The Economic Costs of Alcohol and Drug Abuse in the United States - 1992
Chapter 3: Analytical Principles in Estimating the Costs of Alcohol and Drug Abuse

This study of the cost-of-illness (COI) of alcohol and drug abuse relies on and in some cases further develops the analytical principles and methods employed in previous estimates of the alcohol and drug abuse COI. This chapter reviews these analytical principles and describes the general approach used in this analysis. In this, as in other COI studies, the objective is to increase the understanding of the nature and extent of illness and its consequences for society. The data and discussion included in this report could assist analysts and decisionmakers in identifying the variety and extent of the economic effects associated with alcohol and drug abuse; this in turn can help in assigning appropriate priorities to specific interventions.

Synopsis

This section identifies the questions that need to be addressed in the design and performance of an alcohol and drug abuse COI study and indicates the general approach to developing the necessary answers. Essentially, a COI study must answer the following three questions:

• What adverse outcomes are associated with alcohol and drug abuse?
• What is the degree of causality between alcohol and drug abuse and these outcomes?
• What economic values ought to be assigned to the consequences?

Studies of the economic impact of alcohol and drug abuse typically focus on health care costs, effects on productivity and earnings (as a result of illness and premature death), and other costs (such as crime and social welfare). Although better data are available now than was the case in prior studies, determining how much alcohol and drug abuse contribute to specific adverse outcomes is still considerably limited by a lack of empirical evidence. A major effort in this study is to review evidence that is available. Assigning economic values is relatively straightforward, although, again, further research into the sensitivity and specificity of the economic values is warranted.

This study builds on a strong history of prior studies of alcohol and drug abuse and other illnesses and follows guidelines established 20 years ago for the conduct of these studies. A review of the COI literature reveals several strong conclusions. First, most COI studies build on almost 40 years of experience conducting COI studies in health, generally using an opportunity cost framework. Most recent studies have a common understanding of the concepts at the foundation of such efforts. Second, the specifics of how the studies have been performed differ only somewhat in the details. Most of the differences among studies are attributable to differences in the data that have been available for use in constructing the estimates. The specific methodology used or approach taken in developing a component of a COI estimate is often dictated by the data available to the analyst. Finally, COI studies do not provide sufficient justification for any particular social initiatives directed at alcohol and drug abuse. COI estimates do provide insights to help answer questions such as the following:

• What types of health care services - and in what amounts - are required to treat alcohol and drug abuse and related health consequences? How much do these services cost?
• How many people die as a result of alcohol and drug abuse, and what is the economic impact of these premature deaths?
• What effects do alcohol and drug abuse have on individual productivity in the home and workplace?
• How much crime is due to alcohol and drug abuse, either by definition (e.g., drug trafficking), requirements for money (e.g., robbery), or physiologic effects (e.g., assaults)? What does it cost to protect against these crimes, adjudicate arrests, and punish offenders?
• How much reliance on the social welfare system is caused by alcohol and drug abuse, and at what cost?
• What are the economic dimensions of other effects of alcohol and drug abuse, such as motor vehicle crashes and fire destruction?

The reader should be aware that no single source of information encompasses the variety of costs included in this report. Hundreds of sources of information were compiled and analyzed to guide the selection of appropriate causal factors and economic values. This approach was adopted to provide a broad basis for the estimates even though it also necessarily involves a degree of imprecision. As a result, this study presents estimates rather than precise measures of the economic costs of alcohol and drug abuse. In addition, this approach renders it impossible in a number of instances to obtain some of the usual statistical indicators by which estimates are evaluated (e.g., standard errors). Nevertheless, the estimates reported here are based upon the most accurate and up-to-date information available at the time that this study was conducted.

The next sections discuss the history of alcohol and drug abuse COI studies and draw from them the framework that has been applied in this study. This is followed by a summary of data sources and their strengths and limitations and, finally, by a discussion of how the burden of the costs of alcohol and drug abuse is distributed.

A Short History of Cost-of-Illness Studies

COI studies in general, and those for alcohol and drug abuse in particular, are not newly invented, and the general approach to such studies is well established. In fact, this study, as well as the past several studies, has followed guidelines created by the U.S. Public Health Service expressly to help standardize the methods used in assessments of the costs of many types of illness.

Prior Studies on Alcohol and Drug Abuse

COI studies on alcohol and drug abuse have been conducted for almost 25 years and have built on an additional 15 years or more of COI studies on various health problems. There has been a steady evolution of this literature in the United States, along both theoretical lines and, especially, methodological lines. As the methodology has advanced, there has been greater consistency and comparability across studies (as well as a demand for the same). The general approach to these studies could now be considered nearly standardized, based on the application of well-understood principles of economic valuation.

