How does health insurance affect workers' compensation filing?

Lakdawalla, Darius N. ; Reville, Robert T. ; Seabury, Seth A. 等

I. INTRODUCTION

Workers' compensation serves as a near-universal health insurance system for workers who become sick or injured because of their job. Almost every private sector worker in the United States is covered by workers' compensation, which typically provides full reimbursement for medical expenditures and partial reimbursement for wage losses incurred as a result of a workplace injury or illness. (1)

Given that employer-provided health insurance usually involves cost sharing for medical expenditures and offers no income replacement, one would expect workers to seek compensation for nearly all workplace health conditions.

However, it appears that many injured workers elect not to participate in workers' compensation. Using data from a survey of workers in Michigan, Biddle and Roberts (2003) found that just 55.1% of the workers with lost time reported filing for any workers' compensation benefits and just 39.1% reported filing for wage replacement. This suggests that workers may not view workers' compensation filing as "free." For example, workers might lack information about the availability, or even the existence, of workers' compensation benefits. Employers might discourage participation, and there could be a stigma associated with filing for benefits. Additionally, a worker might be uncertain about his or her ability to prove that an injury or an illness is work related. As a result, pursuing a workers' compensation claim could require that a worker spend money on healthcare consultations upfront, with uncertain prospects for reimbursement. While any of these costs could lead a worker not to file a claim, the magnitude of nonparticipation is still surprising.

The existence of alternatives to workers' compensation, specifically health insurance, is often thought to exacerbate nonparticipation. Injured workers may be unwilling to bear workers' compensation filing costs if their health insurance reduces or eliminates the burden of medical costs. Moreover, individuals without health insurance have greater incentive to pass off nonoccupational injuries or illnesses as work related. If the possession of health insurance explained nonparticipation, there would be fewer normative concerns that poor participation imperils the health of injured workers.

However, a look at data on health insurance and workers' compensation filing from the National Longitudinal Survey of Youth (NLSY), reported in Table 1, calls the conventional wisdom into question. The first two columns of the table show that injured workers without health insurance are 7 percentage points (about 13%) less likely to file a claim than those with health coverage. (2) This is in spite of the fact that uninsured workers are more likely to have lost wages and missed more days of work as a result of their injuries. This result is puzzling, given the argument above that the uninsured worker should be more likely to file.

The last two columns of the table add nuance to this picture, showing that the gap in filing widens substantially when we compare workers in firms that do and do not offer health insurance. Workers in insuring firms are 16 percentage points more likely to file a claim, even though their injuries are still less severe in terms of lost wages and work time. It would appear that the effects of health insurance on workers' compensation claiming are not as straightforward as they might appear and may have more to do with the characteristics of employers who offer health insurance than with the incentives faced by the workers. This table illustrates a result that we confirm in this study; actual health insurance coverage itself is much less important than whether an employer offers health insurance.

We found that injured workers are more likely to file claims when they work for an employer who offers health insurance. This result holds even when we incorporate worker fixed effects and other control variables, suggesting that the characteristics of the employer are not simply proxies for unobserved worker characteristics. Rather, workplaces offering health insurance may be offering more information and/or encouraging injured workers to seek benefits. Worker heterogeneity and incentives appear to be, at most, second order relative to the effect of employer heterogeneity or incentives.

The health insurance offer may causally diminish incentives for employers to fight workers' compensation filing, perhaps because they pay for injuries and illnesses regardless. Alternatively, employers who choose to offer health insurance may have unobserved characteristics that also make them less likely to discourage workers' compensation filing. For example, they may be generally more "worker friendly" or may face higher returns from providing fringe benefits and nonpecuniary compensation.

Our findings also support the idea that the ease of filing and the availability of support for filing are crucial factors in an individual's decision to file. We confirm the findings of previous studies, such as Hirsch, Macpherson, and DuMond (1997), that union membership makes an individual worker more likely to file. We also show that workers with conditions that are harder to document--specifically occupational illnesses--are less likely to file. If workplace environment and employer heterogeneity are dominant factors in the decision to file, this suggests the importance of targeting employers--rather than employees--in efforts to raise the use of the workers' compensation system.

We begin by providing institutional background on the workers' compensation system in Section II. We then describe the data and our empirical framework in Section III. Our empirical findings are presented in Section IV.

II. BACKGROUND

The workers' compensation system is a series of state laws that require employers to provide workers with medical and indemnity benefits when they suffer an injury or an illness at, or because of, their work. (3) Currently, all 50 states and the federal government have workers' compensation systems, although technically, it is optional in New Jersey and Texas. Almost all states provide full medical coverage with no cost sharing. (4) Benefits vary by state, but most provide two-thirds income replacement subject to weekly benefit caps and floors. (5) In 2003, workers' compensation benefit payments (medical and indemnity) totaled approximately $54.9 billion, with total costs to employers of approximately $80.8 billion. (6)

One of the defining characteristics of workers' compensation systems across different states is their inclusiveness. Almost all workers are eligible for compensation if they are injured, with only a few types of workers exempt from coverage. (7) Additionally, virtually all workplace injuries are eligible for compensation. Workers' compensation provides no-fault coverage for workplace injuries, with a few exemptions for certain kinds of worker misconduct such as substance abuse. Eligibility is determined on the basis of workplace causality, the demonstration of which has traditionally required a fairly low standard of proof. Even a relatively minor connection to work generally suffices to ensure compensation.

Given this purported inclusiveness, it is somewhat puzzling to see such low filing rates by injured workers. When deciding whether or not to file, individuals will do so if the expected benefits of filing outweigh the expected costs. There are a number of factors that could affect both the expected benefits and the costs of filing. Biddle and Roberts (2003) found that the injury severity of workers is one of the strongest determinants of the decision to file, with only the more severely injured workers being likely to file for benefits. This is sensible because more severely injured workers will be out of work for a longer period of time and therefore have greater demand for income replacement. Both costs and benefits could explain the result of Hirsch, Macpherson, and DuMond (1997), who showed that unionized workers are much more likely to file. Unionized workers might be more informed about the availability of benefits (increasing the expected benefit of filing) or they may have more protection against employer retribution (decreasing the expected cost).

While these studies indicate that individual characteristics and circumstances influence the filing decision, it also appears that employers have some ability to influence their workers' decisions. Biddle and Roberts (2003) reported that more than 20% of workers with serious injuries report that fear of employer reprisal played a role in their decision not to file a workers' compensation claim. This is consistent with other studies, such as Azaroff, Levenstein, and Wegman (2002), which found that workers fail to report job-related injuries out of fear that this will cause them to be labeled as "careless" or "a complainer," having adverse effects on future promotions and increased earnings.

Given that employers have the ability to influence the filing decisions of workers, what is their incentive to discourage claims? The obvious answer is that workers' compensation claims are costly to firms. Compare the costs of a workers' compensation claim to those of a health insurance claim for the identical injury. If there is sufficient lost time, the workers' compensation claim will have indemnity benefits associated with it, which the health insurance claim will not. These indemnity costs will increase the incentives of workers to remain out of work, as shown by Meyer, Viscusi, and Durbin (1995), increasing the amount of lost productivity or replacement costs by employers. Moreover, previous work, such as Baker and Krueger (1995, 1997) and Johnson, Baldwin, and Burton (1996), has shown that the medical treatment costs for the same injury are higher under workers' compensation than under traditional health care.

The extent to which an employer bears these costs directly will depend on the mechanism through which they insure against workers' compensation losses. Self-insured firms bear the full brunt of costs, while insured firms will face increased premiums through experience rating. (8) While the extent of experience rating varies across employer characteristics such as firm size, virtually all firms will bear some cost from a workers' compensation claim and that cost will typically be larger than the equivalent claim against a group health insurance policy.

