1 Worker Injuries: The Effects of Workers Compensation and OSHA Inspections Leon S. Robertson, Yule University, and J. Philip Keeve? Naval Research Laboratory Abstract. Detailed analysis of work exposure and worker attributes failed to explain changes in injury claims in three plants, each in a separate state, during When relative exposure to hazard and worker attributes are controlled, changes in injury rates are largely explained by increases in Workers Compensation greater than inflation and by inspections made by the Occupational Safety and Health Administration (OSHA). Rises in Workers Compensation greatly increased the claims for injuries, while OSHA citations substantially decreased objectively verifiable injuries in the year following an inspection. Data on days lost due to injury in 167 industrial groupings in 20 states also indicate significant reductions of workdays lost in correlation with OSHA inspections, when increases due to Workers Compensation are controlled. The results thus appear to be generalizable. Previous studies of OSHA failed to adequately control for Workers Compensation effects, and thus underestimated the effects of OSHA. Government has attempted to prevent workplace injuries or reduce their severity by establishing safety standards and then carrying out inspections of workplaces for violations of the standards. The federal imposition and enforcement of standards has been the responsibility of the Occupational Safety and Health Administration (OSHA) since Fines may be imposed for violations, particularly if they are serious or repeated. Should an injury result in lost worktime, the employer is required by state regulations to insure that the worker is compensated according to a schedule of payments for particular disabilities (Workers Compensation). These payments vary from state to state, and there has been a trend in recent years toward increasing them at substantially more rapid rates than the rate of inflation. Large variations in incidence and severity of injuries are found among industries, and among plants and departments within the same industry. This study was supported by a grant to Yale University by the Atlantic Richfield Corporation. The analysis and interpretation of the data are the sole responsibility of the authors and do not necessarily represent the views of the University or Atlantic Richfield. Journal of Health Politics, Policy and Law, Vol. 8, No. 3, Fall by the Dept. of Health Administration, Duke University. 581
2 582 Journal of Health Politics, Policy and Law According to the Bureau of Labor Statistics, the construction industry has the highest injury incidence rate, followed closely by manufacturing. Injuries resulting in one or more lost workdays are highest in manufacturing. Incidence as a percentage of all full-time equivalent workers was virtually constant during the 1970s, but lost workdays increased from less than 45 to more than 60 per hundred workers. These gross trends give the appearance that OSHA has had little effect on incidence, and even a possible adverse effect on lost workdays. Research studies on effects of the OSHA standards on worker injuries have also reported little or no effect. These studies, which have used a variety of strategies to control for problems with the injury data and for other factors that may influence fluctuations in worker injuries, are summarized here, and a major shortcoming in all of them is described. The criteria for reporting injuries were changed in 1970 when the Occupational Safety and Health Act was enacted. Thus before-after comparisons of aggregate series of injuries are of questionable validity. Mendeloff attempted to overcome this problem by looking at before-and-after rates of change in injuries, rather than absolute injury rates. Factors controlled in his model included new-hire rates, percent males 18-24, real hourly earnings, and general trend. He also attempted to separate those injuries that OSHA inspections were more likely to prevent. Comparing projected injury rates based on these factors to actual rates through 1974, he estimated that OSHA accounted for less than a 3 percent reduction in injuries in California.2 Smith compared data on industries with high injury rates that were targeted by OSHA for frequent inspections with others, controlling for prior rates and changes in employment. He found no significant improvement in injury rates in the targeted group.3 Di Pietro reached a similar conclusion, based on a lack of correlation between OSHA inspections in 1972 and changes in injury rates from 1972 to 1973 in selected ind~stries.~ But Smith later reported analysis of data on lost-workday injury rates in specific plants during Lagging the potential effects of inspections over three-month periods, he found about a 16 percent reduction in injury rates in 1973 and a 5 percent reduction in 1974, the latter not statistically ~ignificant.~ These results suggested a greater impact of OSHA than had previously been found. Viscusi used aggregated time-series data from in certain industries, and found inspections and violations uncorrelated with injury rates. Possible exogenous factors controlled were percent production workers; age, sex, and racial mixture; and changes in employment, hours worked per week, and overtime. He did find a significant downward trend in injury rates uncorrelated with inspections, which he interpreted as a possible result of the general effect on industry of OSHA s presence, rather than the specific result of inspection and citation activity of the agency. He noted that the actual fines imposed by OSHA are trivial and could not have forced compliance for