All of the studies of alcohol and drug abuse were preceded by COI studies of other types of illnesses and problems. These prior efforts include a study in Great Britain of road crashes (Reynolds 1956), a study in the United States estimating the costs for each of the internationally recognized major classes of illnesses (Rice 1966), and a U.S. study of mental illness (Fein 1958). Studies prior to these examined various components of what have come to constitute COI studies.

Perhaps the earliest study focusing on the economic costs of alcohol abuse was conducted for Australia in 1969—70 (Pritchard 1971). In the United States, the earliest COI studies date from 1973, including Berry and Boland's (1973) study of alcohol abuse and A.D. Little Inc.'s (1973) study of drug abuse. Since that time, a modest number of studies on the economic costs of alcohol abuse and alcoholism were conducted in other nations. Virtually all of the earlier studies of the abuse of drugs have been conducted in the United States. However, the most recent and comprehensive alcohol and drug abuse COI studies have been conducted in Australia (Collins and Lapsley 1991, 1996), the United States (Rice et al. 1990), and Canada (Single et al. 1996).

The alcohol and drug abuse COI studies performed since 1970 have had the explicit objective of demonstrating that a nontrivial amount of expenditures on health (and morbidity and mortality) are due to alcohol and drug abuse, either directly or indirectly. It should be noted that the study by Fein (1958) did identify and estimate certain "mental health" costs that were attributed to alcoholism and drug addiction and acknowledged certain other nonmedical costs (e.g., criminal justice system costs) that were not estimated because of time limitations.

Underlying the previous studies of alcohol and drug abuse, COI is the premise that an illness results in "costs" when resources are redirected as a result of that illness from purposes to which they otherwise would have been devoted. The next section discusses this point in depth.

The Guidelines and the Opportunity Cost Principle

Costs to society occur when resources are directed away from uses to which they would have otherwise been put or when the total amount of goods and services available to be consumed is reduced.

A central concept in market economics is "opportunity cost," and the studies reviewed employ this basic concept of cost. Opportunity cost is the dollar value of resources that are redirected away from uses to which they would have otherwise been put.

The opportunity cost concept in health studies has been refined throughout numerous studies performed since the 1950's. It achieved its broadest level of application in the United States in a series of COI studies associated with Dorothy Rice. The general approach to this cost methodology in the United States was codified by a task force of the U.S. Public Health Service (PHS) chaired by Dorothy Rice (see Hodgson and Meiners [1979, 1982] for the Guidelines report and a second article that elaborates on the Guidelines). The task force was convened in 1978—79 for the purpose of developing guidelines for COI studies performed or funded by the U.S. PHS. The Guidelines were intended to reduce methodological differences between COI studies performed for different illnesses or for studies of the same illnesses performed by different research teams. It is important to note that the Guidelines did not explicitly contemplate or address the conceptual or methodological challenges involved in constructing COI estimates for alcohol and drug abuse. The principles are general in nature, however, and certainly have been found to be applicable to the alcohol and drug abuse COI studies performed since the Guidelines were developed.

The U.S. PHS Guidelines did not actually set standards of practice or break new theoretical or methodological ground. Instead, the Guidelines recognized and described the mainstream of COI practice (as it developed internationally), noted its strengths and weaknesses, and acknowledged the orthodoxy of this approach. Most alcohol and drug abuse COI studies developed in the United States since the Guidelines were published have been generally based on its approach.

The approach endorsed in the U.S. PHS Guidelines accordingly does not address the conceptual issues associated with any enjoyment or benefit that consumers derive from the use of psychoactive substances. Nor does this approach deal with the fact that resources used to address the problems of psychoactive substances are creating new jobs - actually different jobs, since the funds are taken away from other uses that could themselves create jobs. Also, the theoretical economic constructs of "consumers' surplus," marginal utility analysis, and social welfare functions are beyond the scope of this analysis.

There are, of course, other types of economic studies that include components of the costs of alcohol and drug abuse - for example, the costs and/or benefits of particular types of interventions. These additional types of economic evaluations are described in the next section.

The Framework for Cost-of-Illness Studies

The construction of COI estimates for alcohol and drug abuse can be enormously challenging and complex; however, the framework for the study is relatively clear and easy to understand. The design and performance of these studies need to be built around a set of objectives and standards that have scientific validity and can be communicated to the concerned scientific community, policymakers, the media, and the general population. As noted above, this can be accomplished with the following three-step process:

• Identify the adverse outcomes associated with alcohol and drug abuse;
• Document and quantify the degree of causality between alcohol and drug abuse and the associated adverse outcomes; and
• Assign economic values to the adverse outcomes of alcohol and drug abuse.

Although accomplishing these steps can require significant time, resources, and skill, keeping these objectives and questions in mind during the course of the study should facilitate the effort and ultimately make the final product more accessible. The following sections discuss each of these steps in turn.