The central finding of this study was that workers in firms that offer health insurance are more likely to file for workers' compensation; this seems to be because the health insurance offer signals something about the employer rather than the employee. This raises the question as to why if workers' compensation claims are costlier, some firms might welcome claims more often than others. Generally speaking, an employer with a workforce that places a higher value on fringe benefits and nonsalary compensation should be more likely to tolerate workers' compensation claims, which represent an additional element of nonsalary compensation. As an example, firms that pay high wages provide relatively more value with a given dollar of non-salary compensation because their workers face higher marginal income tax rates. Alternatively, firms that face high replacement or recruitment costs might be more willing to pay for the "restoration" of a worker's health after an injury. Note that all these are motives both for lenient workers' compensation policies and for group health insurance provision. If these motives are strong enough and closely correlated enough, they could produce a positive correlation between health insurance and workers' compensation filing, even though the two might be substitutes from an individual worker's perspective.

III. DATA AND METHODS

While we were interested in the general question of why workers do or do not file for workers' compensation in the event of an injury, the primary focus of our analysis was the role of health insurance. On the surface, Table 1 suggests that contrary to the conventional wisdom, workers with health insurance are more likely to file for workers' compensation. In our analysis, we examined the extent to which health insurance matters for filing and attempted to distinguish between the effects of possessing health insurance and being offered health insurance.

There has been relatively little empirical study by economists on the link between health insurance and a worker's claiming behavior. Most past work has focused on the importance of expected benefits and information in the workers' compensation filing decision. Numerous studies all showed a positive relationship between the workers' compensation benefit levels and the receipt of benefits. (9) In principle, this could come from either a higher propensity to file a claim or increased risk taking by workers; as Butler (1994) noted, however, the evidence suggests that an effect on claim filing is the predominant cause of the relationship. Hirsch, Macpherson, and DuMond (1997) examined the effect of union membership on the receipt of workers' compensation benefits and found that it is positive. They interpreted this result as suggesting that unionized workers are both better informed and better protected from employer retribution.

The other extant work has focused primarily on whether or not workers without health insurance fraudulently pursue workers' compensation benefits for nonoccupational injuries. Smith (1990) documented what is commonly referred to as the "Monday effect," the observation that sprains and strains were more likely to occur on Mondays, but more visible injuries with clearer causation were no more likely to occur on Mondays than on other days of the week. This evidence suggests that workers might suffer strains or sprains during the weekend but report them on Mondays in order to receive workers' compensation benefits. Card and McCall (1996) raised some doubts about this issue, however, with their finding that workers who are more likely to have health insurance are no less likely to file a Monday claim. A key limitation of their study, however, was the absence of direct information on whether or not individual claimants actually possess health insurance. Instead, they were forced to impute the likelihood that an individual possessed health insurance based on observable characteristics.

Using the NLSY, we estimated the impact of health insurance on filing behavior more directly. The NLSY is a rich, nationally representative data source with individual-level information on demographics, health insurance status, and unionization status and detailed information on whether a worker suffered a work-related health condition, whether she filed a claim as a result, and whether she received compensation for that claim. Its true panel aspect is also of value because we can show how individuals behave when their health insurance coverage changes.

A. Data

The NLSY is a longitudinal survey containing a representative sample of people who were between the ages of 14 and 22 yr in 1979. This initial cohort is resurveyed every year until 1994, after which point they are surveyed every other year. In addition to basic demographic information, such as age, sex, race, and marital status, questions are asked about wages, unionization status, and health insurance status. The NLSY also contains some information on employers, including the size of the establishment, whether health insurance is offered, and whether paid vacation or sick leave is offered. In addition, from 1988 onward, there is a substantial amount of data on workers' compensation claims.

First, the NLSY collects data on whether the worker reports a workplace injury or illness. They also collect data on whether a workers' compensation claim was filed for any of those injuries or illnesses, as well as data on whether the worker received a settlement. In addition, they report the ICD-9 code for the illness or injury. We used this to classify a health condition as an injury or an illness. This breakdown is made strictly according to the ICD-9 codes; if the code is for an "illness" or "disease," we classified the health condition as such. Otherwise, the condition is called an injury. Finally, every year, the NLSY reports the total amount of workers' compensation benefits received by each individual. Table 2 summarizes the workers' compensation data contained in the NLSY. Eight years of data contain 4,737 observations in which individuals suffer workplace health conditions.

Occupational injuries and illnesses are self-reported in the NLSY, as is filing behavior. The survey begins by asking respondents: "During the past 12 mo, have you had an incident at any job we previously discussed that resulted in an injury or illness to you?" It then goes on to ask whether this resulted in an injury or an illness, and "What part of the body was hurt or affected?" Then, "For each part of the body listed [previously], what kind of [injury/illness] was it?" The answer to the latter question is used by the survey to assign an ICD-9 code to the condition. After' ascertaining the type of injury, the survey asks a battery of questions about the consequences of the injury, "Did the injury/illness cause you to miss one or more scheduled days of work, not counting the day of the injury/illness?" A similar question is asked about wage loss.

Another important question for our purposes was: "Did you or your employer fill out a worker's compensation form for this (injury/illness)?" This we took to be our measure of claim filing. We preferred an objective e, measure of injury or illness, combined with an administrative record of a claim, but our self-reported data seem reliable, as far as can be ascertained. Biddle and Roberts (2003) used a sample of Michigan workers whose job-related injury status was verified by physicians and whose workers' compensation claims were verified by administrative data. They found that on average, 55% of injured workers filed claims. This is very close to the overall mean filing rate in our nationally representative sample that uses self-reported data. Other studies of self-reported data come to conclusions quantitatively similar to ours. Maier and Reinke (2004) reported on an Oregon state-sponsored study of the 2002 Oregon Population Survey that suggests 54% of workers reporting a workplace injury in the past 12 mo filed a claim.

Table 3 presents the average population characteristics of the NLSY sample (born between 1957 and 1964), using the NLSY sampling weights to construct averages that are representative of the NLSY-aged people in the nation as a whole. (10) In the table, a column to the right of another one is a proper subset of the latter. We begin by presenting statistics for the whole population, proceed to those who were employed in the last year, next to those who were employed in the past year and injured in the past year, proceed to the injured people who filed, and finally to the fliers who received benefits.

The table shows that injured workers are slightly less educated than the average worker; filing workers are less educated still than injured ones; and those receiving compensation are the least educated of all. Similar patterns are evident for earnings. The tables support the results of Hirsch, Macpherson, and DuMond (1997), showing that filing workers are more likely than nonfiling workers to belong to unions. Injured workers appear to work at somewhat smaller firms than average. The raw means also demonstrate that filing status is positively correlated with whether an employer offers health insurance.

Table 4 displays the characteristics of the injured population (i.e., the third column of data in Table 3) broken down by employer health insurance offers and filing decisions. There appear to be few demographic (age, sex, or race) differences between those working at employers who do or do not offer health insurance, though workers at insuring firms are more likely to be married, more likely to be employed in the past 2 wk and slightly more educated. Not surprisingly, being at an insuring firm is associated with significantly higher earnings, a greater likelihood of having health insurance, and being unionized. Interestingly, there is relatively little difference between the two in whether or not any time was missed due to injury, but workers at employers who do not offer health insurance remain out of work for a substantially longer period of time and are more likely to have lost wages.

The table reveals complex relationships among health insurance, filing, and establishment size. (11) Employers who offer health insurance to workers appear to be larger on average than those who do not. However, comparing employers who offer health insurance to those who do not, there are conflicting relationships between filing tendencies and establishment size. Within the group of firms offering health insurance, filing workers are at smaller establishments, but the opposite is true within firms not offering health insurance. This suggests a possible nonlinear relationship between filing and firm size: filing initially rises but eventually falls with firm size. We found some evidence of this later, although the relationship between firm size and filing tends not to be robust.

We merged into the NLSY two measures of job-related strenuousness to investigate whether these affected rates of reported injuries or filing. These measures were job-related exercise and job-related strength requirements. The first measures the number of fitness-related demands (e.g., crawling, reaching, stooping, and climbing) and the second measures the strength-related demands (e.g., lifting) of a job. Both these variables are constructed by the Department of Labor in the Dictionary of Occupational Titles. Each three-digit level occupation is rated on a scale of 1-4 for its job-related exercise and on a scale of 1-5 for its strength requirements.