3 Robertson & Keeve Worker Injuries 583 strictly economic reasons.6 Bacow suggested that OSHA activities in general could have made workers more interested in injury prevention, independent of inspection activity. But Viscusi claimed that according to utility theory workers in environments made safer would actually take fewer precautions (however, he offered no empirical evidence of actual worker behavior under a variety of hazardous conditions). The failure of time-series models based on aggregated data to accurately forecast economic changes does not recommend them as adequate methodology for evaluation of the effects of regulation. One such model was used to generate expected rates of motor vehicle deaths and, when compared to actual rates, led to the conclusion that motor vehicle safety standards had no net effect on overall death rates.* Lower than projected occupant deaths were offset by higher than projected deaths of other road users, which was interpreted as the result of more intensive driving by those more protected. That invalid conclusion was the result of incorrect assumptions regarding the application of the regulations and the changing structure of the causal m0de1.~ Disaggregated data on all fatal crashes in the U.S. found large reductions in fatality rates attributable to the regulations when actually regulated vehicles were compared to non-regulated vehicles, controlling for use. The regulated vehicles struck fewer, not more, other road users per mile used than did unregulated vehicles, apparently because of crash avoidance standards in the regulation. lo Incorrect specification of the variables in a model often leads to false inferences, and the problem is particularly acute when aggregated data are used to test the model. The present study was undertaken partially to investigate the extent to which correlates of injuries to individual workers are predictive of aggregate rates. In the case of aggregated time-series analysis, the series would show little variation if one force acting to increase it were offset by another acting to decrease it. Because of the lack of variation, each could appear to have little or no effect, when in fact the effect of each is quite strong. In some of the noted studies there is an attempt at rigorous derivation of equations from utility theory that might explain worker and management behavior in response to OSHA. But in each case the actual choice of exogenous variables (those that also could contribute to injury rates) for use in the empirical analyses has been mainly ad hoc, based on availability of data and on some knowledge of their correlation to injury rates in prior studies. Despite the economic orientation of these studies, they failed to consider the possible effects of compensation on injury-related behavior. Chelius, noting a positive correlation between Workers Compensation and injury claims, went so far as to say that if there were 100 percent protection against all losses due to accidents, including full compensation for lost salary, pain, and loss of leisure, a worker would tend to be indifferent to accident prevention. 12 In contrast,
4 584 Journal of Health Politics, Policy and Law a National Commission on State Workmen s Compensation in 1972 recommended increased benefits and more experience-rating partly on the grounds that increased costs of Workers Compensation would be an incentive for employers to reduce hazards. l3 The narrow view that human behavior is driven by a hierarchy of utilities, with a sort of gyroscope in the brain that constantly adjusts behavior to their maximization, ignores a vast body of evidence on the complexity of behavior and its motivation. Certainly individuals have some awareness of risk of injury, and on occasion made some adjustment in behaviors that increase or decrease risks, thus making a marginal difference in injury rates in the aggregate. But the limited opportunities for individuals to obtain precise information on the relative risks of a wide variety of behaviors, and their limited ability to maintain constant attention and to react in adequate time, suggest that the marginal effect of deliberate choice of risks would be small. Injury is frequently related to alcohol and drug use that impairs brain function, in terms of both judgment and ability to react to hazardous conditions; addiction to these substances leads to uncontrolled use. Impulse, habit, and preoccupation with matters other than the task at hand are but a few of the other human characteristics that mitigate against constant vigilance and immediate reaction.14 Of course, the most important factor in injury severity is exposure to sufficient energy to do damage. The virtual lack of severe injuries among ofice personnel is mainly a result of the absence in the ofice environment of exposure to such concentrated energy. The major energy source present in damaging amounts in offices is electricity, which is well shielded from the workers. In factories, there is generally far more exposure to far more energy, with far less shielding. The setting and the design of the study The first phase of the present study involved three plants (in New York, Wisconsin, and Connecticut) that produce mainly metal sheets, tubes, rods, and wire. Factory operations include shredding, casting, shaping, molding, and slitting metals. The primary potential sources of injury are moving parts of machines, metal moving in and out of machines, and heat from furnaces and metals. Forklifts and large cranes, often loaded with tons of material, move about in the plants. In some cases, these machines can be operated by remote controls, but at other times the workers are in close proximity to substantial amounts of mechanical and thermal energy. From the logs submitted from each plant to the Bureau of Labor Statistics, all injuries reportable to the Occupational Safety and Health Administration during the period were coded by date, description, anatomic site, number of days away from work, and number of days of restricted activity. Criteria for reporting an injury on these logs are one or more of the following:
5 Robertson & Keeve Worker Injuries 585 loss of consciousness, restriction of work or motion, transfer to another job, or medical treatment (other than first aid). Personnel files in each plant were examined to determine the types of data available on each worker, the uniformity of the data among plants, and the coding system that would be needed to transfer the data into a form for electronic data processing. Data available from the workers job application forms included birth date, sex, years of formal education, military service, height, weight, and number of jobs held prior to employment with the company. Each file also contained a separate sheet for each time the worker was hired, changed jobs, or was terminated, along with dates and codes for departments and jobs in which the worker was employed. Although the present study, like all research, is limited by the data available, it does include correlations of the data on injuries to individuals with data on their histories and with attributes that are suspected or known to be correlated with injury incidence. Age of the worker is a proxy for a mix of impulsiveness, life experience, and world view: younger people are disproportionately involved in a variety of injuries. Formal education is a proxy for accumulation of knowledge and, controlling for age, interest in and ability for achievement: the better educated have been found to take greater precautions, such as use of seatbelts in cars. Military service and number of jobs held prior to employment in the present job may be somewhat indicative of personality types. Ratio of weight to height is a crude proxy for physical condition and physical ability to react quickly in an emergency. A coding scheme was developed to capture the data, and a research assistant was trained in the use of the files and the method of coding information from the application forms and job-change sheets. The research assistant then traveled to each plant and trained other assistants in the coding procedures. The research assistant participated in the coding and supervised the work of others, resolving any problems or incongruities in data with staff of the personnel departments and in telephone conversations with the principal investigator. Files of Workers Compensation claims were checked against injuries reported to OSHA, so as to be sure that trends were not a result of changes in interpretation of reporting criteria. The data were obtained for virtually every person who worked at an hourly rate in the New York and Wisconsin plants, and for a 50 percent random sample of such workers in the Connecticut plant, during the period. (White-collar workers, who are seldom injured on the job, were excluded from the study.) Analysis was accomplished in three stages. First, for each department the total number of injuries was divided by the number of person-years worked in that department. The result was an expected rate of injuries per year for an individual working in a particular area, given the general level of hazard in that area. Second, the difference between the number of injuries actually
6 586 Journal of Health Politics, Policy and Law experienced by each individual and the number expected for that individual based on time spent in various departments was correlated to worker attributes and background variables. Third, for each of the three plants the effects of departmental exposure to hazards and of individual factors contributing to variation from that expectation were combined and summed for each year, to obtain an expected number of injuries in each year for each separate plant. Actual variation from that expected number was then correlated to increased amounts of Workers' Compensation-corrected for inflation-and to number of OSHA citations. Analyses at the individual and plant levels Fatal injuries are very rare in these plants-only one occurred during the eight years of the study. The numbers of injuries and days lost due to injuries, along with rates per person-years worked during in the major departments, are presented in Table 1. (A person-year can be one person working a year, two persons each working six months, etc.) Among some 2,700 workers with about 11,000 person-years of exposure during the eight-year period, there occurred more than 2,500 injuries-about one injury for every 4.3 person-years (0.23 per year), including one lost-time injury for every ten person-years. Table 1. Rates (per Person-Year Worked) of Injuries and Days Lost in Departments with 50 or More Injuries, * Lost-time Department Injuries Rate Injuries Rate Days Lost Rate Casting Copper Mill Brass Mill Maintenance Copper Tube Rod Drawn Copper Alloy Tube Extrusion Tube Mill Machine Shop All Others** ,112 1,983 1, ,166 1, , Totals 2, , , 'Based on 2,711 workers who worked 1 1,055 person-years during the period **Includes unknown
