New methods and data sources for measuring economic consequences of workplace injuries.код для вставкиСкачать
AMERICAN JOURNAL OF INDUSTRIAL MEDICINE 40:452±463 (2001) New Methods and Data Sources for Measuring Economic Consequences of Workplace Injuries Robert T. Reville, PhD, Jayanta Bhattacharya, Lauren R. Sager Weinstein, AB MD, PhD, and Background Evaluation of programs and policies to reduce the incidence of workplace injuries require that the consequences of injury are estimated correctly. Because workplace injuries are complex events, the availability of data that re¯ects this complexity is the largest obstacle to this estimation. Methods We review the literature on the consequences of workplace injuries for both workers and employers, focusing on data sources, particularly linked administrative data from different public agencies. We also review other approaches to obtaining data to examine workplace injuries, including public-use longitudinal survey data, primary data collection, and linked employee-employer databases. We make suggestions for future research. Results Recent advances in the literature on the economic consequences of workplace injuries for workers have been driven to a great extent by the availability of new data sources. Much remains unexplored. We ®nd longitudinal survey databases including the National Longitudinal Survey of Youth, and the Health and Retirement Survey, to be very promising though largely untapped sources of data on workplace injuries. We also ®nd that linked employee-employer databases are well suited for the study of consequences for employers. Conclusions We expect that new data sources should lead to rapid advances in our understanding of the economic consequences of workplace injuries for both workers and employers. Am. J. Ind. Med. 40:452±463, 2001. ß 2001 Wiley-Liss, Inc. KEY WORDS: economic consequences; wage losses; employer costs; administrative data INTRODUCTION Workplace injuries and illnesses impose signi®cant economic costs upon society. Fatalities are the most dramatic measure of the losses from workplace injuries; RAND, 1700 Main Street, PO Box 2138, Santa Monica, California 90407-2138 This paper was first presented at the NIOSH conference, ``Functional, Economic, and Social Outcomes of Occupational Injuries and Illnesses: Integrating Social, Economic, and Health Services Research'', Denver, CO, June13^15, 1999. *Correspondence to: Robert T. Reville, RAND, PO Box 2138, Santa Monica, CA, 90407-2138. Accepted 22 May 2001 ß 2001Wiley-Liss, Inc. according to the Bureau of Labor Statistics's (BLS) Census of Fatal Occupational Injuries (CFOI), 6,218 fatalities occurred at work as a result of an injury in the US in 1997. Mortality from occupational illnesses are much harder to measure, as discussed in Leigh et al. , who estimates over 60,000 deaths in 1992. However, fatal injuries and illnesses are a minute fraction of all workplace injuries. In 1997, according to the BLS Annual Survey of Occupational Injuries and Illnesses, 6.1 million workplace injuries and illnesses occurred nationally, and 2.9 million of these resulted in time out of work (lost-time injuries). All of these injuries potentially impose costs on employers, workers and their families, or both. But how large are the Economic Consequences of Workplace Injuries costs and how does one go about measuring them? These questions are fundamental to any policy evaluation, since economic costs are a key outcome measure. In this discussion, we describe some analytical and methodological issues, new data sources and new approaches to constructing databases for the estimation of the consequences of workplace injuries. A recent study by Leigh et al.  discusses many of the methodological issues in estimating incidence and costs of workplace injuries and data limitations with many of the commonly used databases. We do not attempt to repeat that information here. Our purpose is to describe new methods and data sources, rather than to review comprehensively the older literature and data. We summarize data sources available for workplace injury research, discuss the estimation of losses to workers, focusing on the growing literature on wage loss due to injury and describe the estimation of employer costs, noting the paucity of research in the area, and ®nally make suggestions for research and data collection. Data for Measuring Losses from Workplace Injuries A number of studies have calculated the total annual costs from workplace injuries by summing and combining multiple data sources or workers' compensation administrative databases. These include Miller [1995, 1997] and Leigh et al. [1997, 2000] the letter includes a detailed review of this literature. These studies rely upon many assumptions and extrapolations from estimates of particular costs faced by workers or employers. The data and methods for the component costs of these extrapolations are often quite ¯awed or limited. As these authors recognize, to the extent these aggregations rely upon poorly estimated component costs, the reliability of the aggregate estimates are also limited. More importantly, accurate estimates of the employer or worker costs are also critical for assessment of the adequacy and equity of workers' compensation bene®ts, evaluations of safety programs, return to work programs, or other interventions. Several types of data sources are available for research on worker and employer costs. Leigh et al.  evaluates (and generally ®nds wanting) several of the better known databases, including the BLS's CFOI, NIOSH's National Traumatic Occupational Fatality Study (NTOF), BLS's Annual Survey of Occupational Injuries and Illnesses, the National Health Interview Survey (NHIS), and the National Council on Compensation Insurance (NCCI) Ultimate Reports. In addition to these data sources, there are a number of less common but arguably more satisfactory alternatives, including primary data collection, public-use survey databases other than the NHIS, and linked administrative data. 453 Primary data collection Several characteristics of workplace injuries complicate primary data collection on costs: * * Workplace injuries are complex events. Losses from workplace injuries may occur over many years or an entire lifetime. Payments of costs associated with the injury may be made by both workers and employers, may change over time, and may be dif®cult to differentiate from costs that would have occurred even if the injury had not occurred. Losses for the same injury may differ by socio-economic and demographic characteristics of the worker, economic conditions at the time of the injury, and characteristics of the employer In terms of annual incidence in the population, workplace injuries are relatively rare events. In 1996, according to U.S. Census Bureau , there were 98 million households in the US. Using the BLS annual incidence estimate, an individual household has an injury propensity for lost-time injuries of about .03 assuming no more than one lost-time injury or illness per household. Consequently, to collect a sample of 300 workers with workplace injuries in a given year, a survey would need to contact 10,000 households. Data for rare events usually require large sample sizes. But survey data for complex events usually need to be detailed, and in order to be affordable, these surveys typically have small sample sizes. For instance, they may require longitudinal data for the same individual, and long questionnaires. The net result is that frequently, surveys with detailed information on certain aspects of rare and complex events may lack information fundamental to analyzing the event. For instance, the NHIS 1994±1995 Disability