In order to analyze the impact of workers compensation generosity, we included information on state-level benefit rules. Specifically, we constructed for each worker the fraction of wages lost after an injury that would be replaced by workers' compensation in the event of a successful claim. To arrive at this estimate, we used workers' compensation temporary disability benefits, which are based on weekly wages and typically set equal to two-thirds of weekly pre-injury wages subject to a maximum weekly benefit (though there is some small variation in the rate). The maximum varies across states and over time; the U.S. Chamber of Commerce's annual Analysis of Workers' Compensation Trends reports that in 1999, it was approximately $592.92 a week or about $29,646 annually. A lower benefit maximum is associated with a lower replacement rate of lost income for workers with weekly wages over the benefit cap because benefits fail to increase with earnings once the caps are met. For workers under the cap, variation in replacement rates is minimal. Among all NLSY workers, the average replacement rate is approximately 64%.

B. Empirical Framework

An injured worker will file for workers' compensation benefits if the expected benefits of doing so are at least as great as the expected costs. While we did not directly observe the costs and benefits of filing, which are functions of unobservable characteristics like individual preferences, we did observe many variables that should be correlated with expected benefits, expected costs, or both. Empirically, we modeled the workers' compensation filing decision of injured worker i in firm j at time t as:

(1) [F.sub.ijt] = [[beta].sub.0] + [[beta].sub.1] [H.sub.ijt] + [[beta].sub.3][X.sub.it] + [[beta].sub.4][B.sub.it] + [[alpha].sub.t] + [[theta].sub.i] + [[lambda.sub.j] + [[mu].sub.ij] + [[epsilon].sub.it]

[F.sub.ijt] is a binary variable, indicating whether or not individual i in employer j filed at time t. (12) In some specifications, the variable H measures whether or not an individual has health insurance; in others, it measures whether he/she is offered it by her employer. Distinguishing between these two variables turns out to be interesting and important. [S.sub.it] represents the severity of the condition, which we modeled using the number of days the individual is absent from work. The vector [X.sub.it] contains demographic characteristics, such as age, sex, race, and education, as well as other observable job characteristics such as the amount of physical exertion involved, that might affect the likelihood of filing a claim. The replacement rate offered by workers' compensation benefits is given by [B.sub.it]; these vary for different individuals over time and across states. Year-specific fixed effects are represented by the term [[alpha].sub.t].

We expected that heterogeneity at the individual and employer level likely plays an important role in the propensity to file for workers' compensation. The term [[theta].sub.i] represents individual heterogeneity; individuals with higher values of [[theta].sub.i] are more likely to file a claim than others. The terms [[lambda].sub.j] and [[mu].sub.ij] represent firm-level heterogeneity and match-specific heterogeneity, respectively. The firm-specific term would lead to increased filing if, say, firm j were less likely to challenge claims by any of their workers. The match-specific term would lead to increased filing if firm j were less likely to challenge claims filed by individual i but not (necessarily) any other employee.

Because the NLSY is a true panel, it contains repeated observations on individuals. This allows us to control directly for time-invariant individual heterogeneity. Unfortunately, since the NLSY does not identify employers (and since the sample is too small, in any event, to share many common employers), we were unable to identify employer-specific fixed effects. However, we did include controls for industry, state, and establishment size, which absorb some heterogeneity in employers, but we could not isolate the causal effect of any variable that is correlated with employer-specific or match-specific heterogeneity. Therefore, we can rule out (or in) the importance of individual heterogeneity but not employer-specific or match-specific heterogeneity.

We cannot distinguish between causal effects of health insurance and heterogeneity in the employers or employer-employee pairs that feature health insurance. However, our results do allow us to conclude that individual employee heterogeneity is not of great importance. (13) The positive relationship between employer offers of health insurance and workers' compensation filing must then owe itself to (a) the causal effects of health insurance, (b) selection based on employer characteristics, or (c) selection on match-specific characteristics that are not fixed for an individual worker.

The potential sources of heterogeneity are fairly straightforward. Whether or not the employer offers health insurance is likely to be related to unobserved employer characteristics. The possession of health insurance may also be related to employer heterogeneity, as a significant fraction of workers obtain health insurance from their employers. Both could also be related to match-specific heterogeneity, if better matched individuals receive more benefits of all kinds from their employers. (14)

Identifying the coefficients in this model while controlling for individual heterogeneity depends crucially on the properties of [[theta].sub.i]. If [[theta].sub.i] is uncorrelated with the other covariates, a random effects model is the most efficient choice among consistent estimators. If it is correlated, however, only a fixed effects model will yield consistent estimates, up to the employer fixed effects. In our empirical work, we reported the results of both fixed effects and random effects models and tested for the correlation between [[theta].sub.i] and other covariates using a Hausman specification test.

Finally, note that our results may be influenced by the way we define and measure injuries. Our ultimate object of interest is Pr(Filing | Injury), the probability that an injured worker files. To estimate this, we calculated the probability of filing for the NLSY respondents who report an injury. If the set of people reporting an injury is a random sample of the injured population, this strategy is straightforward. The results might not be generalizable, however, if those likely to report an injury have different characteristics than the average injured worker. Unfortunately, while we can bear this caveat in mind, this limitation cannot be overcome without objective (i.e., not self-reported) injury data confirmed by a neutral third party.

IV. RESULTS

A. Determinants of Injuries

Table 5 provides a first look at the characteristics of the injured population by regressing, for the entire NLSY population, a binary variable for whether the individual reported a workplace injury to the NLSY on a host of covariates. The first four columns help identify the relationship between filing and whether or not the individual's current or most recent employer made health insurance available to him. This we define as an employer's offer of health insurance. In contrast, the next two columns investigate the relationship between filing and the individual's actual possession of health insurance from any source. Finally, Columns 7 and 8 include both an employer's offer and the worker's possession of health insurance as regressors in the same model.

The columns differ according to the way individual heterogeneity is modeled. The model in the first column includes only a year fixed effect. The second adds a three-digit standard industrial classification industry fixed effect. The third adds a state fixed effect and an individual fixed effect. The fourth replaces the individual fixed effect with an individual random effect. The last several rows of the table report the heterogeneity structure for each model.

The random effects model makes the identifying assumption that the individual heterogeneity term is independent of the covariates. We tested this assumption using a Hausman specification test. The Hausman test relies on the fact that the individual fixed effects model is always consistent, regardless of whether [[theta].sub.i] is correlated with covariates, but that the random effects model is only consistent if the assumption holds. If the hypothesis is false, the difference between the two estimates will be non-zero asymptotically. For the model of injury determinants, the Hausman test clearly rejects the null hypothesis that [[theta].sub.i] is uncorrelated with the covariates. The p value in the models using just the employer-offered insurance variables is less than 0.0001; for the models using just health insurance possession, it is 0.001; and for the models using both, it is 0.0035, all rejecting the null at the 1% significance level. Therefore, the fixed effects model is the only appropriate one in this context.

The first four columns of Table 5 reveal that the employer offer of health insurance is associated with a 0.4-1 percentage point increase in the probability of a workplace injury. The relationship is statistically significant at the 5% level in all the models, except the one with individual fixed effects (Column 3). The possession of health insurance is associated with a 0.9-1.3 percentage point increase in the probability of a workplace injury; this relationship is always statistically significant. When we include the offer and possession of health insurance simultaneously, neither is statistically significant.

These results suggest that workers at higher risk for workplace injuries are more likely to possess insurance and to be offered insurance by their employers. It is difficult to determine whether possession of health insurance is more or less important than an employer's offer, although health insurance possession is more robust to individual fixed effects. Overall, the interpretation most consistent with the bulk of the evidence is that individuals facing a higher risk of workplace injury are more likely to possess insurance from some source. The employer's offer of health insurance, in and of itself, is not robust to individual heterogeneity and seems a less likely candidate for explaining injury risk. However, we argue later that the employer's offer is a much better predictor of filing behavior than health insurance possession, even after accounting for individual heterogeneity.

