Despite substantial national efforts, drug abuse remains a serious public health problem for a sizable proportion of the population. Since data presented in Chapter 3 suggest that a sizable portion of the work force uses drugs, reducing use by the active work force would have an impact on drug use overall, reducing the pool of illicit drug users in the United States and moving us closer to the societal goal of eliminating drug abuse. The workplace is thus an obvious site for user-focused interventions.

This societal perspective is seldom used to justify programs to reduce or eliminate drug use by the work force, and there may be constitutional problems with workplace drug-testing programs aimed predominantly at this goal. Interventions aimed at securing a drug-free workplace are justified instead largely on safety and productivity grounds. The data obtained in worker population studies, however, do not provide clear evidence of the deleterious effects of drugs other than alcohol on safety and other job performance indicators. This does not mean there are no deleterious effects; it may reflect the paucity of relevant data and the quality of the research done to date.

The extent to which impaired worker performance due to drug use can affect safety and productivity in the workplace is not well understood, although

a substantial amount of laboratory research has been carried out evaluating the effects of single doses of various abusable drugs on cognitive and psychomotor performance. The results of such research cannot be extrapolated directly to the workplace because the effects of drugs on workplace performance are a complex function of the interaction between the dynamic workplace environment and the multiplicity of other variables impinging on the worker. For example, the job performance of a worker who has slept little the night before, is anxious about a family member's problem, has not eaten breakfast, must work with a dangerous piece of equipment, and has continually changing job demands is likely to be affected differently by a prior night's use of marijuana or cocaine than a well-rested worker performing a routine task. The challenge of modeling such complex interactions and simplifying the issues so that they can be studied in the laboratory is obvious and may never fully be met. Yet laboratory research can provide a base from which to start understanding such problems. Even if it cannot capture the full richness of the occupational world, it can help us understand how the drugs people take interact with different kinds of ongoing behavior; this is knowledge we must have in order to design and implement effective intervention programs.

A second goal of laboratory research on drug effects is to develop reliable measures of the acute impairment associated with drug use. To date, the most commonly used method for identifying drug use by the work force relies on urinalysis to detect the presence of drugs or their metabolites. Such testing does not address the issue of drug-induced impairment. Although there are data relating dose of alcohol to level of impairment, there are no data relating the level of other drugs (or their metabolites) obtained from urinalysis to levels of impairment. Laboratory-developed measures of impairment that lead to the development of a reliable and easily administered performance battery for the detection of workplace performance impairment could be an enormous improvement over the current technologies (discussed in Chapter 6).

A myriad of laboratory performance studies have been carried out to test the effects, under controlled conditions, of such drugs as stimulants, marijuana, sedatives, benzodiazepines, and alcohol (see Table 4.1).¹ However,

even apart from the complex interaction effects mentioned above, these studies have numerous shortcomings as guides to understanding the effects of alcohol and other drug use by the work force. Although the doses studied are sometimes (but not always) the same as those being used by drug users in the work force, patterning of drug use comparable to that of many drug users (i.e., multiple doses, periodically repeated doses, etc.) has not been adequately addressed. Moreover, with few exceptions, no attempt has been made to model the specific task used to measure impairment after specific workplace performances, and multiple variations on similar tasks make generalization across studies difficult.

To further complicate the picture, there has been little effort to model the subject population in laboratory studies after the work force population. The most frequently used research subject is a college student, paid to participate in a research project, or expected to participate in order to fulfill a course requirement. In addition, unlike the worker who is experienced in the task being performed, the subjects in most drug use studies are frequently performing the tasks on which impairment is measured for the first time or after only a brief period of training. Behavioral histories are seldom taken into account in laboratory research. Other common weaknesses of experimental design include inattention to doses used, time points for measurements, and contingencies in maintaining behavior. Despite these problems, however, a few generalizations can be drawn about the likely effects of different classes of drugs on performance.

Stimulant drugs (e.g., caffeine, amphetamine, cocaine) increase general activity, lead to reports of positive subjective effects, and are often used clinically to reduce food intake (Fischman, 1987). Despite users' reports of substantial performance enhancement after stimulant use, this effect has not been systematically replicated in the laboratory (Johanson and Fischman, 1989). When improvement in performance has occurred, the margin of improvement has either been less than 10 percent, or stimulants prevented or reversed a decrement in performance due to fatigue or boredom. Of course in some situations like athletic competitions, a minor improvement in performance could have large positive effects for the performer (Laties and Weiss, 1981), and, when otherwise unavoidable fatigue or boredom are fought off, decrements in performance may be forestalled. In general, however, it is important to point out that significant performance enhancement is not apparent; much of what users report are the subjective effects of stimulants (e.g., increased levels of energy, friendliness), which lead to a

belief that behavior is improved without any actual improvement (Fischman, 1987).

The use of marijuana and products containing δ⁹-tetrahydrocannabinol (THC) has a long history, and the literature on the effects of these substances on performance is voluminous. Concentrated efforts to delineate marijuana-related effects on behavior have yielded variable results, with the most consistent effects being decrements in time estimation and divided attention tasks (e.g., Jones and Stone, 1970; Marks and MacAvoy, 1989). Marijuana interfered with performance on a variety of other tasks on approximately 50 percent of the occasions it was studied (e.g., arithmetic, Chesher et al., 1977; tracking, Barnett et al., 1985), suggesting that experimental conditions play a substantial role in determining the effects of this substance. Although there is some evidence that marijuana can affect performance for several hours after it is used (e.g., Miller and Cornett, 1978), there are almost no data on what behaviors are impaired, for how long, and to what extent.

There has been a general belief that smoking marijuana can lead to a cluster of signs and symptoms often referred to as an amotivational syndrome. If it does, repeated rather than occasional use of marijuana could have severe implications for behavior and productivity in the workplace. The motivational effects of marijuana have provided a focus for research over the past several decades, with variable results. In general, well-controlled epidemiological studies of marijuana use have failed to confirm the existence of such a syndrome (e.g., Comitas, 1976; Stefanis et al., 1977; Page, 1983), and laboratory research suggests that environmental conditions can influence the amotivational effects of marijuana, determining its presence or absence (Foltin et al., 1989, 1990).

The majority of studies evaluating acute effects of alcohol administration have found that single doses cause decrements in a variety of performance tasks, particularly tracking, visual vigilance, divided attention, postural stability, and cancellation tasks, with less robust effects on memory tasks. Since the problems that alcohol use poses for transportation safety are well recognized, it has received substantial attention from the transportation research community. As with other laboratory studies, the magnitude of impairment in transportation-related tasks has been shown to be dependent on the nature of the task, research subject characteristics (e.g., skill level, tolerance) and environmental factors (e.g., fatigue). Overlearned tasks

(e.g., coordination, balance) are relatively resistant to alcohol consumption (Burns, 1992), while divided attention, information processing, and attention processes are highly susceptible to alcohol-induced impairment (Streufert et al., 1992). Performance on these latter tasks are impaired at low blood alcohol levels, implying that relatively small amounts of alcohol can have detrimental consequences for both traffic safety as well as other workplace safety-sensitive positions. Although there is a relationship between blood alcohol level and decrements in performance, there is considerable variability in the alcohol level at which decrements occur. In addition, there is variability in the amount of alcohol required to reach a given blood alcohol level, even when body weight is controlled (O'Neil et al., 1983). This source of variability is largely related to variations in metabolic rate. Furthermore, although the data are not as clear for all the benzodiazepines, data with prototypic benzodiazepines (diazepam, lorazepam, and triazolam) suggest that, as with alcohol, these drugs produce decrements on a full range of performance tasks, from gross motor tasks such as postural stability (Evans et al., 1990) to complex tasks such as divided attention (Erwin et al., 1986).

Although residual effects can refer to any effects that occur a number of hours after major drug effects have dissipated, this has come to mean next-day effects or hangover effects. The issue here is whether substances used at home on one day affect job performance the next day. These effects can either be manifested as prolonged drug effects, similar to the initial drug effect, or can differ from the initial drug effect. This latter change in behavior is best characterized by what is commonly called hangover. Thus, alcohol consumed during the evening can produce intoxication, slurred speech, etc. Six or eight hours later, after some sleep and no further alcohol intake, a different set of symptoms (e.g., headache, irritability, inability to concentrate, etc.) might be apparent. Such hangover effects can be disruptive in the workplace, reducing productivity and perhaps interfering with safe and accurate performance and/or social interactions. In addition, it is possible that hangover or drug withdrawal effects are contributing factors in the maintenance of drug-seeking behavior.

In addressing the issue of residual drug effects, we have to differentiate between chronic, regular, daily drug use and acute, or occasional, drug use. Repeated drug use can result in tolerance to some of the effects of the drug. When tolerance develops, it takes a larger dose of the drug to achieve the same effect. The development of tolerance does not by itself affect workplace performance, although it can moderate what would otherwise be the effects of drug-taking behavior or allow greater consumption of a drug than would otherwise occur. The aspect of chronic drug use that can affect

workplace performance adversely is the development of dependence. Physical dependence is manifested as a syndrome of effects that appear upon abrupt cessation of a drug after chronic use and can be alleviated by intake of that drug. The most widely described drug dependence is probably for the opiates, as seen with chronic heroin use. Comparable dependence is seen with many if not all classes of psychotropic drugs.

The data on the development of dependence to alcohol, sedatives, and opiates are clear. Repeated and regular intake of these substances has been shown to result in physical dependence, manifested by a replicable withdrawal syndrome that can be alleviated by the administration of the substance that the individual has been taking. The data for marijuana are less clear. A number of laboratory studies have been carried out in which research subjects were given marijuana cigarettes to smoke, or δ⁹-THC to consume, repeatedly for 10-30 days. In general, both tolerance to many of marijuana's effects and dependence are seen (Jones and Benowitz, 1976; Mendelson et al., 1976). Withdrawal is manifested as irritability, restlessness, decreased appetite, tremor, etc., and has been described (Jones, 1978) as a clinical picture similar to that seen after withdrawal of the sedative-hypnotics. It is possible that the maintenance of stable THC blood levels is important for the development of dependence, and that cessation of use could result in a withdrawal syndrome with workplace consequences.

Residual effects of occasional marijuana use appear slight if they exist at all. Some researchers searching for hangover effects recount subjective reports of feeling "spacey" or "stoned" or "hung over'' the next day (Cousens and DiMascio, 1973). The few objective measures that purport to show decrements attributable to the consumption of marijuana a day earlier are suggestive at best (Yesavage et al., 1985; Leirer et al., 1991). Thus, we cannot at this time conclude that the occasional use of marijuana will have measurable next-day residual effects, nor can we conclude that some subtle effects are not present.

Accuracy in identifying small amounts of cannabis metabolites in urine is excellent. This means that even occasional use of marijuana is often picked up in urine screens taken in relation to an accident or other workplace problems. We cannot, however, on the basis of the available data, assign particular behavioral consequences to the presence of these metabolites in the urine. Thus, when post-accident drug screening reveals that a responsible person tested positive for marijuana, it does not mean that marijuana use played a causal role in the incident.

Alcohol hangover effects can apparently degrade performance. They have been reported to impair drivers' and pilots' performance (Laurell and Tornros, 1983; Yesavage and Leirer, 1986), although the extent of the impairment was in part related to both age and experience with the task. For example, performance of older pilots was more impaired than that of younger

pilots, but the older pilots were more aware of their impairment for up to 4 hours after reaching peak blood alcohol levels.

There are no data on the residual effects of occasional stimulant use except for fatigue related to secondary sleep deprivation. When stimulants are used repeatedly in binges, a "crash," marked by irritability, hypersomnolence, and some depression can occur (Fischman, 1987). This constellation of next-day effects, however, has not been linked to specific performance changes, and it may be that the effects do not differ from decrements measured after sleep deprivation in the absence of drug use.

Where laboratory conditions are different from the conditions that characterize actual drug use, drug users, and job performance, there is a question of how far one can generalize from laboratory results to predict the actual implications of drug use that are of interest (Berkowitz and Donnerstein, 1982; Dipboye and Flanahan, 1979; Locke, 1986; Sears, 1986; Sackett and Larson, 1991). This is the external validity problem. Not all differences between the laboratory and the outside world pose serious threats to external validity. This depends on whether there is reason to believe the differences are consequential for the generalizations one would like to make. Unfortunately, in assessing the research done to date on the performance implications of drug use, many of the differences between the currently available laboratory studies and drug use outside the laboratory appear large and potentially important. Often, however, these differences, or at least the size of these differences, are not inherent in the laboratory methodology. Thus identifying important differences not only highlights the limitations of extant research, but it also suggests ways in which future studies can be improved.

The failure to examine combinations of drugs constitutes a major gap in the research on drugs' effects on performance. It is becoming increasingly rare to find a single-drug-class abuser (or even drug user). Polydrug use is generally the norm, and this is particularly the case for alcohol. Many substances, such as marijuana, nicotine, and sedatives, are frequently taken in combination with alcohol, and the effects of these combinations are generally unknown. It has recently been reported that the combined intake of cocaine and alcohol results in formation of a metabolite, cocaethylene, which has a half-life of more than 2 hours, and is considerably more toxic than either drug alone (Hearn et al., 1992; Perez-Reyes and Jeffcoat, 1992). Although not yet studied, it is possible that this active metabolite could result in behavioral changes long after measurable cocaine or alcohol levels are present in the body. Other combinations of drugs may have effects that

are different from or longer lasting than the effects of the drugs taken singly.

Research on the performance implications of drug use must also consider carefully the experimental population. There are several research objectives to be met in determining an appropriate research subject population. Initial research studies evaluating a specific substance or a specific task should be carried out in a single well-studied population in which multiple variables can be controlled, motivational variables can be either manipulated or controlled, and basic mechanisms can be addressed. This approach exemplifies much of the research summarized above. Subjects in these studies were generally male college or graduate students. Unfortunately, few studies move on to the next phase of research in which issues of generalizability and predictability are addressed. It is difficult to know how generalizable data from the population of male college or graduate students, tested individually and in isolation, are to the general work force. On one hand, students, who are younger than much of the work force, have commonly learned to perform under less than optimal conditions, often taking tests under conditions of sleep deprivation and substantial stress. On the other hand, student subjects may seldom or never have used the drugs administered, and unlike the occasional or regular user, they may not have learned how to function productively under the influence of the drug. In addition, laboratory performance tests are often novel, with minimal opportunity for practice prior to testing. Members of the work force are generally performing well-learned tasks in familiar, more social environments, with familiar cues designed to enhance performance as the working environment undergoes minor modifications (e.g., illumination changes, personnel changes).

Much of the research being carried out on the effects of psychoactive substances on performance has not been designed with the "at risk" population in mind. For example, at least 20 percent of truck drivers are said to drive under the influence of marijuana, methamphetamine, or cocaine (Beilock, 1988). Yet little research on the effects of these drugs has the specific situations of truck drivers in mind. Prescribed medications (e.g., benzodiazepines) may well impair performance, yet few studies evaluating them in the populations most likely to need them have been carried out. Epidemiologic studies are necessary to help define the populations at risk, the substances most generally used, and the environments in which they are most likely to be used. Data from such statistics can then feed into laboratory research.