Adverse Outcomes Associated With Alcohol and Drug Abuse

The initial step in designing an alcohol and drug abuse COI study is to identify the tangible negative outcomes that are believed to result from alcohol or drug abuse in the area under study. Identification and definition of these outcomes provide a framework for the analysis. This makes it immediately possible to segment the work and to organize multiple and parallel research efforts.

Prior studies based on the PHS Guidelines differentiated "direct" from "indirect" and "core" from "noncore" costs. Direct costs refer to the value of tangible goods and services actually delivered to address consequences of the illness. Indirect costs represent the value of personal productive services that are not performed because of the consequences of the illness. Core costs refer to health care costs and other health-related consequences, and noncore costs refer to costs not manifested through the health consequences of the illness. For example, alcohol and drug abuse treatment were "core, direct" costs. Foregone earnings due to premature death from a drug overdose were a "core, indirect" cost.

This study replaces the terminology used in previous studies but preserves the distinctions between health and nonhealth costs and between out-of-pocket versus foregone earnings. The following are major categories of "consequences," represented in this report as separate chapters:

• Health care costs. These include the costs of treating alcohol and drug abuse as well as illnesses or injuries brought on by alcohol or drug abuse.
• Productivity losses. These include foregone earnings during 1992 among alcohol and drug abusers who suffer from illness or are in prison or jail and among persons who are victims of crime or accidents related to alcohol or drug abuse. Productivity losses also include the foregone lifetime earnings of individuals who died prematurely in 1992.
• Other effects on society. These include nonhealth costs associated with drug and alcohol abuse, such as the costs of crime and criminal justice, social welfare, motor vehicle crashes, or fires.

The outcomes of concern should be quantifiable both in their incidence and in the level of resource use associated with their occurrence. For example, the incidence of particular health problems and utilization of health services is tracked through nationally representative surveys. Mortality statistics are also compiled regularly by the U.S. National Center for Health Statistics, with deaths classified by their primary cause and appending data on age and gender of the decedent. Other data systems track other outcomes, such as criminal activity, property damage, and motor vehicle crashes. However, intangible outcomes, such as pain and suffering, generally are excluded from COI studies because they have not been reliably quantified.

A general set of categories of outcomes associated with alcohol and drug abuse is presented in table 3.1. Each of the items is actually a category comprising further subsets of outcomes - some of which can be quite extensive - that are analyzed later in this report. For example, the potential health consequences of alcohol and drug abuse (by which is meant the health problems that can be caused by alcohol and drug abuse) are quite extensive. There are, of course, other consequences of alcohol and drug abuse, many of which could not be estimated in this report. Table 3.1 refers to several of these consequences.

Table 3.1 mentions that "support" for specialty treatment is included as a cost. Support costs include training, research, and insurance administration. Some experts argue that training and research costs are investments in the Nation's health care system and do not represent a "cost" per se. This study only applies training and research costs as they pertain to the treatment and prevention of alcohol and drug abuse (and not to the health care system as a whole). This is because the training and research programs are specific to the alcohol and drug abuse fields and are not necessarily exportable to the rest of health care. Without alcohol and drug abuse, these investments would not be necessary.

In estimating the costs of alcohol and drug abuse for a particular year (the "base year"), two general approaches are available: use of incidence versus prevalence rates. The important distinction in these approaches has to do with the period in which consequences to be counted occur:

• Prevalence-based estimates count all costs incurred in the base year that result from instances of alcohol or drug abuse originating any time during or before the base year.
• Incidence-based estimates count all costs incurred any time during or following the base year that result from instances of alcohol or drug abuse originating during the base year.

This study, like previous COI studies of alcohol and drug abuse, relies primarily on prevalence-based estimates. That is, this study estimates the consequences - and costs - incurred in 1992 as a result of alcohol or drug abuse that occurred any time during or prior to 1992. For example, this approach tallies all health care costs in a year that are associated with AIDS cases caused by drug abuse (where the "incidence" of needle use very likely occurred in a previous year).

A different approach would be to count the "incidences" during the year that are caused by alcohol or drug abuse - for example, all new HIV infections. Calculation of costs using this approach would generally rely on a determination of the net present value (NPV) of costs associated with the consequences during the expected lifetime of the case from infection until death. Using the example of drug abuse-caused AIDS cases, the NPV of the costs of care for persons with HIV/AIDS would be applied to the incidence, or number of people who contracted HIV during the year.

The case of premature deaths is, however, a "hybrid" case. We first estimate the number of alcohol- and drug-related deaths that occur in the base year regardless of the onset of alcohol or drug abuse. The value assigned to those deaths is calculated as the present value of expected lifetime earnings that would have been generated by those individuals.

Causality

Estimating the role of alcohol and drug abuse as causal factors for various adverse outcomes is a central issue in any study estimating the COI for alcohol and drug abuse. Upon determining the plausible consequences of alcohol or drug abuse, the analyst must assess and quantify the extent to which alcohol or drug abuse may have caused the specific consequences.