The relationship between a workplace injury and the other explanatory variables is largely what we would expect. More educated people are less likely to report suffering an injury. (15) This is likely to be because they tend to work in less dangerous jobs. People in jobs that require more strength are more likely to self-report injuries, but interestingly, people in jobs demanding more fitness are less likely to do so. This could be related to the finding that people in fitness-demanding jobs maintain a healthier weight than others, as found by Lakdawalla and Philipson (2002).

We found that workplace injuries are generally positively related to union status, as expected given the results of past studies, and we found a positive but weak relationship between workers' compensation benefits and injuries. With just the year fixed effects, a 10 percentage point increase in the replacement rate is associated with approximately a 0.37 percentage point increase in the likelihood of suffering an injury. However, this effect essentially disappears in the models that include individual fixed effects. Workers in unions are about 1.1-3.5 percentage points more likely to report an injury, and this effect is significant in most specifications.

B. Filing Determinants

Table 6 presents our key results about the determinants of filing behavior conditional on reporting an injury or an illness. In the table, the sample is the set of respondents reporting a work-related health condition within the past year and reporting employment (i.e., nonzero employment earnings) within the past year. For this sample, filing is regressed on health insurance, a union status variable, whether the condition was an injury or an illness, and a host of other demographic characteristics listed in the table. The table is ordered identically to Table 5, with the columns differentiated by the health insurance measure and the heterogeneity structure.

As with Table 5, we estimated both fixed and random effect models to control for individual heterogeneity. We also tested the assumptions underlying the random effect model using a Hausman test. For the models in Table 6, the Hausman test fails to reject the null hypothesis that [[theta].sub.i] is uncorrelated with the covariates. (16) The test comparing the models in Columns 3 and 4 has a p value of 0.9173; the models in Columns 5 and 6 yield a p value of 0.4708, and those in Columns 7 and 8 yield a p value of 0.4848. All these are clearly much higher than conventional levels of significance would allow. Thus, when modeling the decision to file for workers' compensation benefits, the evidence suggests that the random effects model is consistent and more efficient than fixed effects.

The Effect of Health Insurance on Filing. The first four columns of Table 6 show that the effect of employer's offering health insurance is remarkably stable. Injured workers in firms where health insurance is offered are about 14-17 percentage points more likely to file a claim. This is quantitatively robust to individual fixed effects. Therefore, the positive relationship between employer offers of health insurance and filing seems not to be the result of individual heterogeneity.

When we controlled for the possession of health insurance, we found that it has a much weaker relationship with the decision to file than did the employer's offer of insurance. It is only associated with about a 4-6 percentage point increase in the probability of filing for workers' compensation benefits in the result of an injury, and the result is only statistically significant in the random effects model. In Columns 7 and 8, where we consider both the offer and the possession of health insurance, we see that the offer of insurance dominates possession in explanatory power: employer offers are associated with a 13-19 percentage point increase in the probability of filing a claim, but the actual possession of insurance has virtually no ability to explain filing. Taken together, these results suggest that the mere possession of health insurance is not nearly as relevant for filing as whether the employer offers it.

There are two equally plausible explanations for the importance of employer offers. The first--a causal interpretation--is that offering health insurance makes employers less likely to discourage workers' compensation claims because it gives workers less incentive to pass injuries off onto health insurers. (17) The second--a selection-based interpretation--is that employers who offer health insurance are of a type that encourages workers' compensation claims, health insurance, and other fringe benefits. Either or both of these effects may be operating, hut it seems clear that heterogeneity across workers is not the sole reason for differences in filing behavior.

A selection-based interpretation is aided by the fact that other types of fringe benefits--like paid vacation days--also seem to be associated with higher filing rates. For example, when both health insurance offers and paid vacation offers are included in a random effects specification, both variables are significant (at the 95% level) and have positive coefficients (about 0.10 for the offer of health insurance and 0.07 for the offer of paid vacations). This suggests that workers who switch to "good employers" are more likely to file claims and that being a good employer is not entirely encapsulated by the health insurance offer. It is apparently not the case that health insurance alone makes an employer less concerned with discouraging workers' compensation claims.

Other Determinants of Filing Behavior. The only other direct measure of employer heterogeneity in our sample is the size of the establishment at which an individual works. It is not clear, in principle, whether large firms should be more or less receptive to claim filing. Larger firms tend to have higher wage earners and thus might be more willing and able to accommodate injured workers. This could indicate a more positive workplace environment and be correlated with the same kinds of employer heterogeneity as the offer of health insurance. On the other hand, larger firms are also more likely to be experience rated, which would make their costs of workers' compensation more sensitive to the claims histories of their workers. This means that larger firms could suffer more from a successful workers' compensation claim, making them less receptive to claims. Since these effects offset, the overall relationship is an empirical question.

Figure 1 illustrates the estimated coefficients and 95% confidence intervals of the establishment size categories on filing in the individual random effect and fixed effect regressions that include the employer offer of health insurance variable (i.e., Columns 3 and 4 in Table 6). The omitted category is the set of firms with one to four employees. The figure shows that filing is more common in establishments with five or more employees, compared to the smallest establishments. However, there are no consistent effects of size among the set of establishments with five plus employees. After controlling for individual heterogeneity, therefore, establishment size plays a relatively limited role in filing, discriminating only between the very smallest firms and all others. This is consistent with the findings of Abowd, Kramarz, and Margolis (1999), who noted the substantial role played by employees in heterogeneity among firms of different sizes. Though not reported in the tables, we found similar results for the other specifications, as well as in the analysis of injury rates reported in Table 5. (18)

[FIGURE 1 OMITTED]

The other determinants of filing behavior are generally consistent with our expectations. Belonging to a union or an employee association raises the probability of filing by about 4%-9% points. This is consistent with the findings of Hirsch, Macpherson, and DuMond (1997), who found that workers were more likely to file in union workplaces than nonunion workplaces. Their argument, which seems plausible here, is that unions provide more information about filing and lower the costs of filing for their members.

The effect of unionization is not as robust once we control for individual heterogeneity. This could be due to a lack of statistical power because individuals do not change union status very often. Table 7 shows the number of transitions between health insurance status, union status, and injury or illness status among repeat injury sufferers. These transitions identify the fixed effects, and to a lesser extent the random effects, models. More than a third of repeat injury sufferers switch health insurance status across injuries, but considerably less (24%) switch injury/illness status, and just 14% switch union status. Thus, the union coefficient in the fixed effects model is identified by approximately 151 observations. The random effects model suggests that union status has a positive effect on filing of about 5 or 6 percentage points.

Workers with occupational illnesses are about 11-18 percentage points less likely to file claims than those with injuries. Table 8 provides some insight into the composition of injuries and illnesses in the NLSY. Just more than 20% of work-related conditions in the NLSY are classified as illnesses. Just as in the union case, the coefficient on illness is insignificant in the individual fixed effects model, probably because there are relatively few observations identifying it (according to Table 7). However, the random effects model, which we are unable to reject, produces an estimate that is significant and similar to the simpler specifications.

We found no statistically significant effect of workers' compensation benefits on the decision to file for workers' compensation benefits, contrary to our expectations and past findings. Selection bias is unlikely to explain this result: if anything, most such biases would overstate the impact of the replacement rate. There are two possible explanations. First, our measure of workers' compensation generosity may lack sufficient variation to have any explanatory power: the caps on weekly wages are not binding for approximately 43% of the injured workers in our sample, and for these workers, there will be no within-state variation in the replacement rate. A second possibility is measurement error due to income taxes. Workers base their behavior on the after-tax replacement rate of lost income, but we measure only the pretax replacement rate, which could bias our estimates towards 0.

People who miss more workdays are also more likely to file. One way to interpret this is to argue that people with more severe injuries that keep them out longer are more likely to file, but some caution is warranted here. People who file are intrinsically more likely to receive benefits, and this may induce them to stay out of the workforce longer than people who do not file. This is unlikely to affect the estimated effect of short (1-2 d) stays out of work since nearly all states employ workers' compensation waiting periods of 3 d or more. The comparison between people who missed no work and those who missed 1-2 d of work thus is likely to reflect differences in severity, but the longer stays may also incorporate simultaneity bias since benefit receipt can affect incentives to return to work.