Supplement, which includes detailed information on the consequences of disability, has very limited labor force information and does not include information about workers' compensation bene®ts received. Primary data collection on the scale required is rarely an option. One study that collected primary data at the household level is Hensler et al. . In this study, 26,000 households were contacted in an initial screener, and 2,770 with accidental injuries (including those in the workplace) were administered a detailed interview. The sampling frame included individuals who experienced injuries over several prior years and at least some costs (medical or time loss) in the previous year. It also included individuals with workplace injuries who did not ®le workers' compensation claims. The population prevalence of workplace injuries was .044. Despite the scale of this project, the number of workplace injuries is still relatively small 454 Reville et al. for many analyses, and only a single cross-section is available.1 Since large-scale primary data collection is rarely an option, most primary data collection on the costs of occupational injuries and illnesses uses data collected from a particular worksite or a limited number of ®rms. The generalizability of these kinds of studies is always a question. Public-use survey data Some surveys collected for general demographic, health, or labor force information have longitudinal data on labor force participation and earnings as well as extensive information on disability and workers' compensation. In general, these databases have been underutilized by researchers interested in occupational health and safety. Given the complexity of the impact of a workplace injury, these databases can provide the information necessary to estimate costs in a much more effective and affordable manner than primary data collection. Perhaps pre-eminent among these longitudinal data sets (though not for occupational injury research) is the Panel Study of Income Dynamics (PSID), which has been collected since 1968. The drawbacks to using the PSID for estimating the costs of injuries to workers are considerable and also shared with many other databases. There is minimal medical information, a relatively small sample size, no ability to relate disability information to a potential occupational source, and a limitation of workplace injury information to workers' compensation bene®ts received (which signi®cantly understates injuries). The only study related to occupational injuries and illnesses that we are aware of using the PSID is Leigh . Other databases are more promising than the PSID. One source of data that has not been frequently tapped for analysis of workplace injuries is the Current Population Survey (CPS), which is nationally representative and has considerably larger sample sizes. As with the PSID, though, injury information is limited to workers' compensation, and disability information is not linked to work-relatedness. Medical information is also limited. In addition, the longitudinal dimension of the CPS is limited to two years. Even with the larger sample, though, sample sizes of occupational injuries are still an issue. For example, Krueger  uses two linked years of CPS data to identify new workers' compensation claims and, therefore, has only 290 workers' compensation claimants. Hirsch et al.  adopt the Krueger approach pooled across fourteen years, yielding 1,594 workers' compensation claimants. 1 The collected data was recently used by Marquis and Manning , and is available for public use, but has not been used in any other studies of which we are aware. The most promising databases are the Health and Retirement Survey (HRS), National Longitudinal Survey of Youth (NLSY), and the Survey of Income and Program Participation (SIPP). The HRS, which provides longitudinal data on a sample of individuals approaching retirement age, provides detailed information on disability, modi®ed work, work-relatedness of disability as well as exposure to workplace hazards. It also includes extensive demographic, ®nancial, and labor market information. To our knowledge, it has not been used for the measurement of the costs of injuries [though it has been used to study risk factors for injuries by Zwerling et al. [1996, 1998]. The latest wave of the Health and Retirement Survey contains over 30,000 individuals born before 1948 and their spouses. The NLSY collects extensive longitudinal labor market, income, and demographic information on young people aged 14±22 in 1979. In 1988, it began collecting extensive information on occupational injuries and illnesses. The NLSY data are unique among national public-use databases in that the questions about injuries and illnesses are not dependent upon self-report of disability or of workers' compensation receipt. For this reason, it is the only national survey that would allow analysis of all workplace injuries, including nondisabling and unclaimed injuries. While the SIPP does not have as extensive information as the HRS or the NLSY, it is not limited to a particular age group, has a large sample size, detailed information on disability that can be identi®ed as work-related, and extensive demographic, labor market and income data. We are not aware of any research that uses this database to study occupational injuries and illnesses. The Appendix presents information in widely available national public-use databases for research on workplace injuries. The typical limitation in these databases is that workplace injuries are identi®ed either through workers' compensation (thereby excluding unclaimed injuries) or through work-limiting disability (thereby typically excluding injuries that are non-disabling as of the interview date). If the question of whether a disability results from a workplace injury could be added to national surveys that include disability information (as is done in the NHIS, HRS, and SIPP), the ability to study the impact of workplace injuries on workers would be improved. Alternatively, for those surveys that only identify workplace injuries through workers' compensation (such as the PSID and the CPS), the information would be improved if individuals were asked if they experienced an injury in the last year for which workers' compensation were not received. Administrative data An alternative to survey data is workers' compensation administrative data and OSHA logs, which serve as the primary data source for information on injured workers in Economic Consequences of Workplace Injuries many studies. The data from OSHA logs are surveyed in the BLS Annual Survey, which is reviewed by Leigh et al. . The main advantages of workers' compensation administrative data are that state governments and insurance companies already collect them, they often contain detailed information on the progress of a claim through the workers' compensation system and on the injury, and the population of claims is available instead of only a sample. However, there are several disadvantages of workers' compensation administrative data. First is that information on injuries that do not result in claims is unavailable. According to Hensler et al.  and Biddle and Roberts , as many as 40% of workplace injuries do not result in workers' compensation claims. Therefore, both incidence and costs will be signi®cantly underestimated by workers' compensation data. A second problem is that the amount of demographic information about the injured worker is limited. Third, these data sets include limited outcome measures, only rarely going beyond bene®ts paid. Finally, for workers, these data only collect information on compensated time out of work, which misses the