The value of job safety for railroad workers

Journal of Risk and Uncertainty, 5:175-185 (1992) © 1992 Kluwer Academic Publishers

The Value of Job Safety for Railroad Workers

MICHAEL T. FRENCH Center for Economics Research, Research Triangle Institute, Research Triangle Park, NC 27709 Department of Economics and Business, North Carolina State University, Raleigh, NC and School of Public Health, University of North Carolina, Chapel Hill, NC

DAVID L. KENDALL* Center for Economics Research, Research Triangle Institute, Research Triangle Park, NC 27709

Key words: value of safety, workplace injuries, railroad industry, willingness to pay

Abstract

The value of reducing job risk is estimated using a hedonic wage model and a risk variable that is matched by occupation and state of residence. This study is the first to use the hedonic wage approach to estimate the value of safety in a single industry. Industry-specific estimates will help researchers and labor policymakers better understand the distribution of compensating wage differentials across industries. Our estimated value of job safety for railroad workers is between $21,000 and $26,000 (1980 dollars) per statistical disabling injury, which is somewhat larger than the average of values estimated in previous studies that use cross-industry or cross- occupation aggregate risk data.

Several empirical studies show that workers receive wage compensation for exposure to fatal and nonfatal job-safety risks. 1 But estimated compensating wage differentials for comparable nonfatal injury risks range from $12,000 to $34,000 in 1980 dollars. 2 More- over, Viscusi and Moore (1987) report that the mean value of increased safety would jump to $89,000 if workers' compensation were not present. As Rosen (1988) notes, better empirical estimates of risk valuation are needed to reconcile differences across existing estimates.

One explanation of variation in empirical estimates across studies is that average job risk and risk preferences for particular samples of workers differ from study to study. Dillingham (1985) suggests that higher estimates from industry risk data and lower estimates from occupational data are due to different risk definitions, not differences in risk faced by a given sample of workers. Both explanations are reasonable and may explain part of the variation in empirical estimates.

An alternative but unexplored explanation is that cross-industry and cross-occupation aggregate risk data may lead to different estimates from study to study. Workers' exposure to injury risk on the job varies widely across firms within a particular industry and by

*Earlier versions of this article were presented at the Southern Economic Association Annual Conference, Orlando, Florida, November, 1989, and the Western Economic Association Annual Conference, Lake Tahoe, Nevada, June 1989. We thank Kip Viscusi, Frank Gollop, and an anonymous referee for helpful suggestions and Judy King for editing the manuscript.

176 MICHAEL T. FRENCH/DAVID L. KENDALL

occupation within specific firms. 3 Consequently, researchers would like to use worker- specific injury-rate data to estimate econometric wage-risk models, but such perfectly matched risk data are not generally available.

Most empirical studies resort to average injury rates across industries or averages across occupations in lieu of injury rates for workers within a specific industry, firm, and occupation. Using cross-industry average risk data, analysts assign the same risk measure to workers in the same Standard Industrial Classification (SIC) industry. With cross- occupational average risk data, analysts assign the same risk measure to workers in the same occupational categories.

Either risk measurement strategy may cause estimation errors. Because they are aggregated across firms and occupations, cross-industry average risk data do not reflect variation in injury risk between occupations or between firms within the same industry. For example, two lab technicians employed by different firms in the chemical industry would be assigned the same risk measure even though their firms have different safety records. Even worse, a materials handler in the chemical industry would be assigned the same risk measure as either lab technician.

Cross-occupation average-risk data pose a similar aggregation problem. Because they are aggregated across industries and firms, cross-occupation average-risk data do not reflect variation in injury risk between firms or between industries for the same occupation. For example, two welders--one in the building construction industry and the other in the railroad industry--might be assigned the same risk measure even though their industries have different safety records.

Despite data-aggregation problems, several studies use hedonic wage models to estimate the value of job safety. Hedonic wage models specify a relationship between wage compensation and workers' exposure to job risk. Using multivariate regression, researchers estimate the partial effect of injury rates on wages while controlling for other variables such as age, education, sex, race, marital status, residence, medical insurance, workers' compensation, and occupation--all variables that may affect wages in theory.