Finally, we found that college graduates are less likely to file than other educational groups. Ex ante, it is difficult to say what the expected effect of education on the filing decision is. We might think that more educated workers would have more knowledge about the benefits they are entitled to, thus making them more likely to file. On the other hand, they may end up with less severe injuries or receive lower effective replacement rates (because the statutory caps on benefits mean that higher wages are associated with lower replacement rates), making them less likely to file. A less direct interpretation of the result is that college graduates are healthier than others and healthier people file less often. We explore this issue further in the next section.

C The Relationship Between Health and Health Insurance

Recent research by Biddle and Roberts (2003) suggests that workers who are healthier are less likely to file workers' compensation claims than those who are less healthy. If working for an employer offering health insurance is simply an indicator of poor health, our results might be an indirect special case of this result. To rule this out, we investigated whether our findings for employer-based health insurance could be explained by a correlation between health and filing status. The 1998 NLSY asks questions about health to all respondents older than 40 yr. While this is not enough data to allow us to incorporate health directly into our analysis of filing behavior, we can gain some insight into the relationship between health and health insurance.

Table 9 shows results of an analysis performed for the NLSY population aging more than 40 yr in 1998. These are results from an ordered probit of self-reported health status--where an individual is asked whether his health is excellent (= 1), very good (=2), good (=3), fair (=4), or poor (=5)--on several covariates, including employer-based health insurance. A lower number implies better health. We found no relationship between the offer of health insurance and health, although we found that people who have health insurance are actually in better health than those who lack it. Column 3 does suggest that, conditional on the possession of health insurance, workers who are offered health insurance are in worse health than those who are not, but the relationship is small and not statistically significant. These results suggest that adverse selection in health insurance may be less relevant than heterogeneity; people who are more risk averse or forward looking might both invest more in their health and be more likely to purchase (or choose an employer who offers) health coverage. In any event, there is no evidence that people working at employers who offer health insurance are sicker. Indeed, employers have incentives to avoid such a situation; they would not want to offer health insurance if it would simply attract sicker (and presumably less productive) workers.

While self-reported general health status is used widely as a comprehensive measure of general health (see, e.g., Idler and Benyamini 1997), we also experimented with other measures of health and found exactly the same patterns. These alternative measures included whether the respondent has difficulty climbing stairs and whether the respondent finds it difficult to engage in moderate levels of activities because of his/her health.

V. CONCLUSIONS

We have found that the propensity of injured workers to file workers' compensation claims is affected by certain worker characteristics--like union membership or type of injury--in a predictable fashion but is also independently affected by a worker's health insurance status in ways that are surprising and nuanced. Health insurance is positively related to the probability of filing. What appears more important than simply the possession of health insurance, however, is whether or not it is offered by the employer. This suggests that health insurance acts primarily through the incentives of employers (or perhaps indicates heterogeneity among employers) and has relatively little, if any, effect on the incentives of individuals.

From a policy perspective, these results suggest that efforts to encourage filing, by labor unions or government agencies, should be targeted in part--and perhaps even primarily--at employers. We cannot say for sure whether offering health insurance has a causal effect on employers, but it is clear that employer characteristics and behavior are key determinants of worker filing. It is beyond the scope of this study to identify all the effects associated with various policy prescriptions, but our evidence suggests that incentives or sanctions targeted at employers may be more powerful than incentives targeted at employees. It also appears that policy concerns about workers' compensation being used to cover nonoccupational conditions may be overstated: filing workers are actually more likely to have access to health insurance covering those conditions; moreover, possession of health insurance is not that strongly related to filing behavior, in and of itself. At least as far as we can determine, there is no evidence to suggest that uninsured workers, holding firm characteristics constant, are filing for benefits at an inordinate rate.

Future research on the effects of employer characteristics on workers' compensation filing also seems warranted. Such a research agenda would benefit from matched employer-employee databases; these would allow researchers to identify both employer and employee fixed effects in a single empirical framework. It would also benefit from objective data on workplace health conditions. While it would be costly to compile a nationally representative database with objectively verified data on workplace health conditions, a more affordable and attractive alternative would be to validate injury reports for a randomly chosen subsample of workers. This would provide insight into what it is that "reported" injury rates measure and would allow researchers to refine their interpretation and analysis of variation in reported injury rates.

ABBREVIATIONS

AFQT: Armed Forces Qualification Test

NIOSH: National Institute for Occupational Safety and Health

NLSY: National Longitudinal Survey of Youth

NOIRS: National Occupational Injury Research Symposium

doi:10.1111/j.1465-7295.2006.00011.x

REFERENCES

Abowd, J. M., F. Kramarz, and D. N. Margolis. "High Wage Workers and High Wage Firms." Econometrica, 67, 1999, 251-333.

Azaroff, L. S., C, Levenstein, and D. Wegman. "Occupational Injury and Illness Surveillance: Conceptual Filters Explain Underreporting." American Journal of Public Health, 92, 2002, 1421-9.

Baker, L. C., and A. B. Krueger. "Medical Costs in Workers" Compensation Insurance." Journal of Health Economics, 14, 1995, 531-49.

--. "Medical Costs in Workers' Compensation Insurance: Reply." Journal of Health Economics, 16, 1997, 623-4.

Biddle, J. E., and K. Roberts. "Claiming Behavior in Workers' Compensation." Journal of Risk and Insurance, 70, 2003, 759-80.

Butler, R. J. "Economic Determinants of Workers" Compensation Trends." Journal of Risk and Insurance, 61, 1994, 383-401.

Butler, R. J., and J. D. Worrall. "Claims Reporting and Risk-Bearing Moral Hazard in Workers" Compensation." Journal of Risk and Insurance, 58, 1991, 196-7.

Card, D., and B. P. McCall. "Is Workers Compensation Covering Uninsured Medical Costs--Evidence from the Monday Effect." Industrial and Labor Relations Review, 49, 1996, 690-706.

Chelius, J. R. "The Influence of Workers' Compensation on Safety Incentives." Industrial and Labor Relations Review, 35, 1982, 235-42.

Chelius, J. R., and K. Kavanaugh. "Workers" Compensation and the Level of Occupational Injuries." Journal of Risk and Insurance, 55, 1988, 315-23.

Hirsch, B. T., D. A. Macpherson, and J. M. DuMond. "Workers' Compensation Recipiency in Union and Nonunion Workplaces." Industrial and Labor Relations Review, 50, 1997, 213-36.

Idler, E. L., and Y. Benyamini. "Self-Rated Health and Mortality: A Review of Twenty-Seven Community Studies." Journal of Health and Social Behavior, 38, 1997, 21-37.

Johnson, W. G., M. L. Baldwin, and J. F. Burton. "Why is the Treatment of Work-Related Injuries So Costly? New Evidence from California." Inquiry, 33, 1996, 53-65.

Krueger. A. B. "Incentive Effects of Workers" Compensation Insurance." Journal of Public Economics, 41, 1990, 73-99.

Lakdawalla, D. N., and T. J. Philipson. "Technological Change and the Growth of Obesity." National Bureau of Economic Research Working Paper No. 8946, 2002.

Maier, M., and D. Reinke. "Workplace Injuries and Workers' Compensation Claim Filing: Results from the 2002 Oregon Population Survey." 2004. Accessed January 26, 2005. http://www.cbs.state.or.us/imd/rasums/resalert/wcresults.html.

Meyer, B. D., W. K. Viscusi, and D. L. Durbin. "Workers' Compensation and Injury Duration: Evidence from a Natural Experiment." American Economic Review, 85, 1995, 322-40.

Neuhauser, F., and S. Raphael. "The Effect of an Increase in Workers' Compensation Benefits on the Duration and Frequency of Benefit Receipt." Review of Economics and Statistics, 86, 2004, 288-302.

Roberts, K. "The Structure of and Incentives from Workers' Compensation Pricing," in Workplace Injuries and Diseases: Prevention and Compensation: Essays in Honor of Terry Thomason. Kalamazoo, MI: W.E. Upjohn Institute for Employment Research, 2005, 171-98.