waiting period for bene®ts that is required in every state, and also misses all subsequent uncompensated time out of work (discussed further below). Linking across different administrative databases or surveying workers using a sampling frame drawn from an administrative source can correct many of the limitations of workers' compensation administrative data. The most common linked administrative data in recent workplace injury research are claims data linked to wage data. These are the data used in the wage-loss studies discussed further below. Expanding on the links in these databases can provide considerably more information about the consequences of workplace injuries. Links to data from other disability programs, such as Social Security, or links to public use databases, such as the Census or CPS, can provide additional covariates as well as outcomes for analysis. Besides the wage-loss literature, Biddle and Roberts  and Biddle et al.  provide additional examples of studies that use these types of data. Both demonstrate that a signi®cant number of workplace injuries and illnesses do not result in workers' compensation claims. This ®nding would clearly not be possible from workers' compensation administrative data alone. Biddle et al.  link the Michigan Occupational Disease Reports (ODR), a database of occupational diseases collected independently from workers' compensation, to the Michigan Bureau of Workers' Disability Compensation data. Using the ODR as the sampling frame, Biddle and Roberts  survey workers with occupational diseases to determine if they ®led workers' compensation claims. The main impediment to linking is that many of these data sets are con®dential and distributed among several 455 different bureaucracies within states and the federal government or are considered proprietary data by insurance companies that collect them. If worker and corporate con®dentiality issues can be addressed, and coordination established across the keepers of these data, signi®cant advances in workplace injury research will be possible. Measuring the Economic Consequences for Workers By all indications, workers face considerable losses from injury. For example, for women with indemnity claims in Wisconsin in 1989±90, Boden and Galizzi [1999a] estimate average 10-year pre-tax wage losses of $10,841. Reville  ®nds that a 1991±1992 sample of seriously injured workers with permanent disability claims for workers' compensation experienced average before-tax wage losses of $23,692 over the ®ve years after the injury; the most seriously disabled lost $90,793 in that same period. Marquis and Manning  estimate the life-time costs of non-disabling workplace injuries as $10,032 and disabling injuries as $31,183. According to Leigh et al. , lost earnings constitute one-half of the total losses from workplace injuries. Accurate estimation of these costs is arguably the most important task in the estimation of the economic consequences of workplace injuries, both for surveillance and accounting purposes, and for evaluation of programs and interventions. If only workers' compensation administrative data are available, the typical approach to estimating the economic consequences of a workplace injury for the worker has been to examine the number of lost workdays with compensationÐthe duration of temporary disability bene®ts. This is a common outcome measure used in the evaluation of disability management programs (see for instance, the modi®ed work literature review by Krause et al., 1998]. This outcome measure is inherently interesting because reducing the duration of temporary disability bene®ts is advantageous to employers and if associated with a decrease in the days out of work, it is likely to be advantageous to labor as well. However, the duration of temporary disability indemnity does not count days missed during the waiting period required in every state. More importantly, several recent studies (all using workers' compensation administrative data supplemented with other sources) have shown that unemployment subsequent to initial return to work or after the end of temporary disability indemnity is common among injured workers [Butler et al., 1995; Galizzi and Boden, 1996; Biddle, 1998a; Krause et al., 1999; Reville, 1999]. A better way to estimate the consequences of injuries for workers is to estimate the lost earnings over the years after injuries using longitudinal data on earnings for injured workers. This provides a better measure of consequences for 456 Reville et al. injured workers if the injury leads to a higher probability of work absence subsequent to the end of temporary disability bene®ts. Since post-injury earnings data are not available in workers' compensation data, these studies (referred to as wage-loss studies) require linking claims data to another administrative data source. The ®rst wage-loss studies are Johnson et al.  and Berkowitz and Burton . Both studies link claims data to social security earnings records. The recent literature [Biddle, 1998a,b; Reville, 1999; Boden and Galizzi, 1999a,b; Reville et al., 2000; Biddle et al., 2001; Reville and Schoeni, 2001] link claims data to Unemployment Insurance (UI) earnings ®le, which are available at the state level. Two statistical approaches have been used in the recent literature for estimating wage losses. Boden and Galizzi [1999a] and Biddle [1998b] use a regression approach with workers who suffer minor injuries as the control group. The regression corrects for observable differences between the controls and injured workers. Reville  and Reville and Schoeni  adopt a matching approach to estimation that is useful when the number of covariates available on the data is small and the number of controls relative to injured workers is large. The control group consists of up to ten uninjured workers for each injured worker from the same ®rm as the injured worker (prior to injury) with the same earnings. As an illustration of the advantages of linking to data sources outside workers' compensation, this comparison group is possible by selecting additional workers from the state UI earnings ®le. When the data are available, the matching approach is arguably more intuitive and imposes fewer parametric assumptions than the regression approach. A problem with the matching approach is the possibility that some injured workers will not have controls; consequently, the resulting sample will be unrepresentative. In addition, both matching and regression approaches assume that selection is on observed data alone. In other words, they assume that the differences between injured workers and controls that may affect the post-injury earnings patterns can be controlled for using variables available in the data. This assumption may be wrong if the control group is poorly chosen or if a critical variable is missing. Since objections can be raised to any non-experimental control group, comparisons with alternative control groups and the use of multiple methods are helpful when possible. Another possible source of bias occurs if injuries are caused by or correlated with an unobservable event that leads to wage loss. For example, if workers with grievances against their employers are more likely to make a workers' compensation claim and also more likely to quit or be ®red whether or not a claim is made, then the effect of the injury may be confounded by the effect of the grievance. One solution to this problem is to include characteristics of the employer, to the extent possible, in either the regression or matching functions. For this, employers could be surveyed, and the resulting data linked to the matched claims-wage database. Estimates of wage loss have