Empirical estimates of the value of job safety reported in earlier studies must generally be interpreted as averages across industries or occupations. This study adds to existing information about the value of job safety by estimating a compensating wage differential for a single industry. Industry-specific estimates should help researchers and labor policymakers better understand the distribution of compensating wage differentials across industries. The wide range of empirical estimates for average compensating wage differentials may in part be explained by variation between industries. That is, empirical estimates of the average value of job safety may vary from study to study because the sample for each study comprises workers from a different set of industries. A series of industry-specific estimates, such as those reported here, would add value by revealing how much compensating wage differentials vary from industry to industry.

This study also tries to improve value-of-safety estimates by using occupation- and state-specific injury data to derive on-the-job injury risk for each sample worker. Using less aggregated injury data to estimate on-the-job injury risk should improve empirical estimates. Job injury risk is calculated with injury-rate data for 36 occupational categories in the railroad industry cross-tabulated with injury-rate data by state of residence.

THE VALUE OF JOB SAFETY FOR RAILROAD WORKERS 177

Results indicate that the value of reducing risk of a disabling injury for railroad workers is between $21,000 and $26,000 per statistical disabling injury.

I. Sample and data

We use 1980 panel data from the Census Bureau's Current Population Survey (CPS) (U.S. Bureau of the Census, 1981), augmented with disabling-injury data from the Fed- eral Railroad Administration (FRA) (U.S. Department of Transportation, 1981) to estimate a hedonic wage model. The sample includes only individuals employed full- time in the railroad industry. Records with missing data or for individuals who had more than one employer in 1980 were discarded, leaving a sample of 374 railroad workers.

An ideal data matrix for estimating hedonic wage models would include the following variables matched to each worker in the sample: 1) hourly wage; 2) personal and demographic characteristics such as age, education, sex, race, state of residence, and marital status; 3) job characteristics such as medical and pension insurance coverage and occupational dummy variables; and 4) a measure of worker-specific job risk that is closely matched to each worker in the sample.

The CPS sample includes workers living in the 48 continental states who report wages, demographic characteristics, personal characteristics, and occupation by Census Bureau categories. Creating a job-risk variable matched to each worker in the sample by occupation and state of residence requires additional data and the following intermediate steps: 1) calculate a set of injury rates for occupational categories that can be matched to workers in the CPS sample; 2) calculate a set of railroad injury rates by state that can be matched to sample workers; 3) use the RAS algorithm demonstrated by Bacharach (1965) and Stone and Brown (1962) to estimate a state-by-occupation matrix of injury rates; and 4) assign each sample worker an injury rate based on the worker's state of residence and occupational category.

The FRA reports accident data by Interstate Commerce Commission (ICC) occupational categories. We use these data to calculate injury rates by occupation. Although ICC titles of occupational categories differ slightly from the Census Bureau occupational categories, matching of ICC and Census occupational categories is reasonably straight- forward and obvious.

Appendix A lists 36 railroad occupational groups with man-hours worked (~jMjc), annual disabling injuries (Ic), and disabling injury rates (IRe), where Mjc is millions of man-hours worked within occupation c and firm j; YjMjc is millions of man-hours worked in occupation c by all railroad firms; Ic is annual injuries within railroad occupation c; and IRc = Ic/~jMjc is the estimated disabling-injury rate for railroad occupation c. 4

Workers in the CPS sample do not report firm of employment. Without firm of employment, we cannot match exactly firm-specific injury rates to workers in the sample. Consequently, we use a proxy variable for firm-specific injury rates. The proxy is a state- specific injury rate, calculated as a weighted average of firm-specific injury rates for all railroads operating in the state. So firm-specific injury-rate data are used to calculate injury risk, but assigning injury risk to sample workers is based on state of residence.


Although this procedure is not perfect, it should produce injury-rate estimates that more closely match individual workers than the usual practice of assigning all workers in the same industry or occupation the same risk measure.

Equation (1) shows how we calculate annual injuries in the railroad industry by state.

Is = j[(MjsmsMjs)Ij], (1)

where Is is annual injuries in state s (Mjs is annual man-hours operated by railroad j in state s) and Ij is annual injuries for railroadj. Annual railroad injuries in state s (Is) is the weighted average of injuries across all railroad firms operating in the state. The weights are man-hours operated by a particular railroad divided by total man-hours operated by that railroad in all states. Unpublished data from FRA give annual man-hours worked in each state for all railroads.