Ruser, J. W. "Workers' Compensation Insurance, Experience Rating, and Occupational Injuries." RAND Journal of Economics, 16, 1985, 487-503.

--. "Self-Correction versus Persistence of Establishment Injury Rates." Journal of Risk and Insurance, 62, 1995, 67-93.

Sengupta, I., V. Reno, and J. F. Burton, Jr. Workers' Compensation: Benefits, Coverage and Costs, 2003. Washington, DC: National Academy of Social Insurance, 2005.

Smith, R. S. "Mostly on Monday: Is Workers' Compensation Covering Off-the-Job Injuries?" in Benefits, Costs, and Cycles in Workers' Compensation Insurance, edited by P. Borba and D. Appel. Norwell, MA: Kluwer, Academic Publishers, 1990, 115-27.

(1.) All 50 states and the federal government have a workers' compensation system, though it is optional in New Jersey and Texas. As of 2003, 14 states had numerical exemptions for firms employing a small (always five or less) number of employees. Additionally, ten states do not cover agricultural workers. Almost all states provide full medical coverage with no cost sharing. Arkansas and Florida are the only exceptions, with the former limiting employer liability when no lost time results and the latter requiring a $10 co-pay after maximum medical improvement. Benefits vary by state, but most provide two-thirds income replacement subject to weekly benefit caps and floors. For more, see the U.S. Department of Labor (http://www.dol.gov/esa/regs/statutes/owcp/stwclaw/stwclaw.htm, accessed April 30, 2004).

(2.) In our data, about 58% of workers who suffer work-related injuries file a claim. This is consistent with the results of Biddle and Roberts (2003).

(3.) Prior to the introduction of workers' compensation, which began early in the 20th century, workers could recover compensation only if they could prove in the tort system that their injuries resulted from employer negligence.

(4.) Arkansas and Florida are the only exceptions, with the former limiting employer liability when no lost time results and the latter requiring a $10 co-pay after maximum medical improvement.

(5.) For more information, see the U.S. Department of Labor (http://www.dol.gov/esa/regs/statutes/owcp/stwclaw/stwclaw.htm).

(6.) These estimates are taken from Sengupta, Reno, and Burton (2005). Total employer costs exceed benefit payments because of administrative costs, loss adjustment costs, taxes, reserves for future benefit payments, profits or losses of private carriers, and contributions to special funds.

(7.) As of 2003, 14 states had numerical exemptions for firms employing a small (always five or less) number of employees. Additionally. ten states do not cover agricultural workers.

(8.) For more on the pricing of workers' compensation insurance, see Roberts (2005).

(9.) See, for example, Chelius (1982), Ruser (1985, 1995). Chelius and Kavanaugh (1988). Krueger (1990), Butler and Worrall (1991), Butler (1994). and Neuhauser and Raphael (2004).

(10.) While we used the weights throughout most of the empirical work that follows, the characteristics of the sample by insurance and filing status do not differ substantially with or without the use of weights.

(11.) The establishment size as reported in the summary statistics reflects top coding at 10,000 to reduce the influence of a few instances where individuals report very large establishment size (more than 100,000). More than 98% of observations report establishments less than 10,000. The top coding is not relevant for the regression analysis because the largest category we report is 3,000 or more employees.

(12.) Equation (1) is specified as a linear probability model, although it could easily be written as a latent variable model estimable via a probit or logit specification. Throughout the manuscript, we report the results from linear probability models, as marginal effects from probit and logit models were quite similar in magnitude and statistical significance.

(13.) The regression model makes the important assumption that time-varying individual heterogeneity is absorbed by the model's covariates, such as age and income. If, on the other hand, individuals raise their demand for "good employers" at the same time that they demand more injury compensation, this could confound our results. It is not possible to rule out this particular possibility without matched employer-employee data on injuries and claims.

(14.) The effect of health insurance will be causal if [[lambda].sub.j] = 0 and [[mu].sub.ij] = 0 or if corr([H.sub.ij][[lambda].sub.j]j) = 0 and corr([H.sub.ij], [[mu].sub.ij]) = 0. Otherwise, the effect of health insurance will represent the joint effect of health insurance and employer (or match specific) heterogeneity.

(15.) This does not survive individual fixed effects, but this is probably because relatively few people switch educational attainment over the time frame we are investigating in the NLSY (when the sample is between the ages of 23 and 35 yr).

(16.) The test statistic confirms that the random effects model is more efficient than the fixed effects models: this is a condition for the validity of the test.

(17.) An alternate explanation is that employers with health insurance actually encourage workers to file workers' compensation claims, which they might do if they were trying to reduce the cost of health insurance. This is unlikely for the reasons discussed earlier that workers' compensation is associated with additional indemnity benefits not covered by health insurance and medical treatment under workers' compensation tends to be more expensive.

(18.) We might wonder if employer size matters more than establishment size. if large employers have different practices that are common across all establishments. To address this, we duplicated the regressions in Columns 3 and 4 of Table 6 (not reported here) including a variable indicating whether or not the current employer had at least 1,000 workers at other establishments (the only way to capture employer size in the NLSY). The point estimate for this variable was slightly negative but was statistically insignificant at even the 10% level in both specifications.

DARIUS N. LAKDAWALLA, ROBERT T. REVILLE, and SETH A. SEABURY *

* We are grateful to Les Boden, Karen Roberts, two anonymous referees, and participants in the 2003 NOIRS meeting in Pittsburgh, Pennsylvania, for their helpful comments and suggestions. We are also grateful to the National Institute for Occupational Safety and Health (NIOSH) for research support (Grant 1R03OH07619). The opinions expressed are those of the authors and do not represent those of NIOSH or any institution the authors are affiliated with.

Lakdawalla: Economist, RAND, and Faculty Research Fellow, National Bureau of Economic Research, 1776 Main Street, Santa Monica, CA 90407. Phone 1-310-393-0411, Fax 1-310-260-8155, E-mail [email protected]

Reville: Director, RAND Institute for Civil Justice, 1776 Main Street, Santa Monica, CA 90407. Phone 1-310-393-0411, Fax 1-310-451-6979, E-mail [email protected]