As we have already noted, few performance studies model drug use outside the laboratory. With many drugs, it is rare for a user to take a single dose each day. Stimulants such as cocaine, for example, are taken repeatedly, in binges, for several hours or days at a time (Johanson and Fischman, 1989). Under these conditions, it is likely that tolerance will develop to some of the effects of the drug, although dosing can escalate to

substantial levels. Thus, the single doses often administered in research settings do not accurately reflect, in either pattern or number, those taken by habitual or even occasional users.

Many of the studies evaluating the effects of single doses of drugs on laboratory performance employ contingencies for correct or efficient performance. Points exchangeable for money, for example, are awarded when tasks are completed according to instructions. In the workplace there may be eventual contingencies (e.g., firing for those whose performance is habitually substandard), but there is rarely a performance monitor providing performance feedback on a minute-by-minute basis. This difference also undercuts the generalizability of the data from laboratory to workplace.

Perhaps the most important effect missed in most laboratory performance studies is the interaction of the drug taken with the behavior of the user and others in the environment (i.e., laboratory studies do not involve a social environment). Although drugs have pharmacological effects that at high doses can be substantial, the effects of using a drug often depend on what else is going on in the environment and the feedback given to the individual performing under the effect of the drug. For this reason, studies that emulate workplace conditions can more accurately assess drug effects than those that do not. In an interesting series of experiments, Kelly and colleagues have examined the behavioral profiles associated with using marijuana (Kelly et al., 1990), amphetamines (Kelly et al., 1991), and diazepam (Kelly et al., 1993). Healthy volunteers resided continuously in a laboratory designed for the long-term unobtrusive measurement of human behavior (Brady et al., 1974). The laboratory day was designed to emulate a normal day, with subjects working in their private rooms from about 9:00 a.m. to 5:00 p.m., followed by a social activities period when subjects had access to other subjects and a common social/exercise space. Data were collected simultaneously on a wide range of behaviors under naturalistic living conditions: performance, social behavior, food intake, cigarette smoking, and subjective effects. By comparing drug effects across multiple dimensions of human behavior, it was possible to ascertain a behavioral profile of each drug's action. In addition, this design addresses the risk factors discussed in the introduction of this section and narrows the gap between laboratory studies and the workplace. By using a social setting, distracting events, extensive training with weeks of practice, and a sample of nonstudents, this study comes closer to simulating the workplace and living conditions associated with ordinary drug use than the studies we have discussed thus far.

Kelly and his colleagues found that smoking marijuana cigarettes decreased accuracy on a digit symbol substitution task (DSST), increased food intake, and decreased verbal interaction and tobacco cigarette smoking (Kelly et al., 1990). When comparing the relative potency of marijuana across

these measures, it was clear that DSST performance, food intake, and social behavior were all altered at doses that had no effects on task performance other than the DSST. This suggests that THC doses that affect performance are equal to, if not greater, than doses that affect social and eating behavior.

Oral amphetamine administration decreased food intake, improved accuracy on some work tasks, and increased verbal interaction, cigarette smoking, and verbal ratings of drug effect (Kelly et al., 1991). The relative potency comparisons, in this case, suggested that the doses that were affecting performance also had significant effects on other dimensions of human behavior. Oral diazepam administration, in contrast, increased verbal interaction at low doses and decreased verbal interaction at high doses, disrupted only one measure of task performance, increased food intake at one or both doses, and increased verbal ratings of drug effect (Kelly et al., 1993). The relative potency comparisons in this experiment suggested that diazepam at clinical doses disrupted dimensions of human performance, other than task performance.

The Kelly et al. experiments clearly indicate the utility of multiple measures in studying the performance effects of drugs. The unique aspect of these experiments is that measures of normal social and eating behavior were obtained, providing a source of potency comparisons involving normal nonlaboratory-dependent behavior. The findings suggest that, at least for diazepam and marijuana, experiments that measure only task performance may miss the effects that particular drug doses have on other behaviors. Where significant performance effects are found, other significant changes in behavior are also likely. The use of multiple dependent measures in performance experiments will make it easier to put into the context the changes in performance observed in laboratory experiments.

Measures of the subjective effects of test drugs can also provide essential information about the effectiveness of the dose range being studied. For example, in an experiment on the effects of buspirone on performance, Critchfield and Griffiths (1991) did not observe any performance effects of a high buspirone dose (four to five times the therapeutic dose) in sedative users, but did observe significant ratings of "bad drug effect." Without the self-report data, there would have been no verification that an effective dose range was being tested. Visual analog scales, often used in the assessment of momentary changes in affect (Folstein and Luria, 1973), require little effort, represent minimal intrusion on existing protocols, and can be incorporated readily into most laboratory procedures.

There are no studies that provide direct estimates of the effects of drug use on job performance or on behavior in organizations. As is often the

case with research in the social and behavioral sciences and in medicine, ethical constraints make it impossible to conduct definitive controlled studies of the long-term effects of drug use at work. Rather, it is necessary to infer the impact of drug use at work from a variety of studies conducted in the laboratory and the field. Laboratory studies provide evidence regarding the effects of controlled, short-term exposure to specific drugs on the performance of specific tasks. Field studies provide evidence regarding the links between drug use (either self-reported or detected through other means) and a number of work behaviors, but they lack the controls needed to allow researchers to isolate specific drug effects.

Difficulties in generalizing from behavioral research are by no means unique to research on the impact of drugs in the workplace; these same issues emerge in virtually any area of research that involves human behavior. It is therefore important to keep an appropriate perspective in discussing the methodological limitations of research on drugs and work. Nevertheless, it is important to note at the outset that these difficulties are an important reason why the existing research base does not demonstrate conclusively that the effects of drug use on the work force are either large or small. The challenge is to overcome them.

One way to appraise the generalizability of any specific set of laboratory findings is through a risk factors model, in which each of the potentially important differences between the laboratory and the work setting is treated as a factor that is likely to limit the generalizability of laboratory research. The more risk factors that are present, the greater the likelihood that the effects of the drug examined will be different in the lab than in the field. Table 4.2 lists several risk factors, some of which have already been

mentioned, that commonly threaten the external validity of laboratory studies relating drug usage to task performance.

It is common in the workplace to distinguish between tests of maximal performance and tests of typical performance. In laboratory settings, in which subjects are under scrutiny, people are often motivated to perform well and are able to perform with few distractions. Thus, they are likely to demonstrate maximal performance. In work settings, in which people are not under the same level of scrutiny, their typical performance is not likely to be as high or as consistent. The typical-maximal performance distinction is particularly important for studying the effects of drugs that affect attention, fatigue, and vigilance. Subjects may make special efforts to overcome the effects of these drugs in laboratory settings, which they would not make in work settings. The incorporation of distractions (e.g., multiple attention tasks) in laboratory studies is one way to address this issue.

Laboratory studies often present unfamiliar or novel tasks that require constant attention and monitoring. In contrast, many work tasks are overlearned, and individuals may perform those tasks in an automatic processing mode after extensive training and practice. Also, as already pointed out, laboratory studies often employ novice subjects (e.g., college students), whereas in the workplace most workers are relatively experienced with the tasks that need to be performed. Laboratory studies should include overlearned (i.e., highly practiced) tasks as well as novel tasks.

Laboratory studies often require subjects to perform in isolation, whereas job performance is usually carried out in a social setting. The effects of the presence of other workers on performance are themselves complex (e.g., social facilitation effects can enhance performance, whereas distraction can detract from performance); when jobs involve social interactions and interdependencies, the generalizability of studies in which subjects work alone can be especially hazardous. The addition of studies carried out in social settings could address this issue.

Laboratory studies often involve a single, relatively simple task (e.g., a reaction time task), whereas job performance often involves multiple, complex tasks. Furthermore, the evaluation of performance in the laboratory is often considerably simpler than in the field. That is, the standards that define good performance are often clear and well understood in the laboratory, but the same is not always the case in work settings. What is defined as poor performance in a work setting may not be the result of impairment, but rather the effect of disagreement over the definition of adequate performance.

Laboratory studies often employ convenience samples (e.g., college students, military pilots), which tend to be homogeneous. This may cause those studies to underestimate the importance of individual differences, as well as differences in training and experience when generalizing to the

workplace. Similarly, the exposure of subjects to drugs in laboratory experiments is usually carefully controlled (e.g., fixed doses of specific drugs taken in specific settings), whereas in the workplace there may be extensive variation in patterns and levels of drug use, including different configurations of polydrug use. Studies using experienced drug users with variations in drug use history as subjects may be useful in this regard; polydrug configurations must be tested.

Finally, laboratory studies are usually limited in scope, and subjects often know the approximate time span of the study. It may be easier to maintain maximal performance levels, or to adopt strategies to mitigate the effects of drugs on performance, in tasks whose duration is known (and is short) than it is over the course of a typical workday. Experimental sessions extending over the course of a workday, or at least for several hours, are more likely to capture drug effects than those of shorter duration.

As in all applications of risk factor approaches, a simple count of the factors that are present or absent in a given study does not necessarily determine its external validity. What is crucial is the plausibility of the threats posed by specific factors in the context of the given research. Indeed, it is possible that a study might exhibit all the potential barriers to generalizability shown in Table 4.2 and still produce generalizable results (e.g., effect of cyanide on performance using animals as subjects). The risk factor approach does provide a rough and probabilistic statement about generalizability; the more factors that are present, the less likely that laboratory results will replicate exactly in the field. It is particularly useful for evaluating the existing body of research on drug effects as a whole; the prevalence of studies that share almost all listed risk factors is disheartening. We reiterate, however, that this is not inherent in the laboratory methodology, as some researchers have shown.

Perhaps the most obvious finding of this survey of the literature on performance effects is the lack of consistent and significant effects. While sufficient amounts of almost any substance will have a deleterious effect on work force behavior, this is not the issue. The question is whether alcohol and other drugs of abuse, taken in the doses and patterns that people are using, either occasionally or regularly have detrimental effects on behavior, particularly workplace performance The literature suggests that, in single doses and under laboratory conditions, stimulant drugs (caffeine, cocaine, methylphenidate, amphetamine and nicotine) either have no effect or they moderately improve performance. Smoking marijuana seems to have variable effects, with inconsistent decrements on performance. Sedative drugs (alcohol, benzodiazepines, barbiturates) generally disrupt performance. There

are large differences across experiments and drugs in terms of experimental design, and the proportion of experiments reporting significant drug effects varies widely across tasks and drugs. The extent to which this variation is due to task methodology rather than dose needs investigation. Information gained from such research will also help identify factors that can moderate, as well as accentuate, drug-induced effects on performance.

Given the variability in tasks and procedures, the fact that, at certain doses, drugs affected performance in 40 to 80 percent of the experiments reviewed by the committee indicates that this field, though experimentally mature, requires additional research rooted in more careful attention to methodology. Although an impressive array of tasks has been used in the research to date, experimental protocols have rarely been similar across studies. Suggestions for future protocols include use of control groups, training subjects prior to participation, testing of more than one dose, testing for drug combinations, and testing of performance before and several times after drug administration. A greater range of subject populations beyond the student population should also be sampled. Task standardization, including instructions to subjects, duration of performance, feedback to subjects, motivational conditions, and details of presentation should be maximized across studies at least in some conditions, to provide a common metric for comparison.

The relevance of tasks performed in the laboratory to tasks performed in the workplace is often not clear, and even when there is a plausible link, it has seldom been demonstrated. Clearly it is not possible to model every type of performance by the work force. And yet it should be possible to differentiate several functional categories of tasks to be studied. For example, acquisition of new behavior (i.e., learning) and stable performance of practiced tasks might be differentiated and both aspects modeled in testing. It would be useful for future research designs in which new tasks are being developed to include one or two standardized benchmark tasks for comparison of drug effects across studies. This would provide the systematic replication necessary to generalize across studies. Reliability across tasks could then be evaluated, and importantly, predictive validity could be assessed.

The relationship between drug doses that affect performance and doses that are used clinically, or self-administered for nonmedical purposes, determines the effects of drugs on work force performance. The greater the extent of the overlap between these dose ranges, the greater the potential hazards to both the drug user and society. Therefore, it is important that research in this area incorporate usage information collected in epidemiologic or other laboratory studies so that we can optimize the utility of the data being collected in performance testing for predicting decrements in workplace performance. Future research, using more standardized protocols and

tasks with broad subject populations, should address the varied issues relating to public safety and the creation of work and home environments that minimize potential adverse drug effects.

In sum, laboratory research has the potential for providing valuable information about the relationship between alcohol and other drug use, social behavior, and job performance. To date, despite some research designs that are abstractly elegant, that potential has hardly been tapped. Recognizing the important contribution laboratory research can make to improving our understanding of the impact of individual alcohol and other drug use on work performance, and after having carefully reviewed the current scientific knowledge base, the committee offers the following conclusions and recommendations.

• Laboratory studies of the effects of alcohol and other drugs on behavior have shown inconsistent results. These differences may be due, in part, to differences in the populations tested, the measurements used, and the range of drug doses administered.

Recommendation: Benchmark measures should be included in laboratory studies to permit generalization across studies. Funding agencies should consider holding conferences to establish such benchmarks.

Laboratory studies show small performance-enhancing effects of commonly used doses of cocaine and other stimulants. Commonly used doses of marijuana produce variable decrements in performance. Alcohol and prescribed sedatives produce decreases in performance depending on the dose, time of consumption, and the time-course of circulating concentrations of the drug's active metabolites, relative to the work schedule. All drug effects are influenced by dose and prior experience. The age of individuals and the presence of other drugs may also mediate the influence of particular drugs.

• The use of alcohol and other drugs away from the work site, including prescription drugs and over-the-counter medication, may have detrimental effects during work, especially for those in safety-sensitive positions. Thus, a long-acting drug taken the night before work or alcohol taken at lunch away from the job may have on-the-job effects like those of drugs taken at the work site. In addition, cessation of drug use may produce either withdrawal or hangover effects that affect work site performance. To date there has been little research directed toward any of these issues.

Recommendation: Researchers and funding agencies should devote more attention to the ways in which prescription and over-the-counter medications affect job performance, especially for safety-sensitive positions.

Recommendation: Studies of work site alcohol and other drug use should encompass off-site use that may have on-the-job effects. Hangover and withdrawal effects should also be considered in assessing the workplace implications of alcohol and other drug use.

Alford, C., J.Z. Bhatti, S. Curran, G. McKay, and I. Hindmarch 1991 Pharmacodynamic effects of buspirone and clobazam. British Journal of Clinical Pharmacology 32:91-97.

Barnett, G., V. Licko, and T. Thompson 1985 Behavioral pharmacokinetics of marijuana. Psychopharmacology 85:51-56.