typically counted the earnings during periods of not working as zero. In this sense, the observed loss in wages is not the same as the loss in earnings capacity, and loss in earnings capacity may be closer to the conception of policymakers as the intended target for compensation. Methods of estimating wage capacity for individuals out of work, the problem for which this issue is an example, have occupied labor economists for many years [Heckman, 1979], though have not been applied in this context. One example of when the distinction between wage loss and loss of wage capacity becomes important is when interpreting differences in wage loss across group, such as gender or age groups. For instance, if a pension or social security is available, a workplace injury may lead an older worker to retire when the same or a greater injury would not have at an earlier age. This is an example of greater wage loss without greater loss of earnings capacity. At the same time, it is also possible that older workers take longer to recover from workplace injuries, which would imply greater loss of earnings capacity at older ages. Policy or programs to assist older workers would be different depending upon which explanation was more important in explaining higher losses observed with age. Boden and Galizzi [1999b] and Biddle [1998b] ®nd that the wage losses of men and women are equal on average, but since women earn less than men, the proportional wage losses are higher for women. One possible explanation for this ®nding is that employers are more likely to discriminate against women with injuries, offering them lower wages than men with similar injuries. Another explanation is that women have more alternative opportunities outside work than men, and thus are more likely to stay out of the workforce after an injury (for further discussion, see Boden and Galizzi, 1999b]. Once again, sorting out the different explanations may lead to different policy prescriptions. The wage loss studies conducted thus far have all been limited by the use of workers' compensation administration data to claimed injuries. Indeed, the studies by Reville have been limited to permanent disability claims, while the Boden and Galizzi studies and the Biddle studies have also examined less serious injuries. In future research, publicuse survey data such as the NLSY should be used for examination of the earnings losses associated with unclaimed injuries as well. Directions for Future Research Economists since Adam Smith in The Wealth of Nations have argued that the market must compensate workers for occupational injury or fatality risks. Research on the Economic Consequences of Workplace Injuries existence and magnitude of these ``compensating wage differentials'' has been far more extensive than the research on wage loss and other injured worker outcomes.2 There may be opportunities for cross-fertilization of these two research areas. First, as improved estimates of wage loss improve the estimates of the risks that workers face, detailed estimates of average losses faced for injuries by occupation could be calculated, and it may be possible to measure the ex ante effect on worker's wages more accurately. Second, if workers in risky jobs are receiving higher wages prior to injury, and if an injury tends to cause workers to move to less risky occupations, then measured wage losses may in part re¯ect the loss of compensating wage differentials. There have been no estimates of the quantitative importance of this effect. Another area that would be useful for future research is estimation of income lost instead of wages lost. Evaluations of the adequacy of workers' compensation bene®ts (the purpose of every workers' compensation wage-loss study to date) use the workers' compensation replacement rate as the measure of adequacy. This measure is related to income lost, but it ignores the availability of other income that may follow from a workplace injury. For instance, there may be other program income, such as disability insurance. It may also be possible that a worker's spouse will increase his or her labor force participation in response to the worker's injury. This response would reduce lost income, and may even increase lost wages (relative to unmarried workers), depending upon the substitutability of a spouse's labor force participation. Ultimately, workplace injuries can affect other family outcomes with unmeasured costs. It may be more dif®cult to engage in household activities that are productive, perhaps most signi®cantly child-raising. Spouses may increase labor force participation or decrease it if the uninjured spouse must provide care for the injured spouse. In general, a fruitful avenue for further research would be to consider the effect of injuries on families, taking into account the interdependence of labor force decisions of husbands and wives. Analysis of lost income instead of wages, family income, or bene®ts from alternative sources is dif®cult using administrative data. It would require linking across multiple data sources maintained by different administrative agencies. Linking spouses with administrative data is not likely to be possible. Data on household production is certainly unavailable. However, public use individual and household surveys such as the PSID and the Health and Retirement Survey (HRS) often contain information that can be used for these kinds of analyses. They have been used for studies of 2 Among many economists, compensating wage differentials provide a preferred method for estimating the costs of workplace injuries. Leigh et al.  provides a critique. See Viscusi  and Dorman  for reviews of the compensating wage differential literature. 457 the employment and earnings consequences of disability that do not distinguish between disabilities acquired on and off work [Daly, 1994; Burkhauser and Daly, 1996; Charles, 1997; Charles and Stephens, 2000]. They have also been used for studies of dislocation and downsizing [Ruhm, 1991; Haider and Stephens, 1999]. Stephens  uses the PSID to study the impact of both downsizing and disability on consumption. Haider and Stephens  use the HRS to examine the impact of downsizing on retirement, health insurance, and pensionsÐoutcomes that have not been studied for workers' compensation. Measuring the Consequences for Employers While data and methodological advances have led to a large number of recent improvements in our estimation of the consequences of injury for workers, no comparable progress has been made in the estimation of the costs to employers. By creating a safe working environment and providing workers with equipment, employers may have greater ability to control the number and severity of injuries than workers have, and therefore, accurate estimation of the full costs to employers is critical for education of the employer community and for the design of policies intended to improve safety. Employers face considerable costs arising from workplace injury. One way to measure these costs is to measure the total amount of workers' compensation bene®ts paid (including medical, indemnity, and rehabilitation) in a given year. The National Academy of Social Insurance [Mont et al., 2000] estimates that employer costs estimated in this manner was $41.7 billion in 1998. Alternatively, employer costs can be estimated by summing payments to insurers, state funds, and payments reported to regulators of selfinsured employers; this approach results for 1998 in an estimate of $52.2 billion [Mont et al., 2000]. However, besides costs from workers' compensation, workplace injuries impose many other costs upon employers. Bene®t payments or premium data do not capture lost productivity from time out of work, overtime, retraining, or other costs incurred by employers when injuries disrupt the production process. Despite the potential size of these nonworkers' compensation employer costs, a lack of data has caused the literature on employer costs from workplace injuries to focus on the workers' compensation costs arising from injuries [e.g., Krueger and Burton, 1990].3 A limited description of the costs of workplace injuries to employers 3 Estimates of direct costs such as Mont et al.  rely heavily on aggregate data reported by employers to state Ratings Bureaus or to insurance data collection organizations such as the NCCI or A.M. Best. In addition, data are readily available on premiums paid by employers; these premiums include the bene®ts paid to workers and also capture the insurers' administrative costs and any insurer pro®ts. Data on self-insured employers are usually available only in aggregate form from state regulators. 