7 ~~~~ ~ Robertson & Keeve - Worker Injuries 587 Table 2. A Comparison of Apparent Effect of Age on Injuries, Using Two Different Means of Accounting for Exposure Injury Rate Per Person- Year Worked: Teens 20s 30s 40s 50s 60s Average Standard Deviation F = 0.825, df = 5, p > 0.50 Total Injuries Per Person Minus Those Expected from Length of Departmental Exposures: Teens 20s 30s 40s 50s 60s Average Standard Deviation F = , df = 5, p < The injury rates and rates of days lost were calculated for each job classification within the departments, but the jobs were so finely categorized that the rates were too unstable statistically for analysis at the individual job level; only one job classification had more than 30 injuries during the eight years, and that job accounted for only 3 percent of the total injuries. Among departments, the injury rate varied from about one per seven person-years to one per three person-years. A total of 14,977 days of work were lost due to injury-more than a week on average per injury, and about 1.3 days per person-year. The highest rate of days lost in a number of departments is more than three times that of departments with the lowest rates. To control for differential exposure to hazard in different departments, the expected number of injuries for each worker was calculated by multiplying the injuries per person-year in each department where the worker was employed (from Table 1) times the number of years (or proportion of a year) that the worker was in the department. The total number of expected injuries was then subtracted from the actual number of injuries the worker experienced. This analysis was then repeated using data that included only losttime injuries. The decision to analyze variation from departmental averages was based on the premise that, although exposure to hazard is a necessary condition for injury, individual attributes contribute to variations in susceptibility to injury. Also, the technical issue of lack of homoscedasticity would have arisen had individual rates per person-year been used. Table 2 provides a good
8 588 Journal of Health Politics, Policy and Law illustration. It compares the data on effect of workers ages that appears when average injury rates are used (shown at the top of the table) with the effect of workers ages that appears when average number of injuries minus those expected from departmental exposures is examined. When injury rates are used, the standard deviations are highly variable, and age differences in rates are not statistically significant; also, the relation to age is erratic. In contrast, total injuries minus expected injuries yields more uniform distribution around the average in each age group. With the exception of workers in their 60s, the correlation is as expected from previous research, and the differences among age groups are now highly significant. Apparently, newly hired workers in their teens and twenties are assigned to less hazardous departments but, when exposure is taken into account, have higher than expected injuries than the average workers in those departments. Similar results were obtained when the analysis was limited to lost-time injuries. (It is interesting to note that Viscusi did not find a significant correlation of age and injury rates in his analysis of aggregated data, despite the known correlation of age to injuries at the individual level.) To consider the individual attributes and histories with respect to their predictive power, the data were fitted to the regression equation: I = a + b,a + b,e + b,w + b,m + b,n + b6j + b,-,,y + e where:i = actual minus expected injuries A = age E = formal education W = ratio of weight to height M = military service N = number of previous employers J = job changes in the plant before 1973 Y = years in a given department a = constant 6, = increment in injuries related to the i-th factor per unit of the i-th factor, other things being equal e = residual variation Sex of worker was not included in the equation because less than 4 percent (62) of the workers were women-too few to obtain valid results when distributed among all the possible combinations of the other variables. The number of years in a given department was included as a measure of the potential effect of experience on the job over and above exposure reflected in the expected injury variable. The regression coefficients (bi) are presented in Table 3. As anticipated, older workers and those with more years of formal education had fewer than the statistically expected number of injuries. Those with more previous em-
9 Table 3. Regression Analysis of Worker History and Actual Injuries Minus Injuries Expected from Exposure All Injuries Time-Loss Injuries R = R = < < > 0.50 > < > 0.50 <0.001 Factor Regression Coefficient t P Regression Coefficient t P Age Education No. of Previous Employers No. of Job Changes* Military Service Weight/Height Years in Brass Mill Years in Extrusion Years in Casting Years in Alloy Tube Years in Copper Tube Years in Copper Mill Years in Tube Mill Years in Drawn Copper Years in Maintenance Years in Rod k k k ? L L L k L < > > > * _ f _ * f * Within the plant, prior to 1973
10 590 Journal of Health Politics, Policy and Law ployers had greater than expected injuries. The effect of time spent in various departments is rather consistent among departments, reflecting slightly less injury among those with longer tenure in the same department. Weight-toheight ratio, military service, and job shifts in the plant explained no significant variation. (Since weight was obtained at time of first hire, it may not have been an accurate measure of current characteristics. Job changes in the plant are often not voluntary, occurring either because of bumping during periods of layoffs or reassignment depending on work to be done.) Despite the presence of several highly significant factors, the overall equation is not strongly predictive of individual differences in injuries that would not be expected from the average experience in the departments. The multiple correlation coefficient (R) of 0.26 for all injuries and 0.22 for lost-time injuries