Beilock, R. 1988 RCCC Motor Carrier Safety Survey. Regular Common Carrier Conference, Alexandria, Va.

Berkowitz, L., and E. Donnerstein 1982 External validity is more than skin deep. American Psychologist 37:245-257.

Brady, J.V., G. Bigelow, H. Emurian, and D.M. Williams 1974 Design of a programmed environment for the experimental analysis of social behavior. Pp. 187-208 in D.H. Carlson, ed., Man-Environment Interactions: Evaluations and Applications. 7: Social Ecology. Milwaukee, Wis.: Environmental Design Research Associates.

Burns, M. 1992 Alcohol and drug effects on performance. Pp. 49-59 in Alcohol and Other Drugs in Transportation: Research Needs for the Next Decade. Transportation Research Board, National Research Council. Washington, D.C.: National Academy Press.

Chesher, G.B., H.M. Franks, D.M. Jackson, G.A. Starmer, and R.K.C. Teo 1977 Ethanol and δ⁹-tetrahydrocannabinol: interactive effects on human perceptual, cognitive and motor functions. Medical Journal of Australia 1:478-481.

Collins, W.E. 1980 Performance effects of alcohol intoxication and hangover at ground level and at simulated altitude. Aviation, Space, and Environmental Medicine 51:327-335.

Comitas, L. 1976 Cannabis and work in Jamaica: a refutation of the amotivational syndrome. Annals of the New York Academy of Sciences 282:24-32.

Cousens, K., and A. DiMascio 1973 (-)A9THC as an hypnotic: an experimental study of three dose levels. Psychopharmacologia 33:355-364.

Critchfield, T.S., and R.R. Griffiths 1991 Behavioral and subjective effects of buspirone and lorazepam in sedative abusers: supra-therapeutic doses. In L. Harris, ed., Problems of Drug Dependence. NIDA Research Monograph 105. Rockville, Md.: National Institute on Drug Abuse.

Curran, H.V., and M. Lader 1987 Differential amnesic properties of benzodiazepines: a dose-response comparison of two drugs with similar elimination half-lives. Psychopharmacology 92:358-364.

Dipboye, R.L., and M.F. Flanahan 1979 Research settings in industrial and organizational psychology: are findings in the field more generalizable than in the laboratory? American Psychologist 34:141-150.

Erwin, C.W., M. Linnoila, J. Hartwell, A. Erwin, and S. Gutherie 1986 Effects of buspirone and diazepam, alone, and in combination with alcohol, on skilled performance and evoked potentials. Journal of Clinical Psychopharmacology 6:199-209.

Evans, M.A., R. Martz, B.E. Rodda, L. Lemberger, and R.B. Forney 1976 Effects of marihuana-dextroamphetamine combination. Clinical Pharmacology and Therapeutics 20:350-358.

Evans, S.M., F.R. Funderburk, and R.R. Griffiths 1990 Zolpidem and triazolam in humans: behavioral and subjective effects and abuse liability. Journal of Pharmacology and Experimental Therapeutics 255:1246-1255.

Fischman, M.W. 1987 Cocaine and the amphetamines. Pp. 1543-1543 in H.Y. Meltzer, ed., Psychopharmacology: A Third Generation of Progress. New York: Raven Press.

Folstein, M.F., and R. Luria 1973 Reliability, validity, and clinical application of the visual analogue mood scale. Psychosomatic Medicine 3:479-486.

Foltin, R.W., and S.M. Evans 1992 Performance Effects of Drugs of Abuse: A Methodological Survey. Manuscript prepared for the Committee on Drug Use in the Workplace, Commission on Behavioral and Social Sciences and Education, National Research Council, Washington, D.C. 1993 Performance effects of drugs of abuse: a methodological survey. Journal of Human Psychopharmacology 8:9-19.

Foltin, R.W., M.W. Fischman, J.V. Brady, T.H. Kelly, D.J. Bernstein, and M.N. Nellis 1989 Motivational effects of smoked marijuana: behavioral contingencies and high-probability recreational activities. Pharmacology, Biochemistry, and Behavior 34:871-877.

Foltin, R.W., M.W. Fischman, J.V. Brady, D.J. Bernstein, R.M. Capriotti, M.N. Nellis, and T.H. Kelly 1990 Motivational effects of smoked marijuana: behavioral contingencies and low-probability activities. Journal of the Experimental Analysis of Behavior 53:5-19.

Foltin, R.W., M.W. Fischman, P. Pippen, and T.H. Kelly 1993 Behavioral effects of ethanol and marijuana, alone and in combination with cocaine in humans. Drug and Alcohol Dependence 32:93-106.

Hearn, W.L., S. Rose, J. Wagner, A. Ciarleglio, and D.C. Mash 1992 Cocaethylene is more potent than cocaine in mediating lethality. Pharmacology, Biochemistry and Behavior 39:531-533.

Heishman, S.J., and M.L. Stitzer 1989 Effect of d-amphetamine, secobarbital, and marijuana on choice behavior: social versus nonsocial options. Psychopharmacology 99:156-162.

Heishman, S.J., M.L. Stitzer, and J.E. Yingling 1989 Effects of tetrahydrocannabinol content on marijuana smoking behavior, subjective reports, and performance. Pharmacology, Biochemistry, and Behavior 34:173-179.

Hindmarch, I., J.S. Kerr, and N. Sherwood 1990 Effects of nicotine gum on psychomotor performance in smokers and non-smokers. Psychopharmacology 100:535-541.

Hooker, W.D., and R.T. Jones 1987 Increased susceptibility to memory intrusions and the Stroop interference effect during acute marijuana intoxication. Psychopharmacology 91:20-24.

Johanson, C.E., and M.W. Fischman 1989 The pharmacology of cocaine related to its abuse. Pharmacological Reviews 41:3-52.

Jones, R.T. 1978 Marijuana: human effects. Pp. 378-412 in L.L. Iversen et al., eds., Handbook of Psychopharmacology, Vol. 12. New York: Plenum Press.

Jones, R.T., andN. Benowitz 1976 The 30-day trip—clinical studies of cannabis tolerance and dependence. Pp. 627-642 in M.C. Braude and S. Szara, eds., The Pharmacology of Marijuana. New York: Raven Press.

Jones, R.T., and G.C. Stone 1970 Psychological studies of marijuana and alcohol in man. Psychopharmacologia 18:108-117.

Kelly, T.H., R.W. Foltin, C.S. Emurian, and M.W. Fischman 1990 Multidimensional behavioral effects of marijuana. Progress in Neuro-Psychopharmacology 14:885-902.

Kelly, T.H., M.W. Fischman, R.W. Foltin, and J.V. Brady 1991 Response patterns and cardiovascular effects during response sequence acquisition by humans. Journal of the Experimental Analysis of Behavior 56:557-574.

Kelly, T.H., R.W. Foltin, L. King, and M.W. Fischman 1993 The effects of repeated diazepam exposure on multiple measures of human behavior. Biological Psychiatry. (In press.)

Klorman, R., L.O. Bauer, H.W. Coons, J.L. Lewis, L.J. Peloquin, R.A. Perlmutter, R.M. Ryan, L.F. Salzman, and J. Strauss 1984 Enhancing effects of methylphenidate on normal young adults' cognitive processes. Psychopharmacological Bulletin 20:3-9.

Lane, J.D., and R.B. Williams, Jr. 1985 Caffeine affects cardiovascular response to stress. Psychophysiology 22:648-655.

Laties, V.G., and B. Weiss 1981 The amphetamine margin in sports. Federation Proceedings 40:2689-2692.

Laurell, H., and J. Tornros 1983 Investigation of alcoholic hang-over effects on driving performance. Blutalkohol 20:489-499.

Leirer, V.O., J.A. Yesavage, and D.G. Morrow 1991 Marijuana carry-over effects on aircraft pilot performance. Aviation, Space, and Environmental Medicine 62:221-227.

Linnoila, M., J.M. Stapleton, R. Lister, H. Moss, E. Lane, A. Granger, D.J. Greenblatt, and M.J. Eckardt 1990 Effects of adinazolam and diazepam, alone and in combination with ethanol, on psychomotor and cognitive performance on autonomic nervous system reactivity in healthy volunteers. European Journal of Clinical Pharmacology 38:371-377.

Locke, E.A. 1986 Generalizing from Laboratory to Field Settings. Lexington, Mass.: Heath.

Marks, D.F., and M.G. MacAvoy 1989 Divided attention performance in cannabis users and non-users following alcohol and cannabis separately and in combination. Psychopharmacology 99:397-401.

Mattila, M., T. Seppala, and M.J. Mattila 1986 Combined effects of buspirone and diazepam on objective and subjective tests of performance in healthy volunteers. Clinical Pharmacology and Therapeutics 40:620-626.

Mendelson, J.H., J.C. Kuehnle, I. Greenberg, and N.K. Mello 1976 Operant acquisition of marijuana in man. Journal of Pharmacology and Experimental Therapeutics 198:42-53.

Miller, L.L., and T.L. Cornett 1978 Marijuana: dose effects on pulse rate, subjective estimates of intoxication, free recall and recognition memory. Pharmacology, Biochemistry and Behavior 9:573-577.

O'Neil, B., A. Williams, and K. Dubowski 1983 Variability in blood alcohol concentrations. Journal of Studies in Alcohol 44:222-230.

Page, J.B. 1983 The amotivational syndrome hypothesis and the Costa Rica study: relationship between methods and results. Journal of Psychoactive Drugs 15:261-267.

Patat, A., M.J. Klein, and M. Hucher 1987 Effects of single oral doses of clobazam, diazepam and lorazepam on performance tasks and memory. European Journal of Clinical Pharmacology 32:461-466.

Patat, A., M.J. Klein, M. Hucher, and J. Grainer 1991 Study of the potential reversal of trizolam memory and cognitive deficits by RU 41 656 in healthy subjects. Psychopharmacology 104:75-80.

Perez-Reyes, M., and A.R. Jeffcoat 1992 Ethanol-cocaine interaction: cocaine and cocaethylene plasma concentrations and their relationship to subjective and cardiovascular effects. Life Sciences 51:553-563.

Peterson, J.B., J. Rothfleisch, P.D. Zelazo, and R.O. Pihl 1990 Acute alcohol intoxication and cognitive functioning. Journal of Studies on Alcohol 51:114-122.

Pihl, R.O., and H. Sigal 1978 Motivational levels and the marijuana high. Journal of Abnormal Psychology 87:280-285.

Preston, K.L., J.J. Guarino, W.T. Kirk, and R.R. Griffiths 1989 Evaluation of the abuse potential of methocarbamol. Journal of Pharmacology and Experimental Therapeutics 248:1146-1157.

Sackett, P.R., and J.R. Larson 1991 Research strategies and tactics in industrial and organizational psychology. In M. Dunnette and L. Hough, eds., Handbook of Industrial and Organizational Psychology, 2nd ed., Vol. 1. Palo Alto, Calif.: Consulting Psychologists Press.

Sears, D.O. 1986 College sophomores in the laboratory: influence of a narrow data base on social psychology's view of human nature. Journal of Personality and Social Psychology 51:515-530.

Stefanis, C., R. Dornbush, and M. Fink 1977 Hashish: Studies of Long-Term Use. New York: Raven Press.

Streufert, S., R.M. Pogash, J. Roach, D. Gingrich, R. Landis, W. Severs, L. Lonardi, and A. Kantner 1992 Effects of alcohol on risk taking, strategy, and error rate in visuomotor performance. Journal of Applied Psychology 77:515-524.

Strömberg, C., T. Seppälä, and M.J. Mattila 1988 Acute effects of maprotiline, doxepin and zimeldine with alcohol in healthy volunteers. Archives of International Pharmacodynamics 291:217-228.

Taberner, P.V., C.J.C. Roberts, E. Shrosbree, C.J. Pycock, and L. English 1983 An investigation into the interaction between ethanol at low doses and the benzodiazepines nitrazepam and temazepam on psychomotor performance in normal subjects. Psychopharmacology 81:321-326.

Wilson, J.R., V.G. Erwin, G.E. McClearn, R. Plomin, R.C. Johnson, F.M. Ahern, and R.E. Cole 1984 Effects of ethanol: II. Behavioral sensitivity and acute behavioral tolerance. Alcoholism: Clinical and Experimental Research 366-374.

Yesavage, J.A., and V.O. Leirer 1986 Hangover effects on aircraft pilots 14 hours after alcohol ingestion: a preliminary report . American Journal of Psychiatry 1546-1550.

Yesavage, J.A., V.O. Leirer, M. Denari, and L.E. Hollister 1985 Carry-over effects of marijuana intoxication on aircraft pilot performance: a preliminary report. American Journal of Psychiatry 142:1325-1329.

Use of alcohol or illicit drugs by employees at work or even away from work has long been associated with harmful consequences. In fact, much of the impetus behind the movement for a drug-free workplace has been due to widely publicized accidents that have been tied to the use of drugs. Marijuana use was implicated in the U.S.S. Nimitz accident in 1981, in which a Navy pilot crashed into the aircraft carrier's deck and destroyed several planes, resulting in damage estimates in excess of $100 million. Marijuana use was also implicated in several fatal train accidents, most notably the Conrail-Amtrak collision in 1987, in which 16 people died. Alcohol use is seen as an important cause of the Exxon Valdez catastrophe, one of the worst and most publicized oil spills in U.S. history.

These isolated catastrophic incidents are, however, of little scientific value in assessing the magnitude of the effects of alcohol and other drugs on job-related outcome measures. Studies that have attempted to assess such effects with worker populations have been observational in nature, since, as noted in the previous chapter, ethical constraints prevent definitive studies of the effects of drug use at work (e.g., it would be unacceptable to design a longitudinal double-blind study in which a large number of workers were randomly assigned to various drug use conditions to allow researchers to observe the impact of drug use). But observational filed studies, like controlled laboratory studies, have potentially serious limitations as a means of obtaining true estimates of the effects of alcohol and other drugs on job-related outcomes.

One problem with field studies is that their measures of alcohol and other drug use are often flawed. Self-reports are frequently suspect, although there are conditions under which they can be relatively trustworthy (Rouse et al., 1985; Reinisch et al., 1991; Freier et al., 1991). Bioassays (e.g., urine tests) indicate the recent use of drugs, but they provide little information about the frequency of use or the impairment (if any) associated with drug use.

A second problem is conceptual. The phrase alcohol and other drug use refers to a broad array of patterns of use of different drugs or combinations of drugs. Thus the findings of different studies are often not strictly comparable, and it is difficult to make sensible generalizations about the effects of alcohol and other drug use without numerous qualifications. Finally, while many field studies indicate a statistical relationship between indicators of alcohol and other drug use and work behaviors (e.g., absenteeism), most do not provide a basis for determining the causal role (if any) of alcohol and other drug use in determining these behaviors. Reported correlations between use and job performance could be due to the effects of some third, unmeasured variable rather than to the direct effects of alcohol and other drugs themselves. Job stresses, for example, might lead people both to take alcohol and other drugs and to absent themselves from work. If so, eliminating alcohol and other drug use without decreasing job stress might actually increase absenteeism, because one coping mechanism would be gone. Alternatively, alcohol and other drug use, absenteeism, and carelessness on the job might all result from a rejection of conventional social standards, in which case, attitude change rather then the coerced abstinence from alcohol and other drugs might be the key to better job performance.