458 Reville et al. TABLE I. Classification of Employer Costs Direct costs Ex ante costs Ex post costs Indirect costs Workers'compensation insurance premiums (or their economic equivalent for self-insured firms) Injury-prevention programs Costs of compliance with federal and state regulatory agencies Payment of indemnity benefits (workers'compensation and other benefits) Medical benefits for the injured worker (workers' compensation and other health benefits) Return-to-work programs Costs of job accommodations has resulted. This section describes a classi®cation scheme for employer costs, summarizes studies that estimate the direct costs of workplace injuries to employers, and discusses some conceptual issues involved in estimating indirect costs. Classifying Employer Costs A useful way to think about employer costs from workplace injury is to divide the costs between ex-post and ex-ante costs, those occurring before and after injury, respectively. Estimates of employer costs using premiums paid are estimates using ex ante costs, while estimates using bene®ts paid are based on ex post costs. Costs can be further classi®ed into direct and indirect costs, which is a familiar distinction in the literature. Direct costs show up on an employer's accounting balance sheet in anticipation of workplace injury (if ex ante), or in response to actual workplace injuries (if ex post). These are costs incurred to prevent or compensate injuries. Indirect costs are all other costs incurred by employers due to the existence of workplace injuries, such as changes in wages paid to all workers, and employment or training of replacement workers. Table I classi®es employer costs and provides examples of these costs. It should not be construed as counting up all the costs that employers face due to workplace injuries, since several of the costs in the table double-count. For example, workers' compensation premiums and bene®ts measure the same cost paid by employers. While direct costs are more likely to be included among employer costs, even some of these, such as return-to-work and injury-prevention programs are often not included in estimates of the costs of workplace injuries. The table lists the typical item in the direct ex ante cost category: workers' compensation insurance premiums [used to count costs in studies such as Krueger and Burton, 1990; Mont et al., 2000]. It also lists other items such as the cost of setting up injury-prevention programs and complying with Compensating higher wages to workers for job risks Redundant hiring to insure against workplace injury Lost worker productivity Training other workers to replace the injured worker Decreased company morale Overtime costs paid to other workers covering for theinjured worker federal and state regulatory agencies. It also lists the typical item in the direct ex post costs category: workers' compensation medical and indemnity bene®ts [used to count costs in Mont et al., 2000].4 However, it also lists other indemnity programs and other medical bene®ts since not all occupational injury costs are covered by workers' compensation. In addition, other direct costs listed in the table but not typically counted include the cost of job accommodations and return-to-work programs. The table lists as indirect ex ante costs the payment of higher wages that may be associated with higher risk of injury (compensating wage differentials discussed above). It also lists the possibility of higher staf®ng in order to ®ll positions vacant due to recovering injured workers. This is in contrast to the indirect ex post costs listed including lost productivity, training other workers, reduced morale, and overtime costs. The most comprehensive attempts to measure both indirect and direct costs include Miller  and Leigh et al. [1997, 2000], which for instance include measures of disruption costs to employers. Leigh et al.  includes a review of the literature on training costs, though these costs have never speci®cally been measured in association with workplace injuries. In general, data limitations require that these studies rely on, for instance, aggregate training cost measures and ad hoc assumptions to derive their estimates. No studies that we are aware of have attempted a detailed accounting of both indirect and direct costs at even a single ®rm, and certainly not at a large representative sample of ®rms. This is a glaring omission in the literature on the costs of workplace injuries. It is most likely driven by a lack of appropriate data. 4 Several studies compare medical costs for workers' compensation claims with medical costs for similar injures not sustained in the workplace. For examples, see Baker and Krueger , Zaidman , Johnson et al. , Johnson et al. . Studies in this literature tend to ®nd that medical costs are higher in the workers' compensation sector, but none is able to fully account for unobserved differences in the severity of injury between the two groups or, as noted by Leigh and Ward , unobserved differences in the costs of treatment. Economic Consequences of Workplace Injuries Data Required: Linked Employee± Employer Data A key conceptual issue in examining employer's costs from injury is that employers can trade off costs to minimize total costs. For instance, they may discourage workers' compensation claims in favor of claims against employerprovided health insurance if increases in utilization of health insurance affect total costs by less. They may accept more guarantees of employment after injury in exchange for lower wages paid to the workers at risk of injury (an example of a reduction in compensating wage differentials in exchange for reduced risk). Models for the type of data needed to address these open questions about employer cost are rare. The most promising approach is through the construction of linked employee±employer databases, which combine extensive information about a ®rm with administrative data or survey data on the workers at the ®rm. Ideally, the most powerful database of this type would include longitudinal data on both the workers and the employers. For a large number of studies using this type of database, see Haltiwanger et al. . For a discussion (including a discussion of the application of these data to occupational safety and health issues), see Hamermesh . The simplest example of a database of this type for occupational injury and illness research would be a longitudinal database of workers' compensation claims for a large sample of employers, combined with information about the employers. However, this simple example would not capture most of the costs described in Table I. A better approach would be a longitudinal sample of injured workers' wages and injuries, because if wage data