indicates that less than 7 percent (R2 X 100) of the inter-individual variation from injuries expected from exposure is predicted by the measured worker attributes and histories. The principal question addressed in this report is whether the trend of injuries in these plants during the study period can be explained either by changes in the number of people in more or less hazardous work, or by changes in the work force. To examine the effects of these two factors, an expected number of injuries in each year for each plant was calculated. First, the proportion of a year that each worker was employed in a given department was multiplied times the injury rate per person-year for that department. (Any time not worked, due to either a temporary or permanent termination or to injury, was excluded.) Second, the regression equation was used to calculate for each worker the number of injuries expected above or below those expected from exposure alone. The proportion of the total time worked in a given year was then multiplied by the variation from expected injuries predicted by the worker attributes and history. The first and second values were then added for all workers employed in the given year to obtain for each plant a total expected number of injuries in each year, as predicted by the time worked in given departments by workers with given attributes. To test the extent to which Workers Compensation and OSHA citations explain the variation in aggregate injuries that is not attributable to the cumulative effects of exposure to hazards in different departments and individual attributes or histories, the actual injuries in a year were regressed on expected injuries, maximum Workers Compensation per week corrected for inflati~n, ~ and number of OSHA inspections. The model includes lagged dummy variables for OSHA inspections so as to test for permanance of any effect. The model is: where: N, = number of injuries in year t in each plant E, = expected injuries from exposure and individual variables
11 Robertson & Keeve Worker Injuries 591 W, = ratio of maximum Workers Compensation payment per week in a given state to the GNP price deflator for personal consumption expenditures 0, = 1 if OSHA citation during or just before the beginning of the year, otherwise 0. L, = 1 if OSHA citation the year before, otherwise 0. M, = 1 if OSHA citation two years before, otherwise 0. bi = increment in N per increment in the i-th variable a = constant The coefficients and statistical tests for all injuries are presented in the first column of Table 4. Controlling for expected injuries, the actual number of injuries is significantly higher in association with increments in Workers Compensation above inflation, and significantly lower in years in which the plant received an OSHA inspection. The effect of OSHA is specific to the year and plant, and the effect does not extend to subsequent years. The fit of the data to the model is excellent-94 percent of the variation is explained. Not all injuries, however, are acute events resulting from immediate contact with a hazard. Back strain and pain can develop as the result of years of attempted lifting of weight greater than the musculo-skeletal system can tolerate.i6 Also, in contrast to acute injuries such as lacerations and fractures, cases of pain and strain are usually not objectively verifiable by a physician or nurse. If OSHA inspections result in changes in the workers behavior or environments that contribute to acute conditions, their effects should be more related to objectively verifiable, acute injuries than to strain and pain. Columns two and three in Table 4 provide separate estimates of the effects of the predictor variables on subjective and objective injuries respectively. The data indicate that OSHA inspections had no effect on subjective injuries, but are associated with an average reduction of from 23 to 40 objective injuries per inspection, other factors being equal. In comparison, Workers Compensation is unrelated to objective injuries, but is associated with an increase of from 38 to 60 subjective injuries per doubling of the maximum payment above inflation. The same analysis for lost-time injuries produced almost the same results. Column 4 of Table 4 suggests an increase of about 32 to 58 lost-day injury claims when Workers Compensation doubles above the inflation rate, and a reduction of such claims by about 14 to 33 claims in the year of an OSHA inspection. When subjective and objective injuries are considered separately in columns 5 and 6, the result for subjective lost-day injuries is similar to that for total injuries, except that claims for objective lost-day injuries show an increase associated with rises in Workers Compensation, in addition to the decline associated with OSHA inspections. During the period, the ratio of maximum weekly Workers
12 Table 4. Regression Analysis of Actual Injuries in Relation to Expected Injuries, Workers Compensation, and OSHA Inspections All Injuries Lost-Time Injuries Total Subjective Objective Total Subjective Objective k 6.5 (3.47)* t 5.17 (- 1.01) ( ) R Expected from Exposure and Worker Attributes 1.28 k (1 1.34)* 0.27 k 0.08 (3.29)* ( )* 1.19 k 0.22 (5.3 5)* 0.58 k 0.13 (4.32)* (5.5 6) * Workers Comp./ GNP Deflator (3.64)* (4.2 3) * (0.88)* (3.41)* 22.47? 8.02 (2.80)* First Year After OSHA Inspection (- 2.89)* (- 0.49)* * 8.34 (3.78)* (- 2.56)* k 5.59 (- 1.77) (-3.01)* Second Year After OSHA Inspection (0.84) (- 1.01) (-0.17) (- 1.22) Third Year After OSHA Inspection (0.27) k (- 0.29) 7.25? (0.72) (- 0.26) * 6.38 (- 1.19) (0.29) Figures in parentheses represent t values, significant values 0, < 0.05, df = 18) are indicated by asterisks.