Researchers have found that measures of alcohol and other drug use (both bioassay and self-report measures) are related to a number of problematic work behaviors, notably absenteeism, and perhaps accident rates and performance levels. It is plausible to assume that these work problems result from alcohol-and other drug-induced impairment but, as we have just pointed out, this is not the only possibility, and it may not be the most plausible one. Even if alcohol and other drug use has little or no direct effect on these workplace behaviors, it is still possible to find substantial correlations between such use and absenteeism, poor performance, and the like because these behaviors are all part of a general pattern or syndrome of behavior often referred to as deviance. Knowing whether alcohol and other

drug use makes an independent contribution to undesirable behaviors in the workplace is an important issue, since it bears substantially on how best to treat use by the work force. If alcohol and other drug use is strongly associated with other types of deviant attitudes and behaviors, confronting only the inappropriate use may not correct a worker's problems. In what follows, we first briefly review research on the construct of deviance; we then present a simplified model of the role deviance might play in explaining the relationship of reported associations between alcohol and other drug use and workplace behaviors; finally, we discuss the empirical evidence relating the effects of alcohol and other drugs to job-related outcomes.

Alcohol and other drug use and abuse do not occur as isolated events, but rather seem to be components of a cluster of behaviors and attitudes that form a syndrome or lifestyle referred to as problem behavior or deviance. Problem behavior theory, developed by Jessor and Jessor (1977, 1978), sees alcohol and other drug use as only one aspect of a deviance-prone lifestyle, particularly in adolescents and young adults (who represent the majority of those likely to use illicit drugs in or in connection with the workplace). For adolescents, aspects of the deviance-prone lifestyle include alcohol abuse, illicit drug use, precocious sexual involvement, frequent sexual activity, academic problems, deviant attitudes, and delinquent behavior. This theory has been tested in several studies, which, consistent with the theory, have shown high correlations among these deviant behaviors (Donovan and Jessor, 1985; Gillmore et al., 1991; Newcomb and Bentler, 1988).

Newcomb (1988) found that the use of alcohol and other drugs in the workplace was most highly correlated with low law abidance and selling drugs, followed by thefts and confrontational acts, and then sexual involvement. Alcohol use on the job (which is typically considered less deviant than other drug use) was less tightly bound to other types of deviance. Deviance variables accounted for 6 percent of the variance in self-reported alcohol use on the job, whereas they accounted for 20 percent of the variance in cocaine use at work and for 36 percent of the variance in the use of other hard drugs at work.

Several of the workplace correlates of alcohol and other drug use appear to be part of the general pattern of behavior associated with deviance, including work avoidance, abuse of benefits, irresponsibility, and low workplace rule abidance. Although some of these behaviors may be directly caused by alcohol and other drug use, many of them appear to be aspects of general deviance rather than the consequences of specific alcohol-or other druginduced impairment.

The research reviewed above suggests that deviance is a potentially critical variable in explaining why measures of problem drinking and illicit drug use are related to workplace behaviors such as absenteeism and poor performance. If they are, the simplified model depicted in Figure 5.1 may assist in understanding the complex processes involved in the deviance in alcohol and other drug use/workplace behavior relationship.

This figure includes four critical links. First, general deviance (A) is related to and is probably a causal factor in alcohol and other drug use. In fact, such use is often used as a marker variable or a measure of deviance. Although a reciprocal link is possible (i.e., alcohol and other drug use may increase deviance, which leads to greater exposure to alcohol and other drugs), research on deviance suggests that use is originally and primarily an aspect of or a result of a generally deviant pattern of behavior rather than a cause of that behavior. The magnitude of the reciprocal causal relation probably increases with the individual level of alcohol and other drug involvement.

Second, alcohol and other drug use can lead to impairment (B), which may depend on the dosage and purity of the drug, the time and setting in which drugs were consumed, the individual's prior experience with the drug, the nature of the task, and a host of other factors. Laboratory research suggests that various drugs can either enhance, impair, or have no effect on performance (see Chapter 4). Similarly, the direct effects of alcohol and other drugs on other workplace behaviors (e.g., absenteeism) may be highly uncertain or variable. Nevertheless, it is possible that some of the observed alcohol and other drug use/work behavior correlations are due to the direct effects of these drugs on judgment, coordination, vigilance, overall health, and other variables.

Third, impairment may lead to changes in work behavior (C), although the strength of this effect will depend on the degree of impairment, the subject's experience with performing under impairment, the nature of the task, and the like. Finally, deviance may directly affect behaviors (D) such as absenteeism, performance, counterproductive behavior, etc., independent of any of the pharmacological effects of the drugs taken. For example,

FIGURE 5.1 Relating deviance, drug use, and work behavior.

individuals who engage in deviant behavior patterns may be absent or careless, even on days when no drug or drug withdrawal effects are present.

There is research evidence to justify all four of the linkages shown in Figure 5.1. It is difficult, however, to make well-supported statements about the relative strength or importance of the various links, since, among other things, they may vary in both magnitude and direction, depending on the characteristics of an individual's alcohol and other drug use involvement. However, it is useful to note that the links between use, impairment, and work behavior appear to be more context-dependent than the links between deviance, use, and work behavior. That is, the probable effects of drugs on impairment and work behavior appear to depend on a large number of variables (e.g., dosage, experience, nature of the task), whereas the links between deviance, alcohol and other drug use, and work behavior appear to be more resilient to context, as they would if use and undesirable work behavior are simple manifestations of a broader underlying deviance. This suggests that deviance may be a better explanation than impairment of the links between alcohol and other drug use and undesirable work behavior. If so, confronting deviant behaviors and attitudes may be a more effective strategy than narrow antidrug programs for both preventing workplace decrements and treating poorly performing workers.

Because there are no studies that directly examine the impacts of alcohol and other drug use on work force performance, we must examine the cumulative evidence from laboratory and field studies, each of which examines some facets of the possible connections between alcohol and other drugs and work, to make inferences about the overall impact of their use. For example, to the extent that laboratory studies demonstrate alcohol-and other drug-induced impairment on tasks that are essentially similar to those encountered on the job, it is reasonable to infer the possibility of workplace impairment from the consumption of the studied drugs. The interpretability of field studies on the effects of alcohol and other drugs is enhanced to the extent that multiple studies employing different methods consistently estimate the correlates of alcohol and other drug use in the work environment, and as the probable confounds in reported alcohol and other drug use/work behavior correlations are identified and controlled for. While the individual studies reviewed below are all subject to methodological criticism, the cumulative body of research may nevertheless provide a basis for making reliable, empirically grounded statements about the impact of alcohol and other drug use on work-related behaviors.

In general, alcohol and other drug use by the work force have been associated with absenteeism, accidents, turnover, and other negative work

behaviors. Research support is most consistent for their association with absenteeism. For other types of outcomes, there is mixed or weak support, with very little support from better-designed studies. The research literature for different outcomes is reviewed below and is summarized in Table 5.1.

The relationship of illicit drug use and absenteeism is perhaps the most robust of any of the outcomes that have been linked with employee drug use. A prospective study of preemployment drug testing in the U.S. Postal Service (Normand and Salyards, 1989; Normand et al., 1990) showed after 1.3 years of employment that employees who had tested positive for illicit drug use at the time they were hired had a mean absence rate of 6.6 percent of hours worked, which was 60 percent higher then the mean absence rate of 4.1 percent for employees who tested negative for drugs at the time of hire. When examining positive tests by type of drug, it was found that employees who tested positive for marijuana use were 1.5 times more likely to use leave time heavily, and employees who tested positive for cocaine were more than 4 times as likely as employees who tested negative to be heavy leave users.

Zwerling et al. (1990) examined preemployment drug test results for postal hires in the Boston area and found similar results for absenteeism. Employees testing positive for marijuana had a mean absence rate of 7.1 percent of scheduled hours worked, those testing positive for cocaine had a mean absence rate of 9.8, and those testing positive for other drugs a rate of 7.9, compared with a rate of 4.0 for employees who tested negative.

In surveys of municipal employees in the southwestern United States (Lehman et al., 1990a,b; Rosenbaum et al., 1992), there were strong relationships between self-reported alcohol and other drug use and unscheduled absenteeism. In one municipal sample (Lehman et al., 1990a), employees who reported having unscheduled absences during the last year were more likely than employees without unscheduled absences to get drunk frequently (18 compared with 13 percent), to drink at work (9 compared with 6 percent), and to report alcohol-related problems (26 compared with 15 percent). In terms of illicit drug use, employees with unscheduled absences were more likely to have used marijuana (30 compared with 20 percent) or other illicit drugs (17 compared with 10 percent) in their lifetime and to have used marijuana in the last year (11 compared with 6 percent). They also were more likely to have used illicit drugs in the last month (7 compared with 4 percent) and to have worked under the influence of illicit drugs in the last year (4 compared with 2 percent).

Results in a sample of municipal employees from a different city (Rosenbaum

et al., 1992) also showed strong relationships between alcohol and other drug use and absenteeism. Employees were classified into three groups based on use in the last year: (1) nonusers who did not report use of any illicit drugs and did not drink frequently or experience negative consequences due to drinking; (2) heavy alcohol users who either drank frequently or experienced negative outcomes due to drinking but did not use illicit drugs; and (3) illicit drug users who reported at least some use of any illicit drug in the last year. Illicit drug users were much more likely than others to report having at least three unscheduled absences in the last year (79 percent), and heavy alcohol users (59 percent) were more likely than nonusers (52 percent) to have excessive absences.

Alcohol and other drug use was also more likely among employees with unscheduled absences in a sample of employees from the local housing authority from the same city as the first municipal sample above. Employees with unscheduled absences were more likely than others to drink at work (27 compared with 9 percent) and to report alcohol-related problems (24 compared with 12 percent). They were more likely to report lifetime use of marijuana (30 compared with 11 percent) and cocaine (12 compared with 3 percent), and last year use of marijuana (15 compared with 3 percent) and cocaine (9 compared with 1 percent).

It should be noted that, in the three samples above, alcohol and other drug use and absenteeism were based on self-reports and thus are subject to common methods variance. In addition, the relationships reported are based on two-way cross-tabulations and may possibly be explained by other variables. For example, younger employees are more likely to drink heavily and use illicit drugs, and they are also more likely than older employees to have higher absenteeism rates. It is possible that age or other variables can account fully or in part for the observed relationships.

A drug abuse management program at the Utah Power and Light Company was evaluated by Crouch et al. (1989). In this study, employees who tested positive for alcohol and other drugs in mandatory drug screens (usually for cause) and employees who self-referred to the employee assistance program (EAP) were compared with frequency-matched control groups. Employees in the drug-positive group had significantly higher absenteeism rates than did the matched control groups (75.3 accumulated sick hours versus 55.8 hours). The drug-positive group had a mean of 63.8 hours of unexcused absences compared with 18.7 hours for the control group. For the self-referred EAP group, the mean number of accumulated sick hours was 81.7 compared with 56.3 for the control group; the EAP group had 32.2 hours of unexcused absences on average compared with 10.1 for the control group.

This study is limited in several ways. There were only 12 employees in the drug-positive group and 27 in the EAP group, numbers too small to

make stable estimates. It is not known how the samples were selected, and the time period being evaluated in the study is not stated. Furthermore, members of both the drug-tested group and the EAP group may have been in those groups precisely because of problems like absenteeism. Most importantly, the use of frequency-matched control groups is problematic when attempting to produce equivalent groups for comparison purposes (Salyards and Normand, 1991).

Sheridan and Winkler (1989) report on an evaluation of drug testing at the Georgia Power Company. Their data on absenteeism were less consistent than in other studies. Employees who tested positive on company drug tests were compared with employees testing negative. However, results of for-cause and random tests were mixed, and tests for cause are given for employees who have already been identified as having work-related problems. Results showed that positives were not significantly more absent than negatives, but for absenteeism based on unpaid sick leave, personal time, and disciplinary suspensions, negatives were significantly more absent than the company norm, and positives were significantly more absent than negatives in 1986 but not in 1987. These results are, no doubt, influenced by the presence of for-cause tested individuals in both the positive and negative groups. For-cause negatives probably had the kinds of problems associated with alcohol and other drug use or else they would not have been tested.

The relationship between employee alcohol and other drug use and accidents is mixed, despite the heavy media attention given to accidents by employees working under the influence and the seemingly obvious association between working while impaired and having accidents. Perhaps the most well-known findings linking alcohol and other drug use and accidents are found in the transportation industry. Accidents involving public carriers are often highly publicized in the media. A single accident can have catastrophic consequences, by resulting in significant loss of life or injury to the public, extensive environmental damage, and huge damage costs. A few well-publicized accidents can both seriously damage the public's confidence in the safety of the transportation system and give rise to lawsuits for substantial damages, thus greatly affecting the economic health of transportation companies. For these reasons, the relationship between alcohol and other drug use and transportation accidents has long been a well-funded focus of research attention.

Evidence on the relationship between alcohol and drug use and accidents in industries other than transportation is mixed. There are many possible reasons for the failure to find consistent associations, one being that on-the-job accidents are relatively rare events and thus difficult to

predict. Moreover, there is little evidence that working under the influence is widespread; many employees who do work under the influence may be able to compensate for their impairment, and there is a substantial amount of variation across individuals as to how a specific drug at a given dose affects performance. Given the evidence of higher rates of absenteeism among alcohol and other drug abusers, it is possible that many workers who are too impaired to work and thus more susceptible to accidents stay home from work. Some workers who may be impaired on the job work in positions in which they are not at risk for accidents, such as office settings. Other workers impaired on the job who work in positions in which they are at risk for accidents may be able to avoid dangerous tasks, thus reducing the likelihood of an accident.

In the transportation industry, in contrast, workers who operate vehicles cannot avoid jobs in which they are at risk for accidents. Thus, impairment may become a more important issue for vehicle operators than for some other jobs. Even minimizing risky tasks may be different for vehicle operators. For example, truck or bus drivers need to complete their routes and do not often have alternative activities they can be involved in if they are impaired. Moreover, the stigma that attaches to alcohol and other drug use by transportation workers may lead them to work for fear that excessive absences would call attention to their use, whereas workers in other industries would see absence from work as the best solution. Finally, certain stimulant drugs like caffeine and methamphetamines may in certain circumstances slightly improve performance. Transportation workers may feel that such boosts are necessary to meet job demands, even if in the long run dependence has adverse safety consequences.

Because of the special significance of alcohol and other drug use and accidents in the transportation industry, we review the literature on accidents in transportation first, followed by a discussion on the evidence of alcohol and other drug use and accidents in other industries. (For a more detailed treatment of the transportation literature, the reader is referred to Alcohol and Other Drugs in Transportation: Research Needs for the Next Decade (Transportation Research Board, 1993).