Theoretical Aspects of Causality

Establishing causation has been the topic of considerable theoretical discussion and applied research for decades. The most fundamental concern underlying COI studies is struggling over the distinction between association and causation. Researchers have identified at least three basic conditions that are required before one can infer that an association represents a causal relationship (Berry 1984; Austin and Werner 1974). They are as follows:

• Strong and consistent correlation or covariance between phenomena;
• A coherent logic to the causal linkage, including correct temporal ordering; and
• Elimination of alternative possible causes.

Although the first two requirements are more straightforward, the third presents analysts with the greatest challenge. It is impossible to eliminate all potential other causes; moreover, a "cause" may have precursors that undermine the causal connection. For example, research into the etiology of alcohol and drug abuse has identified several individual and environmental factors that contribute to the likelihood that a person will have problems with alcohol or drug abuse - some argue that these are the "real" causes of a person's problems.

In eliminating alternative possible causes, the analyst must also be concerned with the "counterfactual." The counterfactual essentially is the answer to the question: What would be the likelihood that a person would have a particular negative consequence if he or she were not experiencing problems with alcohol or drugs? The following questions illustrate the complexities:

• If a criminal did not have a problem with alcohol or drug abuse, would that person still have difficulties with the law? Underlying factors in a person's personality or environment may contribute to criminal activity.
• Is it appropriate to assume that a person's wage rates were driven lower because of alcohol or drug abuse, or would the same person still experience lower rates even without the alcohol or drug problems?
• Does the fact that a driver consumed alcoholic beverages shortly before a crash mean that alcohol necessarily caused the crash? Many fatal crashes occur late at night, when drivers are tired and often less attentive. Many of these drivers have also been drinking. Simply counting the proportion of crash drivers that had been drinking fails to account for the generally more hazardous driving conditions at that time.

A "causal" estimate should use logic and analysis to estimate the extra risk posed from drinking and driving. However, even with the best data, most theoreticians concede that certain simplifying assumptions will always need to be made (Asher 1983; Austin and Werner 1974).

Moreover, the analyst must recognize that alcohol and drug abuse definitionally result in some consequences, whereas alcohol and drug abuse may directly or indirectly contribute to other consequences. For example, although alcohol or drug abuse does not cause all HIV/AIDS cases, it may play a direct or indirect role in some cases. Needle sharing among intravenous drug users may directly result in the transmission of HIV; alcohol or drug use may result in unsafe sex practices and thereby indirectly result in the transmission of HIV.

The causal role of alcohol or drugs may be very difficult to disentangle, but it is critical to acknowledge the difference between association and causality. The analyst who wishes to make COI estimates for multicausal consequences will have to evaluate the adequacy of empirical data that differentiates association and causation and decide whether to apply an explicit "judgment factor" in adjusting measures of association into an estimate of causality.

Practical Aspects of Establishing Causality

This study has assembled information that reviews and justifies the estimate(s) of how much of given consequences is caused by alcohol and drug abuse. As noted previously, some consequences of alcohol or drug abuse are definitional or are made by an external source. For example, some health diagnoses have been defined as being caused by alcohol or drug abuse (e.g., alcoholic cirrhosis or withdrawal from heroin). In such cases, analysts need only tally the number of diagnoses for such conditions treated in hospitals or listed as causes of death in mortality statistics. Other examples include criminal offenses that are defined as being related to alcohol and drugs. For some consequences, administrative determinations may be made about the role of alcohol or drugs. This is the case for some disability programs in the United States. Again, for such cases, analysts need only look at tallies of cases from administrative data bases.

A more challenging situation occurs when alcohol or drug abuse is one of multiple causes of particular consequences, such as liver disease, certain cancers/neoplasms, motor vehicle crashes, crime of various types, and employment problems. The objective is to develop an estimate of the proportion of the multicause consequence that can be attributed to alcohol and drug abuse. To accomplish this, we conducted literature reviews for evidence concerning the causal relationships between factors and interviewed experts where necessary and appropriate. In some cases, multiple estimates of causality were available, which were synthesized (generally by selecting a midpoint value).

This study relies on research into the causal role played by alcohol in mortalities to estimate certain hospital care and ambulatory medical care costs. This research into the underlying causes of death (beyond the listed diagnosis) is well documented and supported. Some experts have argued that mortality data are not appropriate for use in estimating health care utilization among the living. However, at face value, the proportion of deaths for a given condition that are caused by alcohol need not be different from the proportion of hospital episodes for the same condition. If alcohol caused 75 percent of malignant neoplasms of the esophagus that resulted in death, it seems plausible to assume that the same proportion of hospital episodes for that condition also were caused by alcohol abuse. However, using attribution factors based on alcohol's role in mortality from various causes to infer alcohol's role in generating morbidity in various diagnostic categories probably imparts a conservative bias to the resulting estimates. This is because such attribution factors do not reflect the causal role of alcohol in other, nonfatal conditions (or in diseases that may be present at death but that are not the cause of death). This is an area where further empirical evidence would be useful.