are also included, it could be possible to estimate the total labor costs to the employer. Also using these data, it would be possible to examine hiring or ®ring patterns. When multiple employers are included with these data available for each, they could provide a comprehensive view of the impact of injury on employer personnel costs. A similar approach could be used to construct a measure of the impact of occupational injuries on employer medical costs. The appropriate database would allow measurement of the medical costs of unclaimed or ineligible injuries, and costshifting from workers' compensation to employer-provided health insurance (and vice versa). Therefore, the linked employer±employee database would be one that combined longitudinal information on health claims and workers' compensation medical claims for a sample of employers. Linked employee±employer data for measuring the costs of workplace injuries to employers may face greater 5 See the (slightly dated but still useful) reviews by Moore and Viscusi  and Krueger . In addition to papers described in this section, some other signi®cant papers since these literature reviews include Ruser  and Ruser . 459 con®dentiality concerns than the linked administrative data that is used for estimating wage losses and other costs to injured workers. This is because issues of con®dentiality of individual medical records or wage records are compounded with employer con®dentiality concerns. However, to the extent that these concerns can be addressed, both employers and employees will bene®t from the construction of these databases, and the research and policy that they would facilitate. Insurance and Incentives One potentially important indirect cost of workers' compensation arises because insurance bene®ts can change the behavior of workers and employers. This situation is referred to as moral hazard. For workers, moral hazard includes failing to take safety precautions that would have been taken in the absence of insurance, spending a longer period out of work while receiving temporary disability bene®ts, or ®ling a workers' compensation claim for non-work disability. Moral hazard for employers includes the failure to provide safety programs or to offer modi®ed work to injured employees when the employer would have if facing full liability without insurance. There is a large literature on moral hazard effects of workers' compensation, though it has tended to ignore moral hazard by employers in favor of focusing on the behavior among employees.5 This literature has shown that higher disability bene®ts increase the number of workers' compensation claims. It has also shown that higher bene®ts lead to greater time out of work for temporary disability claimants. These results are frequently raised in political debates surrounding proposed increases in statutory bene®ts for injured workers because the implication is that costs to employers will rise by more than the percentage increase in statutory bene®ts. Besides its limited scope, the primary shortcoming to the moral hazard literature is that the positive correlation between bene®ts and injury rates might arise for a number of reasons, each with distinct normative implications. Greater time out of work may mean more malingeringÐor more time with suf®cient income to recover from the injury. Increased claims may mean compensation for uninjured workersÐor a greater fraction of legitimately injured workers receiving bene®ts. Finally, since bene®t changes are the outcome of a complicated political process that takes injury rates into consideration, it is quite possible that higher injury rates lead legislatures and regulators to raise bene®t levels.6 To the extent that this relationship exists, the positive correlation between bene®ts and injury rates can be explained without recourse to insurance market imperfec- 6 Butler  and Leigh  suggest this possibility in their studies. 460 Reville et al. tions.7 Some of the dif®culties interpreting the effects in the moral hazard literature are due to data limitations. For instance, no studies have examined the long-term consequences for injured workers (e.g., wage losses and replacement rates or health status) associated with bene®t changes, which would permit a better interpretation of the impact of the increased time out of work that has been estimated. Nor have any studies examined total employer costs associated with changes in bene®ts, because improved recovery by injured workers may simultaneously reduce turnover and retraining costs and confound estimates. Such a study would require information not available in unlinked workers' compensation administrative databases. The CPS data, which is not longitudinal, is also inadequate for this purpose. No public use database exists with suf®cient information on total employer costs. CONCLUSION While an extensive body of research exists on the consequences of workplace injuries, many gaps in knowledge remain. However, increasingly, databases are available that permit analysis of many of the open questions about the impact of injuries on workers. A large literature has developed recently using linked administrative databases to explore the earnings and employment consequences of workplace injuries and the extent of occupational injuries for which workers' compensation is not claimed. We expect that links to additional databases and further analysis using these types of data will permit great strides in our understanding of the consequences of occupational injury and illness for workers. Large, nationally representative, longitudinal databases like the NLSY, HRS, and SIPP contain information about injuries combined with extensive information about wages and employment, as well as family income and participation in public social insurance and income support programs. However, these databases remain largely unused by researchers in occupational safety and health. As researchers become aware of these databases and exploit the information available in them, it will be possible to better understand differences in the impact of occupational injury and illness across demographic groups, the impact on worker's families, and other outcomes. We have noted the paucity of research on the costs to employers of occupational injuries and illnesses outside of workers' compensation costs. Given the role played by employers in providing a safe and healthy workplace, there is much to be gained from research into employer costs. We describe the type of data that would facilitate rigorous research in this area: linked employee± employer databases, which are gaining favor among labor economists. Availability of data often drives the quality of research, and high quality data are increasingly becoming available for research in the economic consequences of occupational injury and illness. We would like to see more public use survey databases include the questions on occupational injury and illness that are in the databases that we have described. Recognizing that con®dentiality considerations need to be addressed, we also would like to see more public agencies (such as the Social Security Administration) allowing their databases to be linked to other administrative databases and made available for research. Finally, we would like to see research funding agencies encouraging the collection of employer data and the creation of linked employee±employer databases with information on occupational injury and illness. Appendix Survey Current Population Survey (CPS) March Supplement 7 Description and Survey Highlights Relating to Disability Monthly survey of approximately 50,000 households. A joint project of the Bureau of the Census and the