Seabury: Economist, RAND, 1776 Main Street, Santa Monica, CA 90407. Phone 1-310-393-0411, Fax 1-310-451-6979, E-mail [email protected] TABLE 1 Health Insurance and Workers' Compensation Claiming Worker Employer Has Health Insurance Offers Health Insurance Yes No Yes No Proportion filing 0.60 0.53 0.62 0.46 Proportion losing wages 0.27 0.45 0.27 0.39 Average work 22.40 38.20 22.70 25.50 days lost Notes: Data are from the 1988-1998 NLSY. Only individuals who report suffering a work-related injury in the previous year were included. TABLE 2 Occupational Injuries and Illnesses in the NLSY Year Number of Workers with Injuries/Illnesses Claims Benefits 1988 849 411 188 1989 614 340 150 1990 620 354 174 1992 563 349 174 1993 449 249 116 1994 469 253 115 1996 610 367 157 1998 563 334 153 Total 4,737 2,657 1,227 Year Proportion of Injured With Claims Benefits 1988 0.49 0.23 1989 0.56 0.25 1990 0.57 0.28 1992 0.63 0.32 1993 0.56 0.26 1994 0.54 0.25 1996 0.61 0.26 1998 0.60 0.27 Total 0.56 0.26 TABLE 3 Weighted Summary Statistics of Characteristics of the NLSY Cohort, 1988-1998 Variable Overall Employed in Past Year Age 31.62 (3.93) 31.63 (3.95) Female 0.49 (0.50) 0.46 (0.50) Black 0.14 (0.35) 0.13 (0.34) Married 0.59 (0.49) 0.59 (0.49) Employed within past 2 wk 0.79 (0.41) 0.90 (0.31) Highest grade attained 13.25 (2.55) 13.39 (2.55) Annual wage income ($) 26,587 (29,715) 31,721 (29,844) Employer offers 0.77 (0.42) 0.78 (0.42) health insurance Has health insurance 0.82 (0.38) 0.83 (0.37) Unionized 0.10 (0.30) 0.12 (0.32) Number of employees 533.61 (1,694.29) 541.36 (1,694.02) Missed work due to an injury 0.04 (0.19) 0.04 (0.20) Number of work days missed 1.45 (21.85) 1.47 (19.81) due to injury Injured Filing Variable 31.57 (4.07) 31.76 (4.05) Age 0.36 (0.48) 0.33 (0.48) Female 0.11 (0.32) 0.13 (0.32) Black 0.55 (0.50) 0.58 (0.49) Married 0.88 (0.32) 0.88 (0.32) Employed within past 2 wk 12.54 (2.26) 12.30 (2.13) Highest grade attained 27,409 (19,572) 27,479 (17,946) Annual wage income ($) 0.77 (0.42) 0.83 (0.38) Employer offers health insurance 0.79 (0.40) 0.82 (0.39) Has health insurance 0.19 (0.39) 0.23 0.42) Unionized 472.98 (1,565.88) 462.40 (1,550.93) Number of employees 0.56 (0.50) 0.65 (0.48) Missed work due to an injury 21.46 (81.48) 31.45 (98.89) Number of work days missed due to injury Variable Compensated Age 31.76 (4.02) Female 0.33 (0.47) Black 0.13 (0.33) Married 0.58 (0.49) Employed within past 2 wk 0.83 (0.38) Highest grade attained 12.13 (2.10) Annual wage income ($) 25.937 (17,338) Employer offers 0.83 (0.38) health insurance Has health insurance 0.78 (0.41) Unionized 0.22 (0.41) Number of employees 776.56 (1,619.34) Missed work due to an injury 0.87 (0.34) Number of work days missed 59.75 (131.69) due to injury TABLE 4 Characteristics of the Injured Population, by Health Insurance and Filing Status Employer Offers Health Insurance Variable Filed Did Not File Age 31.91 (4.08) 31.29 (4.05) Female 0.35 (0.48) 0.37 (0.48) Black 0.11 (0.31) 0.12 (0.33) Married 0.60 (0.49) 0.54 (0.50) Employed within past 2 wk 0.91 (0.29) 0.93 (0.25) Highest grade attained 12.38 (2.11) 13.08 (2.39) Annual wage income ($) 29,794 (17,674) 32,387 (20,930) Has health insurance 0.90 (0.30) 0.91 (0.29) Unionized 0.27 (0.44) 0.20 (0.40) Number of employees 501.9 (1,556.20) 635.1 (1,685.2 Missed work due to an injury 0.64 (0.48) 0.42 (0.49) Number of work 28.87 (93.19) 7.78 (32.24) days missed due to injury Lost any wages 0.30 (0.46) 0.18 (0.38) due to injury Employer Does Not Offer Health Insurance Variable Filed Did Not File Age 31.32 (3.87) 31.32 (4.13) Female 0.37 (0.48) 0.37 (0.48) Black 0.12 (0.32) 0.11 (0.31) Married 0.48 (0.50) 0.48 (0.50) Employed within past 2 wk 0.79 (0.04) 0.80 (0.40) Highest grade attained 11.76 (2.10) 12.21 (2.25) Annual wage income ($) 17,671 (13,406) 18,598 (17,958) Has health insurance 0.43 (0.50) 0.47 (0.50) Unionized 0.08 (0.27) 0.03 (0.18) Number of employees 341 (1,654.8) 256 (1,425.7) Missed work due to an injury 0.71 (0.46) 0.46 (0.50) Number of work 40.00 (114.11) 12.80 (80.15) days missed due to injury Lost any wages 0.45 (0.50) 0.31 (0.46) due to injury Notes: Sample limited to those with an injury and some earnings in the past year. TABLE 5 Determinants of Self-Reported Workplace Injuries and Illness in the NLSY Experienced Workplace Injury or Illness Dependent Variable (1) (2) (3) Employer offers health 0.008 ** 0.009 ** 0.004 insurance (0.004) (0.004) (0.005) Has health insurance In union or employee 0.(135 *** 0.026*** 0.012 association (0.005) (0.005) (0.008) Obese 0.009 ** 0.007 * 0.012 * (0.004) (0.004) (0.007) Replacement rate 0.037 *** 0.022 ** -0.011 (0.010) (0.Oll) (0.019) High school graduate -0.008 ** -0.007 -0.030 * (0.004) (0.004) (0.016) College attendee -0.022 *** -0.021 *** -0.020 (0.005) (0.005) (0.016) College graduate -1.042 *** -0.038 *** -0.021 (0.005) (0.005) (0.018) Physical demands -0.005 * -0.005 * -1.007 * (0.003) (0.003) (0.004) Strength rating 0.012 *** 0.010 *** 0.011 *** (0.003) (0.003) (0.004) Fixed effects Year Year Year Industry Industry State Individual Random effects Other controls Weeks of job tenure, AFQT score, race, gender, age, establishment size Observations 45,357 45,317 45,313 [R.sup.2] 0.27 0.28 0.47 Experienced Workplace Injury or Illness Dependent Variable (4) (5) (6) Employer offers health 0.010 *** insurance (0.003) Has health insurance 0.013 ** 0.009 *** (0.006) (0.003) In union or employee 0.021*** 0.013 * 0.019 *** association (0.004) (0.008) (0.004) Obese 0.010 *** 0.008 0.010*** (0.003) (0.008) (0.003) Replacement rate 0.022 * 0.001 0.024 ** (0.012) (0.018) (0.012) High school graduate -0.010 *** -0.030 * -0.012 *** (0.004) (0.017) (0.004) College attendee -0.016 *** -0.017 -0.017 *** (0.004) (0.016) (0.004) College graduate -0.034*** -0.027 -0.031 *** (0.005) (0.018) (0.005) Physical demands -0.001 -0.004 0.000 (0.002) (0.004) (0.002) Strength rating 0.004 ** 0.009 ** 0.004 ** (0.002) (0.004) (0.002) Fixed effects Year Year Year Industry Industry Industry State State State Individual Random effects Individual Individual Other controls Weeks of job tenure, AFQT score, race, gender, age, establishment size Observations 45,313 42,180 42,180 [R.sup.2] 0.29 0.49 0.30 Experienced Workplace Injury or Illness Dependent Variable (7) (8) Employer offers health 0.006 0.009 ** insurance (0.006) (0.004) Has health insurance 0.010 0.005 (0.007) (0.004) In union or employee 0.011 0.018 *** association (0.008) (0.004) Obese 0.012 0.012 *** (0.008) (0.003) Replacement rate -0.01 0.022 * (0.020) (0.013) High school graduate -0.029 * -0.012 *** (0.017) (0.004) College attendee -0.018 -0.017 *** (0.016) (0.005) College graduate -0.025 -0.034 *** (0.019) (0.005) Physical demands -0.006 -0.001 -(0.004) (0.002) Strength rating 0.010 *** 0.005 ** (0.004) (0.002) Fixed effects Year Year Industry Industry State State Individual Random effects Individual Other controls Weeks of job tenure, AFQT score, race, gender, age, establishment size Observations 38,838 38,838 [R.sup.2] 0.49 0.29 Notes: Robust standard errors are given in parentheses. Time-invariant controls were excluded from fixed