This study determined that for some potential consequences of alcohol or drug abuse, compelling empirical evidence regarding the role of alcohol or drug abuse is not available. Two examples are (1) the role of alcohol or drug abuse as a causal factor in other mental disorders among the dually diagnosed populations (summarized in this report but not included in the totals) and (2) the role of cocaine use in contributing to heart disease (not included in this report).

One of the most difficult issues around causality is the question of whether abuse of alcohol and drugs cause subsequent mental illness or whether preexisting mental illness makes people more susceptible to abusing drugs and alcohol, perhaps in an attempt to "self-medicate" their problems. This is one of the major and largely unresolved epidemiological (and clinical) issues involved with alcohol and drug abuse.

Although prior studies simultaneously developed separate estimates for alcohol abuse, drug abuse, and mental illness, this effort developed estimates only for drug and alcohol problems. In fact, many of the costs estimated in this study occur to individuals who are afflicted with multiple alcohol, drug, and mental disorders. A review of the literature indicates that up to 75 percent of persons with drug problems also have problems with alcohol abuse and/or mental disorders and up to 50 percent of persons with alcohol problems have problems with drugs and/or mental disorders. In contrast, among the population with mental disorders it appears that about 20 to 30 percent have problems with alcohol and/or drugs. How persons are classified with respect to these disorders has a substantial impact on the cost estimates - the impact of the methodology choice is to shift costs among the alcohol, drug, and mental disorders, although this may be more based on data limitations than on the nature of the patients and disorders presenting for treatment.

Researchers in this study chose middle-of-the-road methodologies that split the difference between attributing particular costs solely to mental illness or to alcohol and drug abuse. In the most recent study by Rice et al. (1990) - and to a certain extent in earlier studies - methodologies were used that assigned all of selected costs to mental illness. This has primarily been an issue in estimating health care expenditures, although it is also highly pertinent to understanding the effects of alcohol and drug problems on success in the labor market. Failure to recognize and adjust for co-occurrence of mental disorders probably leads to an overestimate of the costs of alcohol and drug problems, unless the mental disorder has been caused by alcohol or drug abuse.

Cost or Valuation Techniques

The final step in a COI study is to assign values to the consequences and the associated flows of resources. In some cases, such as with goods and services provided by private or public institutions, this valuation is straightforward. In other cases, such as with the value of lost productivity resulting from illness or premature death, this involves more complicated theoretical reasoning and data.

Valuation of Goods and Services

Assigning values to goods and services simply requires that the analyst know the average costs involved. For example, the per diem expenditure in a hospital in 1992 was $816 (National Center for Health Statistics 1996a). If one-half of the episodes of a medical condition are caused by alcohol abuse, then one-half of the annual days of care for treatment of that condition could be allocated to alcohol abuse, at the cost of $816 per day. The most appropriate measures for this purpose are prices that represent the unit cost of purchasing, producing, or replacing the resource flow that has been measured. Average or typical charges for services may be substituted when actual costs are not available, under the premise that in the long run, charges will equal costs.

Expenditures for health and nonhealth goods and services are generally straightforward to value, particularly where resources and services are exchanged in a market. Average price/cost data are usually available for health care services, such as a day of hospital care or a visit to a physician. It is generally acceptable to use such average prices; however, if data are available that can provide more differentiated costs, they should be used. Examples of such data include days and prices for alcohol and drug abuse-related hospital intensive care versus general hospital ward and visits to medical specialists versus general practitioners.

Other publicly provided services, such as criminal justice services and highway safety efforts, often do not have good data on prices and costs that can easily be allocated to alcohol and drug abuse. However, there is generally a total budget for these public services. If a total budget is available, an estimation technique must be developed that divides the budget between alcohol and drug abuse-related efforts and other efforts. For example, one approach to the valuation of police protection is to estimate the proportion of arrests that is attributable to alcohol and drug abuse and then apply this proportion to the total budget for police protective services.

Valuation of Lost Productivity

Economic costs have been estimated using the "human capital" approach to valuation of loss of expected productivity that a person would have achieved in the study year (for morbidity) or over the person's expected course of life (for mortality). The human capital approach to calculation of productivity costs has been generally used in COI studies (Hodgson and Meiners 1982) and in the prior studies of alcohol and drug abuse (exceptions are Miller et al. 1997a,b). Using this approach, a dollar value is assigned to production that is lost because of illness, impaired functioning in the labor force, and premature death. This dollar value is based on data regarding actual earnings found among alcohol and drug abusers and earnings of comparison populations not affected by alcohol or drug abuse using age- and gender-adjusted national data.