13 Robertson & Keeve Worker Injuries 593 Compensation to the GNP price deflator (1972 = 100) increased in periodic increments from 0.96 to 1.48 in Connecticut, from 0.90 to 1.21 in New York, and from 1.22 to 1.84 in Wisconsin. These changes, through their effect on increased claims for subjective injuries, more than offset the reductions in objective injuries attributable to OSHA citations, and thus would mask OSHA s impact in an examination of aggregated data. The distributions of days lost attributable to injuries among individuals are too skewed for a valid statistical analysis at the individual level. Even yearto-year variation in a single plant can vary by several hundred days if, for example, a couple of injuries early in a year result in two workers not working for the remainder of the year. Thus the effect of OSHA and Workers Compensation on total days lost from injury must be examined by aggregation of a number of plants; but this aggregation must be done within states to account adequately for interstate variation in Workers Compensation. Interindustry and interstate variation To estimate the effects of Workers Compensation and OSHA on workdays lost due to injury, each manufacturing industry (by two-digit SIC code, codes 20-39) that had 5,000 or more production workers in a given state during 1975 was selected for analysis. The statistical instability from small numbers is thus lowered. The lost-day rates by two-digit SIC code in each state for 1975 and 1976 are published, * and OSHA provided a list of the number of inspections in those years for each two-digit industry in those states where data were available. OSHA inspection data were missing from several states, and some states with small populations did not have any industries with 5,000 workers. In total, 167 industries in 20 states were included in the analysis. In addition to maximum Workers Compensation payment2 and number of OSHA inspections, the effects of number of establishments,21 hours per worker,22 and number of production workers were included in the analysis. OSHA inspects establishments, not workers, and the number of inspections per worker-used as the primary data in previous studies-is potentially biased by number of establishments relative to number of workers per industry. Changes in hours per worker and number of production workers could be related to fatigue and inexperience, respectively, which may affect the number of injuries. The effects were estimated on the basis of year-to-year changes because of multicolinearity of the independent variables, which might have distorted the coefficients in a cross-sectional analysis. Injury rates and the decisions to inspect are both related to numbers of workers per establishment and to number of establishments in ways that could distort estimates of effects in a cross-sectional analysis of aggregate rates.