effects models. AFQT, Armed Forces Qualification Test. * Significant at 10'%; ** significant at 5'%; *** significant at 1%. TABLE 6 Determinants of Filing a Claim Among Injured NLSY Workers Filed a Workers' Compensation Claim Dependent Variable (1) (2) Employer offers 0.167 *** (0.026) 0.157 *** (0.027) health insurance Has health insurance In union or 0.071 *** (0.023) 0.070 *** (0.025) employee association Replacement rate 0.014 (0.091) 0.036 (0.093) Occupational illness -0.180 *** (0.028) -0.184 *** (0.028) No. of missed workdays 1-2 0.061 ** (0.030) 0.051 * (0.030) 3-9 0.153 *** (0.028) 0.145 *** (0.028) 10-29 0.267 *** (0.031) 0.262 *** (0.031) 30-129 0.264 *** (0.032) 0.255 *** (0.033) 130+ 0.407 *** (0.034) 0.410 *** (0.038) High school 0.017 (0.023) 0.028 (0.023) graduate College attendee -0.007 (0.030) 0.008 (0.031) College graduate -0.094 ** (0.039) -0.054 (0.041) Fixed effects Year Year Industry Random effects Other controls Weeks of job tenure. AFQT score, race, gender, age, establishment size, job physical demands, job strength rating, wage loss Observations 3,645 3,643 [R.sup.2] 0.13 0.22 Dependent Variable (3) (4) Employer offers 0.149 * (0.077) 0.135 *** (0.022) health insurance Has health insurance In union or 0.085 (0.098) 0.056 ** (0.024) employee association Replacement rate 0.108 (0.326) 0.101 (0.091) Occupational illness -0.110 (0.077) -0.157 *** (0.024) No. of missed workdays 1-2 0.086 (0.078) 0.094 *** (0.024) 3-9 0.128 * (0.073) 0.174 *** (0.024) 10-29 0.299 *** (0.086) 0.286 *** (0.029) 30-129 0.205 ** (0.093) 0.317 *** (0.028) 130+ 0.400 *** (0.115) 0.430 *** (0.044) High school 0.055 (0.225) 0.030 (0.021) graduate College attendee 0.052 (0.203) 0.014 (0.028) College graduate 0.151 (0.246) -0.073 ** (0.037) Fixed effects Year Year Industry Industry State State Individual Random effects Individual Other controls Weeks of job tenure. AFQT score, race, gender, age, establishment size, job physical demands, job strength rating, wage loss Observations 3,643 3,643 [R.sup.2] 0.82 0.22 Dependent Variable (5) (6) Employer offers health insurance Has health 0.042 (0.083) 0.062 *** (0.023) insurance In union or 0.044 (0.117) 0.053 ** (0.026) employee association Replacement rate 0.070 (0.400) 0.012 (0.097) Occupational illness -0.141 (0.090) -0.169 *** (0.026) No. of missed workdays 1-2 0.091 (0.099) 0.079 *** (0.027) 3-9 0.195 ** (0.092) 0.158 *** (0.025) 10-29 0.332 *** (0.102) 0.275 *** (0.031) 30-129 0.224 ** (0.097) 0.292 *** (0.030) 130+ 0.421 *** (0.132) 0.433 *** (0.045) High school 0.020 (0.267) 0.042 * (0.023) graduate College attendee -0.177 (0.290) 0.020 (0.030) College graduate 0.080 (0.380) -0.083 ** (0.039) Fixed effects Year Year Industry Industry State State Individual Random effects Individual Other controls Weeks of job tenure. AFQT score, race, gender, age, establishment size, job physical demands, job strength rating, wage loss Observations 3,177 3,177 [R.sup.2] 0.85 0.23 Dependent Variable (7) (8) Employer offers 0.185 * (0.102) 0.129 *** (0.026) health insurance Has health -0.023 (0.096) 0.022 (0.026) insurance In union or 0.048 (0.120) 0.046 * (0.026) employee association Replacement rate 0.099 (0.411) 0.059 (0.100) Occupational illness -0.144 (0.100) -0.165 *** (0.026) No. of missed workdays 1-2 -0.134 (0.103) 0.084 *** (0.027) 3-9 0.194 ** (0.097) 0.163 *** (0.026) 10-29 0.358 *** (0.104) 0.274 *** (0.032) 30-129 0.237 ** (0.102) 0.284 *** (0.031) 130+ 0.426 *** (0.136) 0.418 *** (0.047) High school -0.010 (0.294) 0.047 ** (0.023) graduate College attendee -0.160 (0.294) 0.013 (0.030) College graduate 0.045 (0.395) -0.087 ** (0.040) Fixed effects Year Year Industry Industry State State Individual Random effects Individual Other controls Weeks of job tenure. AFQT score, race, gender, age, establishment size, job physical demands, job strength rating, wage loss Observations 3,003 3,003 [R.sup.2] 0.86 0.23 Notes: Robust standard errors are given in parentheses. Time-invariant controls were excluded from fixed effects models. AFQT. Armed Forces Qualification Test. * Significant at 10%; ** significant at 5%; *** significant at 1%. TABLE 7 Changes in Health Insurance, Union, and Injury/Illness Status Among Repeat Sufferers Number of Injuries 2 3 4 5 6 7 8 Total Insurance 0 511 150 47 14 5 3 1 731 switches 1 218 62 25 7 0 0 0 312 2 0 38 12 7 1 0 0 58 3 0 0 1 1 0 0 0 2 Total 729 250 85 29 6 3 1 1,103 Union 0 649 210 67 24 0 2 0 952 switches 1 80 31 6 4 2 0 0 123 2 0 9 11 1 3 1 1 26 3 0 0 1 0 1 0 0 2 Total 729 250 85 29 6 3 1 1,103 Disease 0 591 168 57 17 4 1 1 839 switches 1 138 54 19 4 1 1 0 217 2 0 28 8 8 1 1 0 46 3 0 0 1 0 0 0 0 1 Total 729 250 85 29 6 3 1 1,103 TABLE 8 Composition of Injuries and Illnesses in the NLSY by Year (1988-1998) Year Condition 1988 1989 1990 1992 Injury Open wound 171 148 102 84 Burn 38 25 24 12 Other visible injury 115 59 85 52 Upper extremity sprain 30 21 20 25 Back sprain 53 37 28 42 Other sprain 88 115 107 78 Fracture 52 38 32 42 Illness Back musculoskeletal disease 16 10 8 6 Upper extremity 3 11 13 5 musculoskeletal disease Other musculoskeletal disease 25 24 30 14 Mental illness 6 6 8 6 Other disease 119 65 65 67 Not elsewhere classified 128 51 84 123 Total 844 610 606 556 Year Condition 1993 1994 1996 1998 Injury Open wound 71 62 92 61 Burn 17 10 17 13 Other visible injury 48 45 42 46 Upper extremity sprain 19 26 34 25 Back sprain 32 32 31 29 Other sprain 65 70 77 62 Fracture 37 44 43 40 Illness Back musculoskeletal disease 9 10 9 13 Upper extremity 20 10 10 13 musculoskeletal disease Other musculoskeletal disease 11 20 21 19 Mental illness 7 5 5 8 Other disease 51 57 95 119 Not elsewhere classified 62 77 101 102 Total 449 468 577 550 TABLE 9 Health Status and Filing Determinants Ordered Probit on Self-Reported General Health Status (a) Employer offers -0.013 (0.108) health insurance Has health insurance -0.203 * (0.120) In union -0.089 (0.117) -0.067 (0.113) or employee association High school graduate 0.004 (0.112) -0.069 (0.106) College attendee -0.134 (0.127) -0.130 (0.121) College graduate -0.374 *** (0.143) -0.398 *** (0.137) Age 0.029 (0.089) 0.068 (0.083) Hispanic origin -0.122 (0.129) -0.140 (0.122) Black -0.128 (0.104) -0.167 * (0.099) Female 0.115 (0.082) 0.085 (0.077) AFQT score -0.007 *** (0.002) -0.007 *** (0.002) Observations 938 1.055 Pseudo [R.sup.2] 0.03 0.03 Ordered Probit on Self-Reported General Health Status (a) Employer offers 0.085 (0.119) health insurance Has health insurance -0.301 ** (0.152) In union -0.078 (0.116) or employee association High school graduate 0.021 (0.111) College attendee -0.111 (0.128) College graduate -0.350 ** (0.143) Age 0.034 (0.089) Hispanic origin -0.128 (0.131) Black -0.128 (0.104) Female 0.122 (0.082) AFQT score -0.007 *** (0.002) Observations 938 Pseudo [R.sup.2] 0.03 Notes: Robust standard errors are given in parentheses, below probit coefficients. AFQT. Armed Forces Qualification Test. (a) General health status takes on values of excellent, very good, good, fair, or poor, where "excellent" takes on a value of 1 and "poor" takes on a value of 5. * Significant at 10%; ** significant at 5%; *** significant at 1%.