The human capital approach is in contrast to the "willingness-to-pay" approach, which attempts to reflect both expected loss of productivity (the essence of the human capital approach) as well as the "value" of pain and suffering. To simplify the difference between human capital and willingness-to-pay approaches, willingness-to-pay counts the value that a person (typically, the alcohol or drug user or, potentially, his or her family) would place on not experiencing impairment or death from alcohol or drug abuse. Cost-benefit studies in highway safety and pollution control use the willingness-to-pay approach (Miller et al. 1991). This approach has not been directly adapted and applied to alcohol and drug abuse.

The human capital approach involves several technical issues, particularly adjustments to average wage rates. First, average wage rates should be adjusted for several factors, including the expected level of employment and the value of "fringe" benefits and taxes that may never appear in a paycheck. Fringe benefits and taxes should be included because they are part of the contribution of labor's productivity. To stay in business, employers have to recover these expenses from the productivity of workers. Also, average wage rates should be adjusted for both age and gender. This is valuable because alcohol abuse and drug abuse are not uniformly distributed across demographic groups; they tend to be concentrated among young males. Finally, the rates should also be adjusted to reflect some value to the household for productivity outside of the workforce.

It could be argued that productivity/earnings estimates should be adjusted down to account for expected living expenses and affected persons' consumption of their earnings. However, this argument implies that consumption by these affected persons does not matter to society, and that the social welfare function only values the surplus of earnings that is passed on to savings, family and friends, and taxes. Although this latter approach may be correct for valuation of life insurance, it is not the approach that has been taken in most COI studies, and no such adjustments have been applied in this study.

Moreover, some of the economic consequences that initially fall on alcohol and drug abusers are redistributed to other persons through insurance mechanisms of different types, including health insurance and unemployment and disability insurance. Even when lost or reduced earnings are not compensated through transfer payments, the impact may be spread beyond alcohol and drug abusers to their family and friends. Researchers need to be very careful in how they work with transfer payments that redistribute resources.

Transfer payments (e.g., the cash value of food stamps and payments from Aid to Families with Dependent Children and other welfare/general assistance programs) are not added into total cost estimates but, nonetheless, are shown as "redistributions" from one segment of the Nation to another (alcohol and drug abusers). There is no net loss from society's perspective, although taxpayers and public decisionmakers may see this quite differently. The primary reason for excluding transfer payments from total cost estimates, however, is that "double counting" may arise if both the initial cost of the programs and the value of lost productivity (which social welfare programs are designed to ameliorate) are counted. The cost of administering transfer payment systems is, however, a real cost.

Contention remains regarding at least two particular issues in application of human capital estimates to alcohol and drug abusers: (1) the expected future course of productivity and (2) the appropriate discount rate to use in the calculations.

• The expected future course of productivity. This term represents the "counterfactual" wage and earning potential for an alcohol or drug abuser. Counterfactual in this context refers to what would plausibly be the case if alcohol or drug abuse were suddenly eliminated or had never existed. It is possible that a person with an alcohol or drug abuse problem may have held (and lost) lower wage jobs than the general population. The question is whether such a factor would justify using values lower than those for the general population in estimating future productivity. The rationale for not making such an adjustment or reduction is that if alcohol or drug abuse caused the initial/current deficit relative to expected productivity, then future deficits will probably also be caused by those problems. Although it is likely that they would have had lower future productivity, this would be caused by and attributable to alcohol and drug abuse, and making the adjustment would miss a material impact and cost that should be counted.
• Selecting an appropriate discount rate for human capital estimates. This is a contentious issue in estimating the value of productivity effects (Burkhead and Miner 1971). A high discount rate gives a lower present value of expected earnings, and a low discount rate gives a higher present value of expected earnings. There is extensive debate about what rate is appropriate for what kinds of studies. Higher rates (such as 8 to 10 percent) are used when there is more risk or uncertainty associated with a stream of values, whereas lower discount rates (2 to 4 percent) are used when there is less risk or uncertainty around events. Also, higher rates give preference to current benefits and consumption, and lower discount rates give more emphasis on future consumption and benefits. The major prior alcohol and drug studies have used a value in between - 6 percent - and the primary justification for using this value may be to allow comparability with previous estimates. However, it appears that lower values, such as 3 percent, are being used in much cost-benefit work for the Office of Management and Budget. This study reports costs using 3 percent, 4 percent, and 6 percent and uses the more conservative 6-percent figure (that is, it gives lower cost estimates) in constructing and reporting the total estimates.

Using the human capital approach, deaths of males and females ages 35 to 39 represent costs of about $700,000 and $500,000, respectively, using a 6-percent discount rate. Using a 3-percent discount rate would yield costs of about $1 million and $700,000, respectively. This clearly illuminates two issues inherent in the human capital values: (1) males tend to have higher values than females, and (2) there is continuing debate over appropriate discount rates.