Much of the public concern regarding alcohol and other drugs in the work force has been precipitated by highly publicized accidents in the transportation industry. Table 5.2 recalls some examples that helped sensitize public perceptions to problems associated with drugs on the job. Although a few accidents involving alcohol or other drug use do not establish a significant relationship for an industry as a whole, the effect of such accidents on public confidence and perceptions and the significant damage even

a single accident can cause mean that such reports are important in their own right.

Alcohol has long been heavily implicated in fatal automobile accidents. In a study of data from 15 states with relatively complete reporting statistics, Fell (1982) estimated that 59 percent of drivers killed in accidents had alcohol in their blood, and 49 percent had blood alcohol content (BAC) at 0.10 percent or higher. A discriminant analysis approach to estimating alcohol-involved fatal crashes developed by Klein (Klein, 1986; Fell and Klein, 1986) estimated alcohol involvement in about one-half of fatal crashes, with 75 percent of those involving BACs at 0.10 percent or higher. Sixteen percent of injury crashes and 8 percent of property damage accidents were estimated to involve alcohol. Although the presence of alcohol in a fatal accident does not necessarily imply that alcohol was a cause of the accident, the data are so overwhelming that a general causal link cannot be denied.

A report by the Transportation Research Board (TRB) (1987) estimated that 750 fatal crashes occur annually in which a commercial vehicle driver has been drinking. Although this is a substantial number of accidents, only 15 percent of commercial vehicle drivers in fatal crashes had detectable levels of alcohol, compared with 45 percent of all drivers in fatal crashes. However, in 75 percent of fatal crashes involving trucks, someone other than the truck driver is killed. In accidents in which the truck driver is killed, 22 percent of drivers are found to have been drinking and 16 percent to have BACs above 0.10 percent. The TRB, we should note, was forced to estimate the number of alcohol-related crashes because the BACs of commercial vehicle drivers involved in fatal crashes are not well reported. Its report states that BACs are reported for only 11 percent of surviving drivers, and 14 percent of those have alcohol in their bloodstreams, 8 percent above 0.10 percent.

The National Transportation Safety Board (1990) reported on a study of truck drivers from 182 heavy-truck crashes in which the driver was fatally injured. One-third of drivers tested positive for illicit drugs. Marijuana and alcohol were found most frequently (13 percent each) followed by cocaine (9 percent), other stimulants (8 percent), and amphetamines (7 percent). Forty-one percent of the alcohol and other drug-positive drivers tested positive for multiple drugs. Impairment from alcohol and other drugs was determined to be a factor in the accident for 87 percent of cases in which the drivers tested positive for drugs.

Mandatory post-accident drug and alcohol testing for railroad accidents was examined by Moody et al. (1990). Overall, 351 events involving 1,398 persons were included in the study. Of these, 6 percent tested positive for illicit drugs or alcohol. However, results showed that, in 27 percent of fatal accidents and 16 percent of nonfatal accidents, at least one employee involved

in the accident tested positive for an illicit drug. Marijuana was the most common illicit drug found (62 percent of positives) followed by cocaine (20 percent). In approximately one-third of the events that involved drug positives, alcohol or other drugs were found to be causally related to the accident. The raw figures suggest that marijuana plays a greater role compared with alcohol than it probably does, possibly because marijuana metabolites are present in urine for days or weeks (for chronic users) after consumption, yet alcohol is rapidly metabolized. Moreover, as Chapter 4 makes clear, while performance decrements attributable to alcohol emerge clearly in laboratory studies, decrements attributable to marijuana are harder to find.

An analysis of the drug testing program of the Southern Pacific Railroad (Taggart, 1989) describes accident rates before and after widespread drug testing was begun. Urinalysis testing was begun in August 1984 and included for-cause and preemployment testing. Alcohol was also included in the testing protocol. Positive tests decreased from a high of 23 percent at the beginning of the program, to 12 percent in the first full year, to 6 percent in 1987. Personal injuries showed similar declines during the same time period, decreasing from 15.5 per 200,000 person hours worked in the last full year before testing began to 5.8 in the first 6 months of 1988. Train accidents attributable to human failure also dropped from pretest levels. In 1983 there were 22.2 human factor train accidents per 1 million train miles, and in the first 7 months of 1988 there were 2.2 human factor train accidents per 1 million train miles. Although these appear to be impressive changes in accident rates, there were no comparison groups and so it is not possible to determine whether the change in accident rates was due to the testing program. For example, new company policies or programs or changes to the old ones may have been implemented or improvements in track beds or other conditions could have been made during that period, resulting in lower accident rates.

Alcohol and other drug use does not appear to be an important factor in aviation fatalities, according to a study by Kuhlman et al. (1991). Blood, urine, and tissue samples from 377 aviation fatalities in 1989 were analyzed. Alcohol was found in 5 percent of cases, marijuana in 1 percent, and cocaine metabolites in 2 percent. There was minimal use of therapeutic depressants or stimulants. The authors concluded that the data showed no consistent pattern of alcohol or other drug use among aviation fatalities.

The incidence of alcohol and other drug use among workers fatally injured on the job in Alberta, Canada, was examined by Alleyne et al.

(1991), and among workers killed on the job in Harris County, Texas, by Lewis and Cooper (1989).

Alleyne et al. (1991) studied 459 deaths occurring at work in Alberta between 1979 and 1986. Of these deaths, 373 were tested for alcohol, 82 for marijuana, and 329 for other drugs. Eleven percent of fatalities tested for alcohol had detectable alcohol levels, with 4.3 percent over 0.10 percent. The only illicit drug identified was marijuana, which was found in 9 percent of cases tested. Prescription drugs were found in 9 percent of cases, and nonprescription drugs in 7 percent. When alcohol was present in a fatally injured worker, 65 percent of deaths were the result of a motor vehicle accident or getting caught in or under an object. When alcohol was not present, 47 percent of fatalities were due to these types of accidents.

A study of 207 fatal work-related injuries occurring in 1984 and 1985 in Harris County, Texas (including the city of Houston), examined the incidence of alcohol and licit and illicit drugs (Lewis and Cooper, 1989). Detectable levels of alcohol were found in 13 percent of fatalities and 9 percent had BACs over 0.10 percent. Only one case of illicit drug use was detected, and one person tested positive for both alcohol and other drugs. Seven percent of the workers tested positive for prescription drugs that could have altered key functions, such as reaction time or the coordination needed to prevent injury. In neither study of fatal injuries do the authors show that the presence of alcohol or drugs was causally related to the fatal accidents, nor did either study seek to determine whether the deceased workers who tested negative were killed in accidents caused by impaired coworkers.

A telephone survey of 1,740 randomly sampled employed adults in the New England region was conducted by Hingson et al. (1985). Respondents were asked about their levels of alcohol consumption, whether they consumed alcohol at work, and general patterns of drug use. They were also asked about any injuries that required medical attention in the previous year, whether these injuries resulted in an overnight hospitalization, whether they occurred on the job, and the nature of the injuries. If respondents reported injuries, they were asked whether they had consumed any alcohol or other drugs in the 6 hours before the accident.

Overall, 17 percent of respondents reported accidental injuries in the last year, and of these 41 percent occurred at work. More than three-fourths of respondents drank alcoholic beverages, but only 7 percent averaged 3 or more drinks per day. Four percent had been drunk on the job in the month before the interview, and 13 percent reported coming to work hung over or high. A total of 8 percent had taken illicit drugs in the previous month that might have affected their functioning at work.

Results of logistic regression analyses indicated that alcohol consumption at work was not associated with job accidents, but averaging 5 or more drinks per day elevated the relative risk of accidents compared with abstainers.

Heavy drinkers were 1.7 times as likely as abstainers to report an accidental injury, 3.8 times as likely to report injuries requiring hospitalization, and 2 times as likely to report a job-based injury. Respondents reporting use of psychoactive drugs were 1.7 times as likely as abstainers to report work accidents and 2.4 times as likely to report having been hospitalized as the result of an accident.

Holcom et al. (in press) examined the relationship between alcohol and other drug use and accidents in a sample of municipal employees. Information on alcohol and other drug use, accidents, personal background, and job characteristics and background was collected via self-report survey of 1,325 randomly selected employees. Measures of use included lifetime use of any illicit substance, alcohol consumption away from work, illicit drug use in the last year, and use of alcohol or illicit drugs at work in the last year. The accident outcome was a composite measure that included three types of accidents in which the respondent was the source of the accident, not just an accident victim: minor personal injury accidents, major personal injury accidents, and equipment damage accidents. Employees were categorized into high-risk (safety-sensitive) and low-risk job groups for analysis.

Discriminant function analyses were used to classify employees within each risk sample into accident and no-accident groups. Variable domains from personal background, including demographic variables and measures of deviance/social maladjustment, job background, including job characteristics and job attitudes, and alcohol and other drug use were used as discriminators. Employees in the high-risk job category were more likely than employees in low-risk jobs to use alcohol frequently (39 compared with 29 percent) and to use alcohol or other drugs at work (13 compared with 8 percent). In the low-risk job category, there were no differences between employees in the accident and no-accident groups on any of the alcohol and other drug use measures. In the high-risk category, employees who had reported an accident in the last year were more than four times as likely as accident-free employees to report illicit drug use in the last year (17 compared with 4 percent). In the discriminant function analyses, drug use at work and illicit drug use in the last year were the best discriminators of accidents for the high-risk jobs; for low-risk jobs, illicit drug use did not contribute to the discrimination of accident groups. As we have noted when discussing other studies, the significant association of alcohol and other drug use with accidents does not necessarily mean that the relationship is causal.

In the evaluation of the Utah Power and Light Company's substance abuse program (Crouch et al., 1989), accident rates increased between 1983 and 1985, but the trend upward stopped in 1985 when the drug-testing program was initiated, and there were statistically fewer accidents in 1986 and 1987. Drug-positive employees were five times more likely to be in an

accident than their control group. However, the drug-positive group was small and included employees tested for cause. It is unknown how many were tested because of an accident. The evidence does not allow one to conclude that the decrease in the accident rate after initiating the drug-testing program was due to the drug-testing program.

The two postal service studies showed mixed results for accidents and injuries. In the Boston area sample reported by Zwerling et al. (1990), marijuana positives showed an increased risk for accidents and injuries during the first year of employment, and cocaine positives showed a slight increased risk for injuries. Marijuana positives were 1.55 times as likely as drug negatives to have an accident and 1.85 times as likely to be injured; cocaine positives were 1.85 times as likely as drug negatives to be injured. The national study of postal service employees (Normand et al., 1990) did not find a significant relationship between testing positive and having an accident at work. This lack of a relationship may, however, be due to the low rate of accidents.

A number of studies have addressed the issue of turnover or job mobility in relation to alcohol and other drug use. One approach involves the measuring of preemployment illicit drug use, using self-report or urinalysis data, and then following employees after employment to evaluate turnover (e.g., Blank and Fenton, 1989; McDaniel, 1988; Normand et al., 1990; Parish, 1989; Zwerling et al., 1990). Other studies have used population studies of several different populations and analyzed job mobility in relation to alcohol and other drug use (e.g., Kandel and Yamaguchi, 1987; Newcomb, 1988; White et al., 1988).

Several studies examining the relationship between preemployment illicit drug use and turnover were conducted with military recruits. Blank and Fenton (1989) collected data on 1,052 naval recruits who were tested for marijuana and other illicit drugs before entering the Navy. A positive marijuana test did not disqualify potential recruits. Half of the studied sample were marijuana-positive; an equal number of marijuana-negative subjects was selected from the same applicant pool as a comparison group. Approximately 2.5 years after entering the service, 81 percent of the marijuana-negative group and 57 percent of the marijuana-positive group were still in the Navy. Fourteen percent of the marijuana-positive group had left the service due to alcohol or other drug problems and 21 percent because of behavioral or performance problems, compared with 1 percent and 8 percent, respectively, for the marijuana-negative group. However, the increased attrition in the marijuana-positive group may be due to the increased surveillance

these positive new recruits were subjected to as a result of the positive preenlistment test.

McDaniel (1988) used self-reports of illicit drug use among over 10,000 military recruits and compared preservice illicit drug use to unsuitability discharges. Preservice illicit drug use correlated .08 with discharge, although discharge was also correlated with cognitive ability (r = -.06). Thus, even though the prevalence rate of previous illicit drug use was relatively high (49 percent), the validity for predicting job suitability was rather low. McDaniel concludes that other selection factors would probably do a better job of selecting employees who are likely to do well on the job.

In the studies of postal employees, Normand et al. (1990) reported that there was not a significant relationship between drug test results and overall turnover or voluntary turnover. However, drug positives were 1.55 times as likely as drug negatives to be fired. Zwerling et al. (1990) found similar results for turnover, with marijuana positives having 1.56 times higher total turnover than did drug negatives and 2.07 times as high involuntary turnover.

Kandel and Yamaguchi (1987) examined job mobility in relation to alcohol and other drug use in a sample of 1,325 young adults aged 24 to 25 in New York state. Three patterns of drug use were investigated, including daily alcohol use, monthly use of marijuana, and monthly use of other illicit drugs. Job mobility was examined as job separations, job changes, and job losses. Other illicit drug use among men and women and daily alcohol use among women were found to have strong associations with job separation, but these were attributed to selection rather than to impairment because the effects of former and current use of these drugs were equally strong. In other words, these data suggest that those who chose to use alcohol and other drugs are people who would probably experience employment instability even if they did not use them. Current daily use of alcohol among men and marijuana use among women were also associated with higher job separation rates; there were differences between those who continued and those who stopped using these drugs, suggesting that alcohol and other drug use may have played a causal role. When other factors were controlled, concurrent use of marijuana, other illicit drugs, and daily alcohol drinking was associated with a reduction of job tenure of 1.3 years for women and just under 1 year for men. Alcohol and other drug effects were more likely to lead to job losses rather than to job changes.

A study of alcohol and other drug use and work patterns for a sample of middle-class, white ''baby boomers" did not find a relationship between alcohol and other drug use and employment status (White et al., 1988). A longitudinal analysis of 376 respondents available in a follow-up found that neither chronic nor current use of marijuana or alcohol had any adverse effect on the respondent's employment status. However, the sample used in

the study was unique, and the authors warn against extrapolating these results to the larger population.

Newcomb's (1988) follow-up study of young adults in the Los Angeles area reversed the hypothesized causal relationship and used work-related factors to predict measures of problem drinking, marijuana, cocaine, and any drug use. Frequently losing jobs in the last 4 years was a significant predictor of any disruptive drug use, disruptive hard drug use, and disruptive cannabis use. Losing a job in the last 6 months was a significant predictor of disruptive alcohol use, disruptive cannabis use, and any disruptive drug use. Although job instability was one of the most consistent predictors of disruptive alcohol or other drug use among the work-related factors, correlations tended to be small.