BLS. Designed to scientifically sample the civilian non-institutional population, the CPS inquires about labor force characteristics. The more extensive March Supplement asks respondents whether they have received workers compensation payments, if any householdmemberhasa healthproblem,andifthatdisabilitylimitsworkability.Injuryidentifiedonly throughworkers'compensation. Despite the conceptual limitations of this literature, two papers in this literature merit particular attention due to data or methodological contributions. Krueger  constructs a nationally representative database of workers' compensation claimants using the CPS and estimates how the decision to claim workers' compensation is affected by the bene®ts using variation across states in bene®t generosity. Meyer et al.  use a ``natural experiment approach'' involving a maximum bene®t change and comparing workers affected by the change to workers unaffected. While it is often politically impossible or expensive to evaluate occupational safety and health programs using experimental methods, there may be many opportunities for evaluation using linked administrative data and the natural experiment method. For more on the natural experiment approach, see, for example, Meyer , Angrist et al. , and Heckman  and Heckman . Economic Consequences of Workplace Injuries Survey Health and Retirement Study (HRS) Waves1, 2, 3 Medical Expenditure Panel Survey (MEPS) National Health Interview Survey (NHIS) NHIS-D1994 Supplement; NHIS-D 1995-D Supplement NHIS-Occupational Health Supplement (1988) National Longitudinal Survey of Youth1979 (NLSY) Panal Study of Income Dynamics (PSID) Survey of Income and Program Participation (SIPP) Survey of Program Dynamics (SPD) 461 Description and Survey Highlights Relating to Disability A nationally representative longitudinal survey that focuses on retirement, aging, health insurance, and economic security.The surveyoverviewincludesmonitoringofworkdisability asoneofitsobjectives.Queriesincludequestionsaskingrespondents if health limits work ability and if the limitation resulted from a work-related injury, about workplace modifications and about the physical demands of the job.Wave1includes questions about lifetime exposure to workplace hazards. A nationally representative probability survey of non-institutionalized U.S. civilians on health-care issues. Collects data on health care use, satisfaction, access, expenditures, and insurance. Asks if a medical condition was caused by an accident or injury, and whether that injury occurred at work. Part of the National Health Survey, which began in1957. A sample survey of the U.S. civilian non-institutionalized population. Roughly 50,000 households are interviewed annually.The NHIS inquires about the incidence and extent of illness; disability and chronic impairments; and health services received. The core NHIS asks whether an individual was injured seriously during the previous three months, and whether that injury was work-related. Supplementstothecore NHISsurvey thatcollectdisability data andorganizedtheminto a two-phase,detailedsurvey.Eligibility for the second phase (titled the NHIS-D Followback Survey) is dependent upon the presence of a disability, as established in the first phase.The NHIS-D asks if an injury was work-related; whether the individual had retired on disability; and whether special accommodations would need to be made at the workplace so the person could work. Cross-sectional supplement tothe NHIS designedtocollect national data on occupational health. Askswhether therespondent filed a worker's compensation claim; how many days of work were missed; whether the respondent has back pain or hand discomfort; and whether the individual had to alter the type of work performed as a result of the injury. A national probability longitudinal sample of youth who were between the ages of14 and 21as of December 31,1978.Designed to gather data on individuals' labor market experience and over-represents blacks, Hispanics and economically-disadvantagednon-blacks/non-Hispanics. Asksrespondentswhetherhealth is limitingthemfromworking; whether they had a workrelated injury; whether they filed a workers'compensation claim; whether they had to changethe type of workperformedas a result; and whether the injury caused the respondent to lose wages or to be laid off. An annuallongitudinalstudyofa nationallyrepresentativesample.Thesurveybegan in1968witha sample of5,000households and has grown to a sample of approximately 8,700 households as of1995.Collects economic and demographic information, such as income,employment,family structure, andresidence. Asks whetherrespondentsreceivedany incomefromworkers' compensation, and asks whether a health or physical problem limits work ability. Anational survey,consisting ofa continuousseries ofpanels,thatcollectsdata on income,laborparticipation, andeligibility and participationin publicbenefitprograms,toestimate governmentprogramcosts andtomeasureprogrameffectiveness.Asks whether respondents received any disability incomeöspecifically from workers' compensationöand asks whether a health or physical problem limits work ability. Queries whether this health limitation is caused by an accident or injury on the job, and asks whether the individual is able to work regularly at the same job they held before. Anationallyrepresentativesampleofnon-institutionalizedcivilians.Interviewedapproximately30,000householdsthat werein the1992 and1993 SIPP panels to determine economic status of individuals within and outside of the labor market. Collects demographic data,labor market information, and data on respondents' workexperience,income and benefits. Asks whether therespondents have received worker's compensation payments, and asks the respondents if they have trouble completing everyday household activities, such as walking three city blocks or carrying a full bag of groceries. ACKNOWLEDGMENTS REFERENCES The support of the RAND Institute for Civil Justice and the State of California Commission for Health and Safety and Workers' Compensation is gratefully acknowledged. We thank Les Boden, Thomas Deleire, Allard Dembe, Beth Giddens, Ted Miller, Seth Seabury, two anonymous reviewers and conference participants for helpful comments. Angrist J, Imbens G, Rubin D. 1996. Identi®cation of causal effects using instrumental variables. J Am Stat Asso 91:444± 455. Baker L, Krueger A. 1995. Medical costs in workers' compensation insurance. J Health Econ 14:531±549. Berkowitz M, Burton JF Jr. 1987. Permanent disability bene®ts in workers' compensation. Kalamazoo, MI: W. E. Upjohn Institute for Employment Research. 