Actually, much higher costs would result from application of the willingness-to-pay approach. A review of the willingness-to-pay literature found an average value per death across 47 studies of $2.3 million in 1989 dollars (cited in Miller et al. 1995). Adjusted for 3 years of inflation (about 12 percent to $2.55 million per death), this approach would indicate losses of $275 billion for alcohol-related mortality and $61 billion for drug-related deaths. These values are about nine and four times greater, respectively, than the estimates obtained using human capital values reported elsewhere in this volume. Both the application of the technique and the estimated values should be given further consideration. It was found that for smoking behaviors, there were substantially lower willingness-to-pay values. Therefore, the questions must be asked, "Would the willingness-to-pay to save the life of alcohol and drug abusers (by themselves and/or by others) theoretically and actually be different from values for other types of risk?" and "How would these values compare with other willingness-to-pay values?"

Sources and Limitations of Data

The ability to develop more or less detailed COI estimates is ultimately determined by the availability of data. This study has utilized dozens of different data sets and previous investigations. Priority is given here to studies that are broadly representative and, at the same time, have the design and content to support the very specialized analyses this project requires. Care is taken in the body of this report to identify all data sets and studies that have been used as inputs. Furthermore, this report provides descriptive information about data sets and studies to give the reader an informed sense of the strengths and limitations of the data used. However,

providing comprehensive assessments and documentation for each data set and study that has been used in the present study is not possible.

This study benefits from a decade of improvements in the quality of data and advancements in areas for which data are available. However, the availability and quality of data remain, unfortunately, uneven. Simply put, there is no single source of data that addresses the numerous health and related sequelae of alcohol and drug abuse. As a result, this study has sought out many complementary sources of information that include the following: general-purpose surveys, administrative data sets, special studies, and expert judgment.

The most basic data come from general-purpose surveys and administrative data sets that often have limited information about alcohol and drug abuse. Large surveys used here are generally representative of the United States but often do not include certain data elements about alcohol and drug abuse and its role in the issue under investigation in the survey. Administrative data provide useful information about resources and services that complements survey data, particularly regarding services provided in the public sector (e.g., social welfare). Special studies (e.g., studies of clinical populations in a few treatment settings) are useful for exploring intricacies but are not designed to represent the United States. Nonetheless, all such data are extremely valuable, because they provide both bases from which estimates of alcohol and drug abuse-related costs can be derived and - if good, specific data bases on alcohol and drug abuse exist - standards against which estimates can be compared. Table 3.2 lists examples of surveys and administrative data sources used or analyzed in this report.

One type of data that is not used in this report is health care claims data. Health care claims data would provide an alternative to survey measures of specialty treatment services and for all types of costs associated with medical consequences. However, many alcohol and drug abusers do not have health insurance and, accordingly, no claims records exist for care they receive. Moreover, the survey data that are used provide a nationally representative picture of health care. Claims data are idiosyncratic to the insurers or employers that have provided the data. Nonetheless, claims data may constitute a potential source of information for future analyses of special topics.

The most critical data for alcohol and drug abuse COI studies are data concerning the relationship of alcohol and drug abuse to the consequences of concern. These data are likely to vary greatly in their utility and quality and in the approach used to obtain them. The most basic studies will be those that simply examine and report on the involvement of alcohol and drug abuse in the consequence of concern (Did the victim of throat cancer drink? Had the driver of the crashed vehicle consumed alcohol before driving?). Even studies with such limited objectives can be performed more or less rigorously, based on factors such as the nature of the sample (representative or purposive sample), size of the sample, and how the data were collected (self-report or from tests, such as blood alcohol or urinalysis).

However, studies that move beyond association toward causation are the most critical for the credibility of this study. As discussed previously, alcohol and drug abuse are behaviors that often occur in combination with other problems, most notably risk-taking or criminal activities. This suggests the possibility and, indeed, the likelihood that alcohol and drug abuse and certain negative consequences may be partially related to other underlying factors, values, or predispositions.

Studies of causality, therefore, need to use measures of alcohol and drug abuse behaviors that may be logically related to the consequence at hand. In the case of motor vehicle crashes and other accidents, the standard measure is usually blood alcohol concentration. Use of such measures must be supported by meaningful scientific literature, such as studies demonstrating that physical coordination and the ability to perform certain tasks (mental or physical) are likely to be progressively impeded above specified blood alcohol concentration values.

Studies of alcohol and drug abuse effects on worker productivity should look beyond simple patterns of consumption and toward signs that alcohol and drug abuse is chronic and severe, to the degree that it regularly affects individual lives and a person's ability to function in the home, in the workplace, and in society at large. Such patterns of problems are integral to making clinical diagnoses of alcohol and drug abuse and dependence, so studies of the relationship of alcohol problems to employment and productivity should attempt to use clinical criteria for abuse and dependence if the specific data base has sufficient detail and quality of information.