A number of studies have found significant associations of alcohol and other drug use with employees' job satisfaction. Depending on the theoretical perspective, job satisfaction can be considered a precursor or an outcome of use. However, the same can be true of many of the other work-related outcomes associated with use. No studies have been able to adequately assess causal direction. However, job satisfaction is an important work-related variable that has been associated with a variety of other work outcomes, such as intentions to quit, turnover, and job performance.

Mangione and Quinn (1975) studied responses from a sample of 1,327 wage and salary workers in the Quality of Employment Survey. Men and women were analyzed separately, and within gender, employees younger than 30 were analyzed separately from employees age 30 and older. There was a significant but small (r = –.12) correlation between job satisfaction and illicit drug use at work, but only for men age 30 and older.

In a laboratory simulation task, Perone et al. (1979) first surveyed industrial workers about drug use, and then, to enhance the likelihood of finding a relationship, placed only those subjects with extreme scores on illicit drug use into their user and nonuser groups. The subjects in these groups were given a laboratory simulation involving a repetitive task. Job satisfaction was measured for both the subject's real jobs and their simulated jobs. There were no major differences between illicit drug users and nonusers with regard to either real or simulated jobs.

Hollinger (1988) examined 9,175 responses to a mailed survey of employees in 47 organizations representing 3 industries. Working while intoxicated was predicted in a logistic model by variables representing age, gender, social interaction with coworkers, and job satisfaction. Dissatisfied employees reported that they worked while intoxicated at a rate 75 percent higher than satisfied employees.

The Lehman et al. (1990a) study of municipal employees also showed significant relationships between job satisfaction and alcohol and other drug use. Employees with low satisfaction were more likely than employees with high satisfaction to report using alcohol (79 compared with 72 percent), getting drunk (52 compared with 44 percent), and having alcohol problems (23 compared with 13 percent). Dissatisfied employees also reported more often that the use of prescription or over-the-counter drugs adversely affected their ability to work (11 compared with 5 percent for over-the-counter; 12 compared with 3 percent for prescription drugs). Finally, dissatisfied employees were more likely than satisfied employees to report lifetime use of marijuana (28 compared with 16 percent) and other illicit drugs (18 compared with 7 percent), marijuana use in the last year (10 compared with 3 percent), last year use of other illicit drugs (7 compared with 2 percent), use of illicit drugs within the last month (7 compared with 2 percent), and use of illicit drugs at work in the last year (3 compared with 0 percent). In a multivariate analysis, including other work and personal background variables, job satisfaction was not found to be a significant predictor of use, suggesting that bivariate relationships may not be causal in nature (Lehman et al., 1991).

Employees' alcohol and other drug use has also been associated with a variety of other job outcome measures. These include counterproductive behaviors such as theft, vandalism, and purposely doing work wrong (Mangione and Quinn, 1975; Newcomb, 1988); job withdrawal behaviors (Lehman and Simpson, 1992; Rosenbaum et al., 1992); low job productivity and difficult interpersonal relations at work (Blum et al., 1993); higher medical claims, EAP referrals, and disciplinary infractions (Salyards, 1993); and negative coworker attitudes (Lehman et al., 1992).

Mangione and Quinn (1975) and Newcomb (1988) found that alcohol and other drug users were more likely than nonusers to engage in counterproductive behaviors. In Mangione and Quinn's study, men over age 30 who reported using illicit drugs at work were more likely than their nonusing counterparts to engage in such deviant behaviors as spreading rumours or gossip, intentionally doing poor work, stealing merchandise or equipment, not reporting accidentally damaged equipment or merchandise, and damaging equipment or merchandise on purpose. There was no significant relationship between these behaviors and use at work for women or for men under age 30. Newcomb found that engaging in vandalism at work was a significant predictor of problem drinking, marijuana use, hard drug use, and any drug use.

Several different measures of on-the-job behaviors were analyzed in

Lehman and Simpson's (1992) study of municipal employees. Job behaviors included positive work behaviors (doing more work than was required, volunteering to work overtime, trying to improve working conditions), psychological withdrawal behaviors (thinking of being absent, daydreaming, chatting with coworkers, spending work time on personal matters), physical withdrawal behaviors (leaving work early, falling asleep at work, taking long lunch breaks), and antagonistic behaviors (arguing with coworkers, filing formal complaints, insubordination). A measure of counterproductive behaviors was examined but not used because of restricted variance.

Examination of bivariate relationships between the work behavior factors and alcohol and other drug use indicated that employees reporting alcohol or other drug use at work, illicit drug use in the last year, lifetime illicit drug use, or heavy alcohol use tended to engage in more frequent psychological and physical withdrawal and antagonistic behaviors. There were no differences between user groups on positive work behaviors, except that employees who reported lifetime illicit drug use reported slightly higher levels of positive work behaviors than did employees who never used illicit drugs.

Each of the four job behavior factors was then regressed on domains of predictor variables, including personal background, job background, job climate, and alcohol and other drug use. The unique contribution of use to the prediction of job behaviors was assessed by adding the block of variables on alcohol or other drug use to the regression equation after the other variable blocks had been entered. These results indicated that use uniquely accounted for significant although small amounts of variance in each of the four job behavior factors, with the amount of uniquely explained variance ranging from 4 percent in the case of physical withdrawal behaviors to 1 percent for positive work behaviors.

An examination by Rosenbaum et al. (1992) of bivariate relationships between a variety of job behaviors and alcohol and other drug use in a sample of municipal workers showed that illicit drug users were more likely than heavy alcohol users, who in turn were more likely than nonusers, to engage in negative work behaviors. For example, 73 percent of illicit drug users argued with coworkers compared with 63 percent of heavy alcohol users and 52 percent of nonusers. Forty percent of illicit drug users disobeyed supervisors' instructions, 32 percent stole work supplies, and 22 percent intentionally did their job wrong compared with 29 percent, 12 percent, and 10 percent of heavy alcohol users and 19 percent, 11 percent, and 6 percent of nonusers, respectively. Likewise, illicit drug users were more likely than others to let coworkers do their work (57 compared with 44 percent of heavy alcohol users and 36 percent of nonusers), complain of illness at work (69 compared with 54 and 48 percent), daydream about nonwork activities (40 compared with 17 and 15 percent), and fall asleep at

work (42 compared with 21 and 17 percent). It should be noted, however, that these measures are based on self-reports and thus are subject both to the biases this method entails and to common methods variance.

Alcohol use and job performance measures were examined by Blum et al. (1993), in their study of 136 employed men. Measures of alcohol consumption and job performance, including assessments of conflict avoidance, technical performance, self-direction, and interpersonal relations, were obtained from the workers and the collateral sources they had named for referrals. Although light and heavy drinkers did not generally differ on their self-assessments of job performance, heavy drinkers were rated as significantly lower on all four scales by their references. Thus heavy drinkers were seen by others as having lower technical performance, self-direction, interpersonal relations, and higher conflict avoidance than light drinkers. More consistent differences were found, however, for the two social aspects of job performance (conflict avoidance and interpersonal relations) than for the technical aspects (technical performance and self-direction).

An expanded analysis of the postal service data examined use of medical benefits, EAP referrals, and disciplinary infractions an average of 3.3 years after hire (Salyards, 1993). Results indicated that employees who tested positive at the time of hire were more than 2.7 times as likely as employees who tested negative to subsequently experience problems requiring EAP intervention, with marijuana positives being twice as likely and cocaine positives being 6.27 times as likely to do so. Positives were 3.5 times as likely as negatives to use the EAP because of alcohol problems and 5.7 times as likely to do so because of illicit drug problems. Although drug positives were more likely than negatives to be disciplined, much of the difference was due to attendance-related offenses. However, positives were 1.6 times as likely as negatives to be disciplined for conduct offenses.

Those who tested positively for illicit drugs at the time they were hired also showed significantly higher utilization of medical benefits than those who tested negative. Comparisons between positives and negatives indicated that, compared with negatives, positives were 1.7 times as likely to be above the median on the total number of claims filed, 1.9 times as likely to be above the median on the total dollar amount of their claims, and 3.4 times as likely to file claims involving alcohol or other drug-related diagnoses (Salyards, 1993).

Alcohol and other drug use has also been linked to employees' job attitudes and perceptions of coworkers. Lehman et al. (1992) asked municipal employees about their awareness of use among their coworkers, whether use in their work group affected their ability to get the work done, and their attitudes toward company policies regarding employee use of alcohol and other drugs. More than 40 percent of respondents reported that coworker use sometimes caused poor-quality work in their work group; more than 30

percent blamed coworker use for poor communication, more chances for injuries, and damaged equipment. Employees exposed to higher levels of alcohol and other drug use by coworkers also were much more likely to attribute work-related problems to coworker use.

Employees exposed to alcohol and other drug use in their work group viewed management efforts to deal with the problem much more negatively than did employees who were not exposed. Employees not aware of coworker use were much more likely not to have an opinion about company efforts to deal with the matter. Employees exposed to coworker use were also more likely to favor preemployment and random urine testing programs. A discriminant analysis used attitudes such as job satisfaction, job involvement, faith in management, and organizational commitment to discriminate between employees not exposed to coworker use, and those exposed to low or high levels of coworker use. Job attitudes significantly discriminated between coworker use groups. Employees with a high exposure to coworker use had lower job satisfaction and less faith in management than did employees not aware of coworker use.

A wide variety of job outcomes and behaviors has been associated with employees' use of alcohol and other drugs. These include absenteeism, accidents, turnover, job satisfaction, counterproductive behaviors, psychological and physical withdrawal behaviors, and coworker attitudes. Despite the wide variety of research in the studies reviewed above, few definitive statements can be made about the impact of using alcohol and other drugs on job performance. The abundance of evidence presented here indicates that the relationship between use and job behaviors and outcomes is clearly negative. However, the magnitude of the relationships found is generally small, and causal spuriousness and direction are problems that have not been adequately addressed in the literature.

Part of the problem is that the research designs and methods are not amenable to establishing causality. Even when reliable relationships can be reported, the current research does not allow unequivocal causal assertions about those relationships. Perhaps the most reliable relationship involves use and absenteeism. However, the evidence does not necessarily show that use causes higher absenteeism. It is possible that other variables can account for the relationship, for example, general deviance or subjective job stress that induces both use and absenteeism.

The other job outcome that has shown consistent results involves accidents in the transportation industry. Although the presence of alcohol and other drugs in transportation accidents is considerable, it is much lower than that found generally in motor vehicle accidents. However, accidents in

the transportation industry, perhaps to a greater extent than most other industries (except possibly the nuclear industry), have the potential to cause great harm to the public and thus are more likely to result in public concerns regarding employees' use of alcohol and other drugs and lack of confidence in public carriers. Alcohol is by far the most implicated drug in transportation industry accidents. Laboratory studies on the effects of alcohol as well as predicted differences in the role alcohol plays in single versus multiple car fatalities allow cause to be ascribed with more confidence in transportation accidents.

When attempting to understand the complex relationships between alcohol and other drug use and job performance across diverse industries, the issues of deviance and impairment are often pertinent. Untangling the magnitude and direction of causal links—which may include deviance and impairment and which may go from behavior to alcohol and other drug use or vice versa or both—poses difficult but critically important issues that must be attended to when attempting to attribute negative work outcomes to use. And yet, as we point out in Chapter 7, it is less critical if the goal is simply to decide whether a drug-testing program will have an impact on an organization's overall productivity level.

To date, most of the research efforts that have sought to shed some light on the potential causal relationship between alcohol and other drug use and job performance have used designs and research methods that do not allow these relationships to be satisfactorily untangled. Relationships that are reported are often mixed and generally not very strong. For example, some studies have found positive relationships of use to turnover; others have not. Some of the inconsistency can be attributed to different or flawed research designs. Inconsistencies can also be attributed to the highly skewed distributions of most variables of interest. Most employees do not abuse illicit drugs or alcohol, most employees do not have accidents at work, and most employees do not abuse absence policies. Thus, finding consistent relationships between relatively rare events such as alcohol and other drug abuse and accidents requires a carefully designed study with a large sample size and reliable measures—a difficult task indeed. It is not enough, however, to show consistency or even causality. From a policy standpoint, attention must be paid to the magnitude of these effects as well.

The use of flawed research designs is also an important issue to be considered when interpreting these studies. Unreliable variables and common method variance plague studies based on self-report data; specification errors abound in models attempting to find the most relevant predictors of job-related behaviors; urine test results are erroneously used to classify employees as users or nonusers (i.e., high rate of false negatives); and evaluation studies use inadequate comparison or control groups, or no comparison groups at all, to name some of the problems that a reader encounters.

Each of the studies reviewed here contains serious limitations. However, consistent results across even seriously flawed designs can help increase our understanding of this difficult area. This is fortunate because, even though we can and should do much better than we have done, flaws are often unavoidable in studies of complex human behaviors.

It is thought that alcohol and other drug use by the work force has a significant impact on society. This perception appears corroborated by the extant research, for virtually all of it, including the best-designed studies, report some associations between alcohol or other drug use and distressing, dangerous, or other dysfunctional behaviors. However, it is difficult, given the current research base, to make definitive statements regarding the magnitude of the impact of alcohol and other drug use at work. Many of the effects found, though significant statistically, are small to moderate. Indeed, the available research, taken as a whole, should soften the concern about employee alcohol and other drug use often found in the popular media. But the picture that science presents of alcohol and other drug problems in the workplace may change in either direction. Our current understanding of the area is limited, and much more research needs to be done.

The field studies reviewed above have attempted to assess the causal link between alcohol and other drug use and work-related behavior. Other researchers, primarily economists, have tried to go further and estimate the costs that such use imposes on society. This research has received substantial attention both in the scientific community and among the public. This section describes the nature of such studies and critically evaluates their methods and conclusions. We begin by briefly summarizing one of the best and most widely cited such studies. We then explain why this study and related studies are by themselves inadequate to guide public policies. Finally, we examine in detail how such studies attempt to evaluate the effect of alcohol and other drug use on worker productivity.

Rather than summarize the cost-of-drug-use studies done to date, we focus on one recent, widely cited study that subsumes the results of most previous studies and represents perhaps the best example of such studies. This study (Rice et al., 1990) determines the costs of alcohol and other drug use to society by totaling estimates of treatment, morbidity, mortality, and crime costs. We discuss each of these cost estimates in turn. For brevity, our discussion of those cost estimates focuses most strongly on results related

to illicit drugs, and we only briefly discuss the results related to the use of alcohol. We note, however, that Rice et al. estimate costs of alcohol use at $70 billion compared with $44 billion for illicit drug use.

Treatment and support costs are those health care expenditures that result from illicit drug use. They include expenditures for short-term hospitalization, outpatient care, medicine, research and training, program administration, and the net cost of private insurance related to illicit drug treatment. For the year 1985, Rice et al. estimate illicit drug treatment and support costs of approximately $2.1 billion. The vast majority of this total stems from visits to hospitals and other institutions devoted specifically to the treatment of illicit drug-abuse-related conditions.