462 Reville et al. Biddle J. 1998a. Description and analysis of measures of the postinjury return to work patterns of washington state workers' compensation claimants. Mimeo, Performance Audit of the Washington State Workers ' Compensation System. Biddle J. 1998b. Estimation and analysis of long term wage losses and wage replacement rates of washington state workers' compensation claimants. Mimeo, Performance Audit of the Washington State Workers ' Compensation System. Biddle J, Roberts K. 1999. Evidence on the level and determinants of workers' compensation take-up rates. Mimeo, Michigan State University. Biddle J, Roberts K, Rosenman KD, Welch EM. 1998. What percentage of workers with work related illnesses receive workers' compensation bene®ts? J Occup Environ Med 40:325±331. Biddle J, Boden L, Reville R. 2001. Permanent partial disability from occupational injuries. In: Budetti et al., 2000. Boden L, Galizzi M. 1999a. Economic consequences of workplace injuries and illnesses: lost earnings and bene®t adequacy. Am J Ind Med 36:487±503. Hamermesh DS. 1999. LEEPing into the future of labor economics: the research potential of linking employer and employee data. Labour Economics 6(1):25±41. Heckman JJ. 1979. Sample selection bias as a speci®cation error. Econometrica 47(1). Heckman JJ. 1997. Instrumental variables: a study of implicit behavioral assumptions used in making program evaluations. J Hum Re 32(3):441±62. Heckman JJ. 1998. Addendum to ``instrumental variables: a study of implicit; assumptions used in making program evaluations.'' J Hum Res 33(1):247. Hensler DR, Marquis MS, Abrahamse AF, Berry SH, Ebener PA, Lewis E, Lind EA, MacCoun RJ, Manning WG, Rogowski JA, Vaiana ME. 1991. Compensation for accidental injuries in the United States. Santa Monica, CA: RAND, R-3999-HHS/ICJ. Hirsch BT, MacPherson DA, Dumond JM. 1997. Workers' compensation recipiency in union and nonunion workplaces. Ind Labor Rel Rev 50(2):213±236. Boden IL, Galizzi M. 1999b. Income losses of women and men injured at work. Mimeo, Boston University. Johnson WG, Cullinan PR, Curington WP. 1978. The adequacy of workers' compensation bene®ts. Research report of the interdepartmental workers'. Compen Task Force 6:95±121. Budetti PP, Burkhauser RV, Gregory JM, Hunt HA. 2001. Ensuring health and income security for an aging workforce. Kalamazoo: W.E. Upjohn Institute for Employment Research MI. Johnson WG, Burton JF Jr., Thornquist L, Zaidman B. 1993. Why does workers' compensation pay more for medical care? Bene®ts Quarterly 9(4):22±31. Burkhauser RV, Daly MC. 1996. ``Employment and economic wellbeing following the onset of a disability, the role of public policy.'' Mashaw L, Jerry Reno, Virginia Burkhauser, Richard Berkowitz, editors. Disability, work and cash bene®ts. Kalamazoo: W. E Upjohn Institute. p 59±101. Johnson WG, Baldwin ML, Burton JF. 1996. Why is the treatment of work-related injuries so costly. New Evidence from California Inquiry 33:53±65. Krause N, Dasinger LK, Neuhauser F. 1998. Modi®ed work and return to work: a review of the literature. J Occ Rehab 8(2). Butler RJ. 1994. Economic determinants of workers' compensation trends. J Risk Insu 61:383±401. Krause N, Dasinger LK, Deegan L, Brand R, Rudolf L. 1999. Alternative approaches for measuring duration of work disability after low back injury based on administrative workers' compensation data. Am J Ind Med 35:604±618. Butler RJ, Johnson WG, Baldwin ML. 1995. Managing work disability: why ®rst return to work is not a measure of success. Ind Labor Rel Rev 48:452±469. Charles KK. 1997. The longitudinal structure of earnings losses among disabled workers, mimeo. University of Michigan. Charles KK, Stephens MS. 2000. Job displacement, disability and divorce, mimeo, University of Michigan. Daly MC. 1994. The economic well-being of men with disabilities: a dynamic cross-national view. Dissertation from the Maxwell School at the University of Syracuse. Dasinger LK, Krause N, Deegan LJ, Brand RJ, Rudolf L. 1999. Duration of work disability after low back injury: a comparison of administrative and self-reported outcomes. Am J Ind Med 35:619± 631. Dorman P. 1996. Markets and mortality: economics, dangerous work and the value of human life. Cambridge University Press. Dorman P, Hagstrom P. 1998. Wage compensation for dangerous work revisited. Ind Labor Rel Rev 52(1):116±135. Galizzi M, Boden LI. 1996. What are the most important factors shaping return to work? In: Evidence from Wisconsin. Cambridge, MA: Workers Compensation Research Institute. Haider S, Stephens M Jr. 1999. The impact of displacement for older workers''. Mimeo, RAND. Haltiwanger C, Lane JI, Spletzer JR, Theeuwes JJM, Kenneth R, Troske KR. 1999. The creation and analysis of linked employeremployee data, contributions to economic analysis. Amsterdam, London, and New York: Elsevier Science, North-Holland. Krueger AB. 1990. Incentive effects of workers' compensation insurance. J Pub Econ 41:73±99. Krueger AB, Burton J. 1990. The employers' costs of workers' compensation insurance: magnitudes, determinants, and public policy. Rev Econ Stat 72(2):228±240. Leigh JP. 1985. Analysis of workers' compensation using data on individuals. Ind Rel 24(2):247±256. Leigh JP. 1997. Review of markets and mortality: economics, dangerous work, and the value of human life, In: Industrial and labor relations review 50: Cambridge University Press. p 519±521. Leigh JP, Ward MM. 1997. Medical costs in workers' compensation insurance: comment. J Health Econ 16:619±622. Leigh JP, Markowitz SB, Fahs M, Shin C, Landrigan PJ. 1997. Occupational injury and illness in the United States: estimates of costs, morbidity, and mortality. Arch Int Med 157: 1557±1568. Leigh JP, Markowitz SB, Fahs M, Landrigan PJ. 2000. Arbor Ann, editor. Costs of occupational injuries and illnesses. University of Michigan Press. Marquis MS, Manning WG. 1999. Lifetime costs and compensation for injuries. Inquiry 36:244±254. Meyer BD. 1994. Natural and quasi-experiments in economics. working paper. Natl Bur of Econ Res 170. Meyer B, Viscusi WK, Durbin DL. 1995. Workers' compensation and injury duration: evidence from a natural experiment. Am Econ Rev 85(3). Economic Consequences of Workplace Injuries 463 Miller T. 1995. Injuries to workers and their dependents. J Saf Res 26:75±86. Reville RT, Polich S, Seabury S, Giddens E. 2000. Permanent disability at private self-insured ®rms. RAND MR-1268-ICJ, 2000. Miller T. 1997. Estimating the costs of injury to U. S. employers. J Saf Res 28:1±13. Ruhm C. 1991. Are workers permanently scarred by job displacements. Am Eco Rev 81:319±323. Mont D, Burton JF Jr., Reno V. 2000. Workers' compensation: bene®ts, coverage, and costs, 1997-1998 new estimates. Natl Acad Soc Ins. Ruser JW. 1993. Workers' compensation and the distribution of occupational injuries. J Hum Res 28(3):593±617. Moore MJ, Viscusi WK. 1990. Compensation mechanisms for job risks: wages, workers'. In: Compensation and product liability. Princeton: Princeton University Press. Peterson MA, Reville RT, Stern RK, Barth P. 1997. Compensating permanent workplace injuries: a study of California's system. Santa Monica, CA: RAND MR-920. Reville RT. 1999. The impact of a permanently disabling workplace injury on labor force participation and earnings. In: Haltiwanger et al., editors. 1999. The creation and analysis of linked employer-employee data, contributions to economic analysis. Amsterdam, London, New York: Elsevier Science, North Holland. Reville RT, Schoeni R. 2001. Disability from injuries at work: the effects on earnings and employment. RAND Labor and Population Working Paper, DRU-2554-LP. Reville RT, Inkelas M, Rogowski J, Karoly L. Data sources to evaluate the effect of the welfare reform legislation on disabled children. RAND Report (Restricted Draft), February 1998. Ruser JW. 1998. Does workers' compensation encourage hard to diagnose injuries. J Risk Insu 65(1):101±124. U.S. Census Bureau. 1997. Statistical Abstract of the United States: Washington D.C: US. Government Printing Of®ce. (117 th edition.). Viscusi WK. 1993. The value of risks to life and health. J Eco Lit 31:1912±1946. Zaidman B. 1990. Industrial strength medicine: a comparison of workers' compensation and blue cross health care in Minnesota. Minnesota Department of Labor and Industry. Zwerling C, Sprince NL, Wallace RB, Davis CS, Whitten PS, Heeringa SG. 1996. Risk factors for occupational injuries among older workers: an analysis of the health and retirement survey. Am J Pub Health 86(9):1306±1309. Zwerling C, Sprince NL, Wallace RB, Davis CS, Whitten PS, Heeringa SG. 1998. Occupational injuries among older workers with disabilities: a prospective cohort study of the health and retirement survey, 1993 to 1994. Am J Pub Health 88(11):1691±1695.