The need to use clinical criteria becomes particularly clear given that some studies have detected a positive relationship between earnings/employment and alcohol consumption. For example, greater proportions of higher income populations tend to be consumers of alcoholic beverages than is the case among relatively lower income populations. Such a finding does not necessarily contradict the hypothesis that abuse of alcohol and drugs is likely to have a negative impact on employment and/or earnings. Instead, clinical diagnoses are needed to differentiate persons for whom alcohol consumption has led to detrimental symptomatology that adversely affects employment and/or earnings. Analysis of alcohol and drug abuse and employment will need to recognize and model or control for both a clinical phenomenon associated with pathological consumption and the more pervasive "consumption-function" relation. Without modeling these complex relationships, studies will probably yield inconsistent results.

Conclusion

COI studies serve to summarize and highlight the many disparate dimensions of alcohol and drug abuse. Data derived from these studies are considered quite valuable to both public figures and private advocates who are working to address the problems of alcohol and drug abuse. COI studies also help place the problems of alcohol and drug abuse into perspective. For instance, decisionmakers may be more interested from time to time in specific aspects of health costs, deaths, or other disease sequalae - all of which can be informed by COI studies.

Still, COI studies cannot tell us how much to spend on prevention and treatment. The real burden in using economic logic to justify social policy is on the analysts and advocates of prevention and treatment services to document that these services can be, and are, effective. If these services are not effective, they can never be cost-effective or cost-beneficial. Knowing only that there is a massive cost associated with the abuse of alcohol and drugs is insufficient justification for any particular policy without knowing the potential effectiveness and cost of the proposed policy intervention.

Table 3.1: Consequences Assessed in Cost-of-Illness Analyses

Table 3.1: Consequences Assessed in Cost-of-Illness Analyses

Value of Goods and Services	Value of Lost Productivity	Generally Nonquantifiable Costs
Health
Specialty drug/alcohol treatment and prevention Support for specialty treatment, including training, research, and insurance administration Health consequences of alcohol and drug abuse, including hospital care, physician services, nursing home care, and pharmaceuticals, or the continuum of services for certain special disease categories such as HIV/ AIDS, fetal alcohol syndrome, drug-exposed infants and boarder babies, hepatitis, and tuberculosis	Reduced or lost earnings while impaired or unemployed Lost earnings due to premature death or to institutionalization	Pain and suffering Bereavement Psychosocial development impairment among alcohol and drug abusers and their children Familial health Out-of-pocket costs other than deductibles and copays, such as transportation, child care, and other factors associated with health care use.
Other (Non-health)
Criminal justice system expenses, including protection, adjudication, and corrections Victim expenses Crime-related property destruction Administration of income transfer programs Motor vehicle crashes Fire destruction	Lost earnings while crime victims cannot work Lost earnings while criminals are incarcerated Lost legitimate earnings, including lost tax dollars due to "careers of crime"	Reduced product quality Secondary market effects Productivity consequences for family members Productivity consequences for coworkers and firms that are not reflected in the earnings of alcohol and drug abusers

Table 3.2: Selected Data Sources Used in This Report

Table 3.2: Selected Data Sources Used in This Report

Data Source	Period Covered by Data	Reference
Budget of the United States Government	1992	Office of Management and Budget (1993)
Drug Use Forecasting	1992	U.S. Department of Justice (1994d)
Fatal Accident Reporting System	1992	Blincoe and Faigin (1992)
Inventory of Mental Health Organizations	1990	Manderscheid and Sonnenschein (1994)
Justice Expenditure and Employment	1990	U.S. Department of Justice (1992b)
Mortality Data	1992	National Center for Health Statistics (1996b)
National Ambulatory Medical Care Survey	1992	National Center for Health Statistics (1994a)
National Crime Victimization Survey	1990	U.S. Department of Justice (1993c)
National Drug and Alcohol Treatment Unit Survey	1991	Substance Abuse and Mental Health Services Administration (SAMHSA; 1993a)
National Hospital Discharge Survey	1990-1992	National Center for Health Statistics (1993, 1994b)
National Household Survey on Drug Abuse	1974-1995	SAMHSA (1993b, 1994, 1995, 1996a,b)
National Longitudinal Alcohol Epidemiologic Survey	1992	Grant and Dawson (1996)
National Nursing Home Survey	1992	National Center for Health Statistics (1991)
Federal Drug Control Budget	1992	Office of National Drug Control Policy (1995)
State Alcohol and Drug Abuse Profile	1992	National Association of State Alcohol and Drug Abuse Directors (1994)
Survey of Prisoners and Inmates	1991	U.S. Department of Justice (1994e)
Uniform Crime Reports	1992	U.S. Department of Justice (1994b)