Morbidity costs reflect the "lowered" productivity that results from illicit drug use. Cost-of-drug-use studies equate productivity with income. They compare the income of illicit drug users and nonusers to determine the reduction in income due to illicit drug use. Personal characteristics such as age, sex, and education are commonly controlled in making these comparisons. Rice et al. estimate morbidity costs for 1985 of just under $6 billion.

Mortality costs equal the income that would have been earned by individuals who die from illicit drug use had they not died prematurely. To estimate mortality costs, it is assumed that a person of a given age and sex who dies of illicit drug use would have had a future income stream consistent with the current cross-sectional age-sex distribution of income. The estimated mortality costs for 1985 are $2.6 billion.

Crime-related costs have two main components. The first consists of public and private expenditures on the portion of the criminal justice system that deals with illicit drug-related offenses, including federal efforts at drug traffic control. This component of crime-related costs is estimated at $13.2 billion for 1985. The second component is the losses imposed on the victims of illicit drug-related crimes and the forgone earnings of individuals who are incarcerated for their illicit drug use or who engage in criminal activity, rather than legal employment, because of their illicit drug use. This component of crime-related costs is estimated at $19.3 billion for 1985, of which almost $14 billion represents forgone earnings of individuals who engage in criminal activity because of illicit drug use.

Although their conclusions are widely cited by the media, politicians, and others in the public policy arena, cost-of-drug-use studies do not by themselves provide an economic justification for any particular public policy toward alcohol and other drugs. The reason they do not is that, as emphasized by Harwood et al. (1984), Harwood (1991) and Sindelar (1991), economically based policy recommendations should reflect an evaluation of the

costs and benefits of particular policies, not an evaluation of the total costs of the activity the policy seeks to change. This point is fundamental and well recognized by the authors of many cost-of-drug-use studies (e.g., Harwood et al., 1984), but it is often overlooked by those who cite these studies to justify government actions.

Measuring the costs associated with alcohol and other drug use does not, in other words, determine the extent to which these costs might be increased or decreased by specific public policies. For example, consider the policy proposal that funding for alcohol and other drug abuse treatment centers be increased. If those who would receive treatment as a result of additional funding are likely to be program dropouts, additional funding might be ineffective in reducing alcohol and other drug use and its associated costs. If, however, the new clientele consists of people who are strongly desirous of reducing their alcohol and other drug use but are unable to afford existing treatment programs, such expenditures might be highly cost-effective. Estimating the total amount of money currently spent on treatment, which is essentially what occurs in studies like those by Rice et al. (1990), tells us nothing about the likely effects of increased funding for treatment.

The component of cost-of-drug-use studies that is most relevant to understanding the interactions between alcohol and other drugs and the workplace is the "diminished" worker productivity attributable to alcohol and other drug use. Existing studies estimate this cost component by comparing the wage rates (or income) of individuals (or households) who use illicit drugs or alcohol with the wage rates (or income) of those who do not, controlling in the better studies for other observable characteristics such as age, sex, race, education, health status, marital status, and family background. A finding that illicit drug or alcohol use is correlated with wage rates or income, possibly controlling for other factors, is usually interpreted as showing that illicit drug or alcohol use affects worker productivity. It should be noted, however, that, for the purpose of this report, productivity effects and earning effects are used interchangeably. Ideally, one would like to be able to obtain a direct measure of productivity; however, because such measures are not easily obtained, researchers must settle for using less than perfect indices or proxies of the construct of interest (i.e., productivity). That is, earning differentials are typically evaluated in attempts to estimate the impact of alcohol and other drug use on productivity.

Despite important differences in data sets, productivity measures, control variables, sample periods, and estimation procedures, a consistent set of

empirical regularities emerges from this literature. These are described in the sections that follow.

Problem or heavy drinking is negatively related to household income, although the strength and statistical significance of the relation depends on the measure of problem drinking considered. Berry and Boland (1977), using a 1968 National Household Survey, show that the mean income of households with a male problem drinker (defined as having a large number of alcohol-related consequences) is 18 percent lower than the mean income of households without a problem drinker. Harwood et al. (1984), extending work by Cruze et al. (1981), use 1979 National Household Survey data from NIAAA to show that most available indicators of problem drinking are negatively correlated with household income (controlling for other characteristics) and that a few of these indicators display large and significantly negative correlations with household income. Rice et al. (1990), using 1980-1984 Epidemiological Catchment Area (ECA) data, find that a measure of lifetime alcohol abuse or dependence is significantly negatively correlated with household income. These results do not, however, exclude the possibility that low income either contributes to or proxies for variables that contribute to problem drinking.

Mullahy and Sindelar (1989) also examine the relation between alcohol abuse and income using ECA survey data. They find that alcoholism has no direct relation with earnings but does have an indirect relation: alcohol abuse is associated with lower educational attainment and marital stability, which in turn are associated with lower earnings. In a follow-up study, Mullahy and Sindelar (1991) examine gender differences in labor market responses to alcoholism. Using multiple-site data from the ECA survey, they find that alcoholism is typically negatively related to both labor force participation and income for the full sample. However, the relation varies by stage of the life cycle and by gender. They demonstrate that their results depend on the variables controlled for and whether the dependent variable is labor force participation or income.

Low to moderate drinking is associated with higher wages. Berger and Leigh (1988) use the Quality of Employment Survey to estimate separate wage equations for drinkers and nondrinkers among both men and women. They find that drinkers earn higher wages than nondrinkers, after controlling for differences in observable characteristics. Their result is robust to changes in model specification, identification assumptions, and definition of drinker status. Harwood et al. (1984) find that over a range of alcohol consumption of up to two ounces per day, the amount of alcohol consumed

is positively correlated with household income. This might occur because higher income leads to increased discretionary consumption of alcohol.

Heavy or long-term marijuana use is negatively related to both household income and wages for men but appears to be positively related to wages and household income for women. Harwood et al. (1984), considering a population that aggregates men and women, find that daily marijuana use for a period of at least a month at some point in the past is negatively related to household income. Register and Williams (1992), considering men only, report that in the 1984 National Longitudinal Survey of Youth (NLSY), marijuana use for longer than 8 years is negatively related to current wages. Kaestner (1991), who also examines the 1984 NLSY, finds that heavy lifetime or past-30-day marijuana use is insignificantly negatively related to wages for men and significantly positively related for women.

Current or moderate lifetime marijuana use is either essentially uncorrelated with wages or income or is modestly positively correlated. Harwood et al. (1984) report that all measures of current marijuana use or lifetime marijuana use other than having smoked marijuana daily for at least a month display insignificant correlations with household income, with some point estimates positive and some negative. Rice et al. (1990) state that results based on use of particular drugs are ill defined and therefore do not report any details. Register and Williams (1992) find that, averaged over the whole sample, marijuana users and slightly lower wages than nonusers. Controlling for a range of observed characteristics, however, reveals a positive and significant correlation between marijuana use and current wages. Kaestner (1991) reports negative, insignificant correlations for men and positive, generally significant correlations for women with all measures of marijuana use.

The relation between any level of cocaine or other nonmarijuana drug use and wage rates or income is either insignificant or positive. Harwood et al. (1984) state that they could find no significant results relating illicit drug use to household income for any illicit drug other than marijuana. Again, Rice et al. (1990) state that results based on the use of particular illicit drugs were ill defined and therefore do not report any details. Gill and Michaels (1992), using 1980 and 1984 NLSY data, find that a wage differential that favors illicit drug users over nonusers increases when the comparison is made between hard drug users (cocaine, heroin, etc.) and all other illicit drugs. Register and Williams (1992) document a higher average

wage for cocaine users than nonusers, and they show that cocaine use is positively although insignificantly correlated with wages after accounting for observable characteristics. Kaestner (1991) finds that virtually all measures of cocaine use are positively correlated with wage rates, and these correlations are substantially larger and more significant for women than men.

In studies for which comparisons are available, the relation between alcohol and other drug use and productivity is usually more negative (less positive) for men than for women. Rice et al. (1990) show that alcohol abuse is more negatively correlated with household income for men than women up through age 34, after which the relation reverses, and that illicit drug abuse is negatively correlated with household income for men of all ages but positively correlated for women of all ages. Kaestner (1991) documents a negative and usually insignificant correlation between various measures of marijuana use and wages for young men and a positive and usually significant correlation for young women. He documents a positive but small and usually insignificant correlation between various measures of cocaine use and wages for men and a positive, larger, and usually significant correlation for women.

The comparisons that employ household income as a measure of productivity suggest, on average, associations that are more negative between alcohol and other drug use than comparisons based on wages. However, no study uses data on both measures of productivity, so it is difficult to isolate the effects of using a particular productivity measure. In comparing studies with one measure or the other, one also faces differences in data sets, drug variables, control variables, sample periods, and population characteristics (e.g., only young people in the wage regressions whereas all ages in household income regressions). Nevertheless, when similar alcohol and other drug variables are employed, household and individual measures of productivity display similar correlations with alcohol and other drug use. In large part, earlier studies appear to have obtained substantially more negative relations than more recent studies because they emphasized those results that used measures of heavy use or abuse. Even in the early studies, most measures of use for alcohol and other drugs show small negative to small positive correlations with productivity.

The failure of alcohol and other drug use to display the ''expected" negative correlation with productivity does not appear to result mainly from

an effect of alcohol and other drug use on productivity via the labor supply. Register and Williams (1992), using the NLSY data, find that marijuana use is associated with a decreased probability of being employed but that cocaine use has no significant relation to the probability of employment. Zarkin et al. (1992), using the 1990 National Household Survey on Drug Abuse, examine the prevalence of illicit drug use by work status and the relation between illicit drug use and measures of labor supply: weeks worked during the past year, number of sick days taken during the past month, and number of days of work skipped for nonmedical reasons during the past month. Their prevalence estimates indicate that the working population has a slightly higher rate of alcohol and other drug use than the total household population. Their regression results indicate that illicit drug use is associated with fewer weeks worked, whereas alcohol use is associated with more weeks worked. Illicit drug use is not related to the number of sick days taken, but past-month illicit drug use is related to an increased number of work days skipped. Kaestner (1992) finds a negative relation between illicit drug use and labor supply in cross-sectional estimates but little significant relation in longitudinal estimates.

Despite the substantial differences in results across studies, careful reading suggests some consistencies in the empirical results. Heavy or problem use of marijuana or alcohol is generally associated with lower productivity (with the possible exception of women in the case of marijuana) and low to moderate use of any illicit drug or alcohol is either positively associated with productivity or simply not significantly related.

Despite the consistency of the empirical evidence, however, interpretation of this evidence must proceed cautiously. The key problem of interpretation may be called the heterogeneity effect. Put simply, individuals differ along several dimensions, some observable and some not. Observable characteristics include sociodemographics such as education, age, race, and gender. These variables are likely to influence job compensation and are usually included in wage equations. However, several unobservable characteristics may be equally or even more likely to influence wages. These unmeasured variables include motivation, aggression, intelligence, ambition, discipline, and the like. If characteristics such as these influence wages, and if alcohol or other drug use is correlated with one or more of these variables, then the estimated relation between such use and wages will tend to pick up these latent effects. Thus, an estimated negative relation between use and wages or income may simply indicate that less ambitious people are likely to both use alcohol or other drugs and have low wages; the inference that such use causes low wages would not be justified. Similarly, an estimated positive

relation could indicate that creative or gregarious individuals are likely to both use alcohol or other drugs and have high wages without suggesting any causal effect.

Another problematic aspect of some of these studies and others (Mills and Noyes, 1984; Newcomb and Bentler, 1986, 1988; Kandel 1984; White et al., 1988) that have examined the association between alcohol and other drug use and income level is that most results are based on cross-sectional data largely among youthful samples. Potential problems associated with cross-sectional designs and the use of restricted age range samples is illustrated by the results of recent analyses of longitudinal data. In analyzing income and alcohol and other drug use data, Newcomb and Bentler (1988) found that greater polydrug use by teenagers was associated with increased income 4 years later. They explained this finding by noting that adolescent polydrug users were more likely than those who used few or no drugs to begin working right out of high school and not attend college. Those who used few or no drugs as teenagers were more likely to go to college and delay entry into the work force. Therefore, adolescent heavy alcohol or other drug users may be expected to earn higher incomes than less heavy users as young adults, since they will have been in the work force for 4 years while those less involved in drugs continued their education.

In a follow-up of this same sample, Newcomb and Bentler (1992) found that 4 years later the relationship between income and teenage polydrug involvement had reversed. Both adolescent alcohol and other drug use and increased use of alcohol and other drugs into adulthood was associated with reduced income by the time people reached their mid-to-late twenties. The researchers explain this reversal by assuming that the greater education of the low alcohol and other drug users eventually resulted in higher income (an elevated earning potential ceiling) than that enjoyed by those who were involved with alcohol and other drugs as teenagers and maintained or continued such involvement. The latter had a short-term income benefit but suffered over the long run from a low ceiling to their earning potential.

Clearly, more prospective data must be analyzed to characterize more adequately and precisely the dynamic relationship between alcohol and other drug use and income. Reliance on cross-sectional data is inappropriate and possibly gravely misleading. Furthermore, a much wider age range must be studied than the primarily young groups represented in these studies.

In addition to these general problems, there are some more technical but nonetheless fundamental problems with most cost-of-drug use research. Wage rates as a measure of productivity are biased because they are observed only for employed individuals and do not reflect the productivity costs that would be incurred by individuals out of the labor force. While researchers today commonly recognize and attempt to address this problem (Gill and Michaels, 1992; Register and Williams, 1992; Kaestner, 1991),

the methods used, so-called Heckman corrections (Heckman, 1976, 1979), do not seem adequate to the task. Not only are they sensitive to model misspecification, but at best they can only estimate how alcohol and other drug use would have affected the wages of the unemployed if they were employed. What matters, however, is the extent to which alcohol and other drug use leads individuals to become unemployed. But even these criticisms are less important than the fact that Kaestner (1991) demonstrates that the corrections appear to make no difference to the results in any event.

Some of the studies that use wages as the productivity measure attempt to control for the fact that the relationship between wages and alcohol and other drug use is potentially bidirectional; each may affect the other (e.g., Kaestner, 1991). Such similarity, if it exists, will inflate the estimated correlation between alcohol and other drug use and wages. Correcting for the simultaneity problem requires the availability of "instrumental" variables that affect alcohol and other drug use and not wages. Spousal income is commonly treated as such a variable, but Gill and Michaels (1992) demonstrate that the spouses of alcohol and other drug users typically have low income. If, holding other demographic characteristics constant, users of alcohol and other drugs tend to marry people with low incomes, then simultaneous equation estimation techniques will incorrectly adjust for the effect of income on alcohol and other drug use and could lead to overestimates of the negative relation between such use and wages. Simultaneous equations estimates may thus be worse than the estimates that make no correction for simultaneity bias.