14 594 Journal of Health Politics, Policy and Law The regression equation to which the data were fitted is: a + b,w, + b,o, + b,e, + b,h, + b,p, + e Change in days lost (1976 minus 1975) for injury in the j-th industry of the i-th state. Change in Workers Compensation in the i-th state Change in OSHA inspections of thej-th industry in the i-th state. Establishments of thej-th industry in the i-th state. Change in hours per worker of thej-th industry in the i-th state Change in production workers in thej-th industry in the i-th state. The results are presented in Table 5. As in the plant-level data, both Workers Compensation and OSHA Inspections showed significant effects, but in opposite directions. When a state s Workers Compensation maximum payment doubled above inflationary increases, lost workdays from injury claims increased by some 600 to 900 days in each industrial grouping in the state. Each OSHA inspection resulted in 1.7 to 3.1 fewer days lost due to injury. Number of establishments and changes in numbers of production workers were related to lost workdays, but hours per worker were not. A separate, similar analysis of changes in lost-time injuries produced essentially the same results, although the margin of error in the estimate of OSHA effects was larger than in the estimate of lost days. Considering the potential for biases in the estimation of effects from aggregate data, the results are remarkably consistent with the results at the plant level. Conclusions These results have several implications for public policy and policy analysts. Although the goal of more permanent changes in workplaces, which would Table 5. Regression Analysis of Changes in Days Lost from Injuries, Interstate and Interindustry-20 States in Change (1975 minus 1976) in Days Regression Lost Related To: Coefficient t Workers Compensation/GNP Deflator (1976 minus 1975) t * OSHA Inspections (1976 minus 1975) * Number of Establishments * Hours Per Worker (1976 minus 1975) Production Workers (1976 minus 1975) 0.9 t * *p <0.01 (d = 161), R = 0.57
15 Robertson & Keeve 9 Worker Injuries 595 extend the effect of an OSHA inspection beyond the year after it occurred, has apparently not been realized, the enforcement of OSHA standards did result in significantly reduced injuries and fewer days lost from injury in the year after each inspection. This apparent success was offset in the aggregate by increases in Workers Compensation above inflation; and in the case of lost workdays, the effect was more than offset. The relatively large reduction in injuries following an OSHA inspection could not have been simply a result of corrections of the relatively few hazards for which citations occurred in the three plants studied. Apparently, the embarrassment of being cited, or the fear of a large fine for repeated citation, resulted in general management attention (for a time) to hazardous conditions or worker behavior. The actual fines amounted to only a few hundred dollars each, and in some cases more money was spent in travel and time in an effort to reduce the fines than would have been spent to pay the fine. A review of correspondence associated with OSHA citations in the three plants revealed great concern for the cost of any suggestion by OSHA for modification to the workplace, and no reference whatever to the cost in compensation to the workers when hazards result in injuries. Prior to the period of the study, the company had adopted required use of safety glasses, helmets, and special shoes, although whether this was for economic or humanitarian reasons is not known. Unsystematic interviews with managers revealed no instance in which Workers Compensation costs had been factored into decisions about capital expenditures for less hazardous environments or equipment, the cost of which might have been more than offset by savings in Workers Compensation. (The company studied was self-insured, and was paying around $1OO,OOO per month for Workers Compensation in the three plants during 1980.) Apparently, future decisions to increase Workers Compensation ought to be based solely on consideration of fair compensation for injuries incurred and for associated lost worktime. There is no evidence that increasing Workers Compensation is an incentive for management to reduce hazards. Although the standard litany in neoclassic economists writings on behavior of the firm tells us that managers will act to reduce injuries to the extent that marginal costs of Workers Compensation and other reductions in profit are offset by the marginal costs of injury reduction,23 no evidence was found to support such beliefs. Indeed, increasing the costs to the firm of compensable injuries has a multiplier effect on the firm s expenses, with no benefit in reduced injuries: incrementing injury compensation to workers at a rate greater than inflation substantially increased both the number and the cost of injury claims. The data are insufficient to conclude that the additional claims are unjustified. Most of the claims of unverifiable pain and strain were for back injuries. Although the increases in compensation may have stimulated additional claims for back problems that actually originated outside the workplace, or
