be enhanced as a result, or that more widespread drug testing would be cost-effective.

The Cobb-Douglas estimating equation is derived in the context of a factor augmentation model of production, with inputs redefined in terms of efficiency units, and factor effort functions assumed to be of the following form:

The productivity of each factor may be influenced through interactions with labor, and so effort functions are specified for capital as well as labor. The parameters a_l and a_k are the effort parameters for labor and capital, respectively. These represent the labor and capital channels whereby drug testing programs are potentially effective. This is a convenient form for estimation because it reduces to the standard model with E(F)=1 if D=0 or if a_j=0. The potential effect of drug testing on productivity is not restricted a priori to be positive, neutral, or negative; the properties of the effort functions are sufficiently general to allow for each of these possibilities. Factor augmentation is incorporated into the model without imposing it, and the effect of drug testing on productivity depends on the sign and magnitude of the effort parameters as well as the production coefficients.

The estimating equation is based on a two-factor model where firms employ labor (L) and capital (K), to produce output (Q). With CD technology, the production function can be written as: 2) Q = A[(1+a_l)^D× L]^α[(1+a_k)^D× K]^βe^u

The intensive form represents a direct productivity measure and is convenient for empirical estimation. Dividing by L, taking logs, and making use of an approximation, the following estimating equation is obtained:

3) ln(Q/L) = lnA + (αa_l + βa_k)D + βln(K/L) + (Θ-1)lnL + u

The error term represents random deviations from technical efficiency by the individual firm, and Θ is a measure of returns to scale (i.e. α + β). Output per worker depends upon the capital/labor ratio, scale of output, and the drug testing variable D. Estimates of (αa_l + βa_k) provide measures of the elasticity of output per worker with respect to changes in the drug testing variable. Control variables for union status, capital quality, or other factors are readily incorporated into the equation, following approaches commonly used in other applied studies.

References

Ackerman, D. (1991) "A History of Drug Testing," in R.H. Coombs and L.J. West, eds., Drug Testing: Issues and Options, New York, Oxford University Press.

Akerloff, George A. and Janet Yellen, eds. 1986 Efficiency Wage Models of the Labor Market, Cambridge, England, Cambridge University Press.

Arthur and Doverspike (1997) "Employment Related Drug Testing: Idiosyncratic Characteristics and Issues", Public Personnel Management, Vol. 26, No. 1 (Spring), pp. 77-87.

Barnum,D. and Gleason, J (1994) "The credibility of drug tests; a multi-stage Bayesian analysis." Industrial and Labor Relations Review, Vol. 47, no. 4 (July) pp.610-621.

Brown, Charles and James L. Medoff. 1978 "Trade Unions in the Production Process." Journal of Political Economy, Vol. 84 no. 3, (June), pp. 355-378.

Bureau of Labor Statistics, 1989 Survey of Employer Anti-drug Programs, Report 760 (January)

Clark, Kim B. 1980 "The Impact of Unionization on Productivity: A Case Study." Industrial and Labor Relations Review, Vol 33 no. 4, (July).

Conte and Svejnar, 1988 "Productivity Effects of Worker Participation in Management, Profit Sharing, Worker Ownership of Assets, and Unionization in U.S. Firms." International Journal of Industrial Organization, Vol 6, March 1988, 139-151

Dreher (1982) "Working Men and Ganja: Marijuana Use in Rural Jamaica." Philadelphia, Institute for the Study of Human Institutions.

Fines, Reeves, and Harney (1996) "Employee Drug Testing: Are cities complying with the courts." Public Administration Review, vol. 56, no. 1, (Jan-Feb), pp. 30-37

Griliches, Z. 1967 "Production Functions in Manufacturing, Some Preliminary Results," The Theory and Empirical Analysis of Production; Studies in Income and Wealth, ed. Murray Brown. New York: National Bureau of Economic Research.

Grinspoon, L.,1997 Marijuana; The Forbidden Medicine Yale University Press,

Hanson (1990) "What Employees Say about Drug Testing" Personnel, (July), pp. 32-36.

Hartwell, Steele, French, and Rodman, 1996, "Prevalence of drug testing in the workplace" Monthly Labor Review, (November), pp. 35-42.

Hausman, Jerry .A. 1978 "Specification Test in Econometrics." Econometrica, Vol. 46 (November), pp. 1251-1272.

Hayghe, H. 1991 "Anti-drug programs in the workplace-Are they here to stay", Monthly Labor Review, (April), pp. 26-29.

Kaestner, R. (1994) "New Estimates of the Effect of Marijuana and Cocaine Use on Wages" Industrial and Labor Relations Review, vol. 47, no. 3, (April)

Kruse, Douglas. 1992 "Profit Sharing and Productivity, Microeconomic Evidence", Economic Journal 102 (January): 24-36.

Normand, Lempert, and O'Brien, eds., Under the Influence:Drugs and the American Work Force, National Research Council and Institute of Medicine, National Academy Press, 1994.

Register and Williams, 1992 "Labor Market Effects of Marijuana and Cocaine Use Among Young Men.", Industrial and Labor Relations Review, Vol. 45, No. 3, (April), pp. 106-123.

 

Scientific American, 1990 "Testing Negative, A look at the evidence justifying illicit drug testing" (March)

Shepard, Edward. 1994 "Profit Sharing and Productivity: Further Evidence from the Chemicals Industry", Industrial Relations (Oct): .

White, H. 1978. "A Heteroskedasticity Consistent Covariance Matrix and a Direct Test for Heteroskedasticity." Econometrica, Vol. 46 pp. 817-838.

Zellner, A., J. Kmenta and J. Dreze. 1966 "Specification and Estimation of Cobb-Douglas Production Function Models." Econometrica, Vol.34 no.4, (October), pp. 79-92.

Zimmer and Morgan, 1997 Marijuana Myths, Marijuana Facts: A Review of the Scientific Evidence, Lindesmith Center.