In sum, there are serious difficulties with estimating the causal relations between alcohol and other drug use and productivity and with attempts to estimate the costs of use in general. Current estimates can be taken only as ballpark figures. They are probably correct in suggesting that there are great costs, even if the most concretely measured costs are those incurred in the attempt to control alcohol and other drug use because of the assumed magnitude of nonenforcement costs. Work to refine cost-of-drug-use measures should proceed, but even if substantial progress is made, we should not confuse measures of the cost of drug use with the expected net benefits of policies aimed at limiting those costs.

The committee's conclusions and recommendations that follow are based on a critical review of the literature on the impact of alcohol and other drug use on employees' work-related behavior as well as studies that have explored the relationship between use and productivity/cost estimates. They are intended to highlight the need to expand scientific knowledge of how alcohol and other drug use affects work-related behavior and to improve the quality of research aimed toward this end.

• Field studies have consistently linked alcohol and other drug use to higher rates of absenteeism; they also provide evidence of an association between alcohol and perhaps other drug use and increased rates of accidents, particularly in the transportation industry. Less consistent evidence exists linking alcohol and other drug use to other negative work behaviors, although the current research base is insufficient to support firm conclusions. When associations between alcohol and other drug use and counterproductive workplace behavior are found, relationships are most often of moderate or low strength even when they are statistically significant.

• The empirical relationships found between alcohol and other drug use and job performance are complex and need not imply causation. Relationships may exist for some job performance outcomes like absenteeism but not for others. Alcohol and other drug use may be just one among many characteristics of a more deviant lifestyle, and associations between use and degraded job performance may be due not to drug-related impairment but to general deviance or other factors.

Recommendation: To intervene more effectively in improving job performance, we must develop a better research base from which to assess how alcohol and other drug use and other factors act alone and in combination to degrade job performance.

• Widely cited cost estimates of the effects of alcohol and other drug use on U.S. productivity are based on questionable assumptions and weak measures. Moreover, these cost-of-drug-use studies do not provide estimates of potential savings associated with implementing particular public policies toward alcohol and other drugs.

Recommendation: Further research is needed to develop refined, defensible estimates of how much alcohol and other drug use costs specific organizations and society at large. Business decision makers and policy makers should be cautious in making decisions on the basis of evidence currently available.

Alleyne, B.C., P. Stuart, and R. Copes 1991 Alcohol and other drug use in occupational fatalities. Journal of Occupational Medicine 33:496-500.

Berger, M.C., and J.P. Leigh 1988 The effect of alcohol use on wages. Applied Economics 20:1343-1351.

Berry, R.E., Jr., and J.P. Boland 1977 The work-related costs of alcohol abuse. In C.J. Schramm, ed., Alcoholism and Its Treatment in Industry. Baltimore, Md.: The Johns Hopkins University Press.

Blank, D.L., and J.W. Fenton 1989 Early employment testing for marijuana: demographic and employee retention patterns. In S.W. Gust and J.M. Walsh, eds., Drugs in the Workplace: Research and Evaluation Data. NIDA Research Monograph No. 91. Rockville, Md.: National Institute on Drug Abuse.

Blum, T.C., P.M. Roman, and J.K. Martin 1993 Alcohol consumption and work performance. Journal of Studies on Alcohol 54:61-70.

Crouch, D.J., D.O. Webb, L.V. Peterson, P.F. Buller, and D.E. Rollins 1989 A critical evaluation of the Utah Power and Light Company's substance abuse management program: absenteeism, accidents, and costs. In S.W. Gust and J.M. Walsh, eds., Drugs in the Workplace: Research and Evaluation Data. NIDA Research Monograph No. 91. Rockville, Md.: National Institute on Drug Abuse.

Cruze, A., H. Harwood, P. Kristiansen, J. Collins, and D. Jones 1981 Economic Costs of Alcohol and Drug Abuse and Mental Illness-1977, Research Triangle Institute, Research Triangle Park, N.C.

Donovan, J.E., and R. Jessor 1985 Structure of problem behavior in adolescence and young adulthood. Journal of Consulting and Clinical Psychology 53:890-904.

Fell, J.C. 1982 Alcohol involvement in traffic accidents. In Alcohol and Highway Safety: A Review of the State of the Knowledge. Report DOT-HS-806-269. Washington, D.C.: U.S. Department of Transportation.

Fell, J.C., and T. Klein 1986 The Nature of the Reduction in Alcohol in U.S. Fatal Crashes. SAE Technical Paper 860038, Society of Automotive Engineers, Warrensdale, Pa.

Freier, M.C., R.M. Bell, and P.L. Ellickson 1991 Do Teens Tell the Truth? The Validity of Self-Reported Tobacco Use by Adolescents. Santa Monica, Calif.: The RAND Corporation.

Gill, A.M., and R.J. Michaels 1992 Does drug use lower wages? Industrial Labor Relations Review 45:419-434.

Gillmore, M.R., J.D. Hawkins, R.F. Catalano, Jr., L.E. Day, M. Moore, and R. Abbott 1991 Structure of problem behaviors in preadolescence. Journal of Consulting Clinical Psychology 59(4):499-506.

Harwood, H.J. 1991 Economics and drugs: promises, problems, and prospects. In W.S. Cartwright and J.M. Kaple, eds., Economic Costs, Cost-Effectiveness, Financing, and Community-Based Drug Treatment. Rockville, Md.: National Institute on Drug Abuse.

Harwood, H.J., D.M. Napolitano, P.L. Kristiansen, and J.J. Collins 1984 Economic Costs to Society of Alcohol and Drug Abuse and Mental Illness: 1980. Report to Alcohol, Drug Abuse, and Mental Health Administration. Research Triangle Institute, Research Triangle Park, N.C.

Heckman, J.J. 1976 The common structure of statistical models of truncation, sample selection, and limited dependent variables and a single estimator for such models. Annals of Economic and Social Measurement. 5:475-492. 1979 Sample selection bias as a specification error. Econometrica 47:153-161.

Hingson, R.W., R.I. Lederman, and D.C. Walsh 1985 Employee drinking patterns and accidental injury: a study of four New England States. Journal of Studies on Alcohol 46:298-303.

Holcom, M.L., W.E.K. Lehman, and D.D. Simpson in press Employee accidents: Influences of personal characteristics, job characteristics, and substance use. Journal of Safety Research.

Hollinger, R.C. 1988 Working under the influence (WUI): correlates of employees' use of alcohol and other drugs. Journal of Applied Behavioral Science 24:439-454.

Jessor, R., and S.L. Jessor 1977 Problem Behavior and Psychosocial Development: A Longitudinal Study of Youth. New York: Academic Press. 1978 Theory testing in longitudinal research on marijuana use. In D. Kandel, ed., Longitudinal Research on Drug Use. Washington, D.C.: Hemisphere.

Kaestner, R. 1991 The effect of illicit drug use on the wages of young adults. Journal of Labor Economics 9:381-412. 1992 The Effect of Illicit Drug Use on the Labor Supply of Young Adults. Unpublished manuscript, Rider College, Trenton, N.J.

Kandel, D.B. 1984 Marijuana users in young adulthood. Archives of General Psychiatry 1:200-209.

Kandel, D.B., and K. Yamaguchi 1987 Job mobility and drug use: an event history analysis. American Journal of Sociology 92:836-878.

Klein, T. 1986 A Method for Estimating Posterior BAC Distributions for Persons Involved in Fatal Traffic Accidents. Washington, D.C.: Sigmastat.

Kuhlman, J.J., B. Levine, M.L. Smith, and J.R. Hordinsky 1991 Toxicological findings in Federal Aviation Administration general aviation accidents. Journal of Forensic Sciences 36:1121-1128.

Lehman, W.E.K., and D.D. Simpson 1992 Employee substance use and on-the-job behaviors. Journal of Applied Psychology 77:309-321.

Lehman, W.E.K., M.L. Holcom, and D.D. Simpson 1990a Employee Health and Performance in the Workplace: A Survey of Municipal Employees of a Large Southwest City. Unpublished manuscript, Institute of Behavioral Research. Texas Christian University, Fort Worth. 1990b Employee Health and Performance in the Workplace: A Survey of Employees of the Housing Authority in a Large Southwest City. Unpublished manuscript, Institute of Behavioral Research. Texas Christian University, Fort Worth.

Lehman, W.E.K., D.J. Farabee, M.L. Holcom, and D.D. Simpson 1991 Prediction of Substance Use in the Workplace: Unique Contributions of Demographic and Work Environment Variables . Unpublished manuscript, Institute of Behavioral Research. Texas Christian University, Fort Worth.

Lehman, W.E.K., D.J. Farabee, and D.D. Simpson 1992 Co-Worker Substance Use and Its Relationship with Employee Attitudes and Morale. Unpublished manuscript, Institute of Behavioral Research. Texas Christian University, Fort Worth.

Lewis, R.J., and S.P. Cooper 1989 Alcohol, other drugs, and fatal work-related injuries. Journal of Occupational Medicine 31:23-28.

Mangione, T.W., and R.P. Quinn 1975 Job satisfaction, counterproductive behavior, and drug use at work. Journal of Applied Psychology 60:114-116.

McDaniel, M.A. 1988 Does pre-employment drug use predict on-the-job suitability? Personnel Psychology 41:717-729.

Mills, C.J., and H.L. Noyes 1984 Patterns and correlates of initial and subsequent drug use among adolescents. Journal of Consulting and Clinical Psychology 52:231-243.

Moody, D.E., D.J. Crouch, D.M. Andrenyak, R.P. Smith, D.G. Wilkins, A.M. Hoffman, and D.E. Rollins 1990 Mandatory post accident drug and alcohol testing for the Federal Railroad Administration (FRA) . In S.W. Gust, J.M. Walsh, L.B. Thomas, and D.J. Crouch, eds., Drugs in the Workplace: Research and Evaluation Data, Vol. II. NIDA Research Monograph No. 100. Rockville, Md.: National Institute on Drug Abuse.

Mullahy, J., and J.L. Sindelar 1989 Alcoholism and Human Capital. Discussion paper QE89-06, Quality of the Environment Division, Resources for the Future, Washington, D.C. 1991 Gender differences in labor market effects of alcoholism. AEA Papers and Proceedings May:161-165.

National Transportation Safety Board 1990 Safety Study: Fatigue, Alcohol, Other Drugs, and Medical Factors in Fatal-to-the-Driver Heavy Truck Crashes, Vol 1. NTSB/SS-90/01. Washington, D.C.: National Transportation Safety Board.

Newcomb, M.D. 1988 Drug Use in the Workplace: Risk Factors for Disruptive Substance Use Among Young Adults. Dover, Md.: Auburn House.

Newcomb, M.D., and P.M. Bentler 1986 Cocaine use among adolescents: longitudinal associations with social context, psychopathology, and use of other substances. Addictive Behaviors 11:263-273. 1988 Consequences of Adolescent Drug Use: Impact on the Lives of Young Adults. Newbury Park, Calif.: Sage Publications. 1992 Adolescent and Subsequent Changes in Young Adult Drug Use: Effects on Psychosocial Functioning in Adulthood. Unpublished manuscript, University of Southern California .

Normand, J., and S.D. Salyards 1989 An empirical evaluation of preemployment drug testing in the United States Postal Service: interim report of findings. In S.W. Gust and J.M. Walsh, eds., Drugs in the Workplace: Research and Evaluation Data. NIDA Research Monograph No. 91. Rockville, Md.: National Institute on Drug Abuse.

Normand, J., S.D. Salyards, and J.J. Mahoney 1990 An evaluation of preemployment drug testing. Journal of Applied Psychology 75:629-639.

Parish, D.C. 1989 Relation of the pre-employment drug testing result to employment status: a one-year follow-up. Journal of General Internal Medicine 4:44-47.

Perone, M., R.J. DeWaard, and A. Baron 1979 Satisfaction with real and simulated jobs in relation to personality variables and drug use. Journal of Applied Psychology 64:660-668.

Register, C.A., and D.R. Williams 1992 Labor market effects of marijuana and cocaine use among young men. Industrial Labor Relations Review 45:435-448.

Reinisch, E.J., R.M. Bell, and P.L. Ellickson 1991 How Accurate are Adolescent Reports of Drug Use? Santa Monica, Calif.: The RAND Corporation.

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

Rosenbaum, A.L., W.E.K. Lehman, K.E. Olson, and M.L. Holcom 1992 Prevalence of Substance Use and Its Association with Performance Among Municipal Workers in a Southwestern City. Unpublished manuscript, Institute of Behavioral Research. Texas Christian University, Fort Worth.

Rouse, B.A., N.J. Kozel, and L.G. Richards, eds. 1985 Self-Report Methods of Estimating Drug Use: Meeting Current Challenges to Validity. NIDA Research Monograph No. 57. Rockville, MD.: National Institute on Drug Abuse.

Salyards, S.D. 1993 Preemployment drug testing: associations with EAP, disciplinary, and medical claims information. PharmChem Newsletter 21(1):1-4.

Salyards, S.D., and J. Normand 1991 Methodological issues in the evaluation of employee drug-testing programs. In M.J. Burke, chair, Drugs in the Workforce. Symposium conducted at the 15th Annual Convention of the International Personnel Management Association Assessment Council, Chicago.

Sheridan, J.R., and H. Winkler 1989 An evaluation of drug testing in the workplace. In S.W. Gust and J.M. Walsh, eds., Drugs in the Workplace: Research and Evaluation Data. NIDA Research Monograph No. 91. Rockville, Md.: National Institute on Drug Abuse .

Sindelar, J. 1991 Economic cost of illicit drug studies: critique and research agenda. In W.S. Cartwright and J.M. Kaple, eds., Economic Costs, Cost-Effectiveness, Financing, and Community-Based Drug Treatment. Rockville, Md.: National Institute on Drug Abuse.

Taggart, R.W. 1989 Results of the drug testing program at Southern Pacific Railroad. In S.W. Gust and J.M. Walsh, eds., Drugs in the Workplace: Research and Evaluation Data. NIDA Research Monograph No. 91. Rockville, Md.: National Institute on Drug Abuse.

Transportation Research Board 1987 Zero Alcohol and Other Options: Limits for Truck and Bus Drivers. Special report 216. Washington, D.C.: Transportation Research Board, National Research Council. 1993 Alcohol and Other Drugs in Transportation: Research Needs for the Next Decade. National Research Council. Washington, D.C.: National Academy Press.

White, H.R., A. Aidala, and B. Zablocki 1988 A longitudinal investigation of drug use and work patterns among middle-class, white adults. Journal of Applied Behavioral Science 4:455-469.

Zarkin, G.A., M.T. French, and J.V. Rachal 1992 The Relationship Between Illicit Drug Use and Labor Supply. Research Triangle Institute, Research Triangle Park, N.C.

Zwerling, C., J. Ryan, and E.J. Orav 1990 The efficacy of preemployment drug screening for marijuana and cocaine in predicting employment outcome. Journal of the American Medical Association 264:2639-2643.