16 596 Journal of Health Politics, Policy and Law may simply have provided greater incentive for malingering, it is also quite conceivable that workers with painful backs continued to work in spite of legitimate occupational injuries when Workers Compensation was insufficient to support them and their families. Total incidence of verifiable injuries was not related to Workers Compensation, but objectively verifiable time-loss injuries did increase in relation to increased compensation. Although those who believe that degree of worker caution is related to level of compensation may take heart in the latter finding, the lack of correlation to total incidence suggests an alternative explanation: workers may be more demanding of time off for an injury when the compensation is such that they can afford the time off, whereas they might continue to work with a similar injury when compensation is inadequate. The issue of the effect of compensation on caution would be better resolved by actually observing a set of relevant precautions taken by workers performing similar activities in states with widely differing compensation schedules. In this as in many other cases, both social science and public policy would be better served if social scientists were more cautious about imputing individual motives solely on the basis of theory or aggregated data. If a social scientist misspecifies an equation or makes an error in a computer program, he or she suffers no ill consequence, other than embarrassment if the error is discovered by others. If a worker in a metal-working plant inadvertently places a hand in the wrong place, the hand can be disabled or even lost. Social scientists might make fewer mistakes in generalizations about workers and managers if they would spend more time in workplaces. Both time spent in the workplace and properly disaggregated data lead to our final conclusion: OSHA inspections do appear to work in the short term; and, primarily, they seem to work not by permanently correcting specific safety violations, but by temporarily instilling a greater consciousness of transient factors related to injuries. While the analysis presented here does not make entirely clear whether an increase in the frequency of OSHA inspections would bring about a further reduction in the number of injuries (since at some level the increase in corrective actions might diminish as inspections became routine), it certainly does indicate that OSHA inspections are efficacious. That a further reduction of injuries in the workplace would occur, all other things being equal, if inspections were made more frequently, is a hypothesis well worth testing. Notes 1. Bureau of Labor Statistics, Occupational Injuries and Illnesses in the United States by Industry, 1978 (Washington, D.C.: US. Department of Labor, 1980). 2. John Mendeloff, Regulating Safety: An Economic and Political Analysis of Occupational Safety and Health Policy (Cambridge, Mass.: The MIT Press, 1979). 3. Robert S. Smith, The Occupational Safety and Health Act: Its Goals and Its Achievements (Washington, D.C.: American Enterprise Institute for Public Policy Research, 1976).
17 Robertson & Keeve Worker Injuries Aldona Di Pietro, An Analysis of the OSHA Inspection Program in Manufacturing Industries, , Mimeo. (Washington, DC: U.S. Department of Labor, 1976). 5. Robert S. Smith, The Impact of OSHA Inspections on Manufacturing Injury Rates, The Journal of Human Resources 14 (Spring 1979): W. Kip Viscusi, The Impact of Occupational Safety and Health Regulation, Bell Journal of Economics 10 (Spring 1979): Lawrence S. Bacow, Bargaining for Job Safety and Health (Cambridge, Mass.: The MIT Press, 1980). 8. Samuel Peltzman, The Effects of Automobile Safety Regulation, Journal of Political Economy 83 (August 1975): Leon S. Robertson, A Critical Analysis of Peltzman s The Effects of Automobile Safety Regulation, Journal of Economic Issues 11 (September 1977): Leon S. Robertson, Automobile Safety Regulations and Death Reductions in the United States, American Journal of Public Health 71 (August 1981): Laura I. Langheim and Allan J. Lichtman, Ecological Inference (Beverly Hills, Cal.: Sage Publications, 1978). 12. James R. Chelius, Workplace Safety and Health: The Role of Workers Compensation (Washington, D.C.: American Enterprise Institute, 1977). 13. Louise B. Russell, Safety Incentives in Workmen s Compensation Insurance, The Journal of Human Resources, Spring 1974, p Leon S. Robertson, Injuries: Causes, Control Strategies and Public Policy (Lexington, Mass.: Lexington Books, 1983). 15. Economic Indicators, June, 1982 (Washington, D.C.: U.S. Government Printing Ofice, 1982). 16. Zdenek Hrubec and Blaine S. Nashold Jr., Epidemiology of Lumbar Disk Lesions in the Military in World War 11, American Journal of Epidemiology 102 (November 1975): Bureau of the Census, Annual Survey of Manufacturers: Statistics for States, Standard Metropolitan Statistical Areas, Large Industrial Counties and Selected Cities (Washington, D.C.: U.S. Department of Commerce, 1977). 18. Bureau of Labor Statistics, State Data on Occupational Injuries and Illnesses in 1976 (Washington, D.C.: U.S. Department of Labor, 1979). 19. Data provided by the Office of Management Data Systems, Occupational Safety and Health Administration, U.S. Department of Labor, Washington, D.C. 20. Analysis of Workmen s Compensation Laws (Washington, D.C.: Chamber of Commerce of the U.S., 1975, 1976). 21. Bureau of the Census, 1977 Census of Manufacturers, Volume IiI, Geographic Area (Washington, D.C.: U.S. Department of Commerce, 1981). 22. Bureau of Labor Statistics, Employment and Earnings, States and Areas, (Washington, D.C.: U.S. Department of Labor, 1979). 23. Smith, The Impact of OSHA Inspections.