Dividing Is by Y~jMjs (millions of man-hours operated in state s by all railroads) gives IRs, an estimate of the railroad injury rate for state s. If each sample worker is employed by one of the railroads operating in the worker's state of residence, IRs should be a reasonable estimate of injury risk incurred by railroad workers who live in state s. Al- though workers may be employed by a railroad that does not operate in the worker's state of residence, this is a low-probability event for most railroad occupations. Appendix B reports annual man-hours (EjMjs), disabling injuries (Is), and disabling-injury rates (IRs) by state.

Vermont had the lowest annual state injury rate (14.55) and New York had the highest (48.76). The source of the variation in state injury rates can be traced to the concentra- tion of different railroads that operated in each state. For example, five different railroads operated in Vermont in 1980, while 11 railroads had operations in New York. Only three railroads operated in both states in that year. The data and calculations for state injury rates are available from the authors.

The next step is estimating occupation-by-state injury rates and then matching injury rates to workers in the sample. We derive a 48 x 36 (state by occupation) injury-rate matrix (R) by dividing each element in a 48 x 36 injury matrix (I) by the corresponding element in a 48 × 36 man-hour matrix (M). Each element (Msc) in the man-hour matrix is millions of man-hours worked in state s within railroad occupation c. Likewise, each element (Isc) in the injury matrix is the number of injuries occurring in state s with individuals working in railroad occupation c.

Man-hour and injury data from the FRA provide Ms. = EcMsc, man-hours worked in state s summed across all railroad occupations; M.c = EsMsc, man-hours worked in railroad occupation c summed across all states; Is = Eclsc, injuries in state s summed across all railroad occupations; and Ic = Eslsc, injuries in railroad occupation c summed across all states. But unfortunately, values of cell entries in either the man-hour matrix or the injury matrix are not available.

Using the RAS algorithm demonstrated by Bacharach (1965), we estimate cell entries for M and I from knowledge of marginal row and column sums and initial starting values for elements Isc and M~. The algorithm iteratively adjusts starting cell entries--fist an


iteration by row, then an iteration by column. Adjustments in each iteration are propor- tional to the difference between calculated marginal sums and sums reported in FRA data. Iteration continues until the adjustment differential is arbitrarily small,

The mean disabling injury rate for railroad workers in our sample was 30.982 injuries per million worker hours. Assuming these individuals worked 2000 hours per year, on average, the annual injury frequency rate was 0.062 per worker. In comparison, the annual injury frequency rate for all railroad employees and for all private sector employees was 0.072 and 0.040 per worker in 1980. Coincidentally, the average lost-workday injury resulted in 16 days away from work for both the railroad industry and the aggregate of all private sector industries (U.S. Department of Labor, 1990).

In theory, compensating wage differentials for perceived job risk exposure should be smaller if workers' compensation is available. 6 Workers' compensation benefits are ex post compensation for job risk, while compensating wage differentials are ex ante compensation. The relative importance of the two forms of compensation depends on the degree to which workers wish to insure against income loss from job injury.

Unlike most U.S. workers, railroad workers are not covered by state workmans' compensation laws. When workplace injury occurs, railroad workers have two options for recovering lost income. First, they may file for partial indemnification under the Rail- road Unemployment Insurance Act (RUIA). This act covers primarily minor injuries, because wage loss is indemnified for only a short period. Second, for more serious job injuries, workers may file lawsuits against their employers under the Federal Employers' Liability Act (FELA) (1988). Injured railroad workers can file a claim against a railroad in their state of residence or in any state where the railroad does business. If workers prove employer negligence, the amount of damages they can recover under FELA is unlimited. However, less than 25% of all cases filed by railroad workers under FELA go to trial.

This unique form of workers' compensation prevents us from estimating the effect of workers' compensation on wages. Because all railroad workers are covered by RUIA and FELA, we cannot include a workers' compensation variable on the right-hand side of the hedonic wage model. But as Viscusi and Moore (1987) note, one can omit workers' compensation from wage equations if benefit levels are uniform. In fact, because there is no systematic variation in workers' compensation across railroad workers, a hedonic wage model should attribute none of the variation in wages to compensation for workplace injuries.

2. Estimation and results

We estimate several hedonic wage equations in both linear and log-linear form. Each model specifies a particular combination of injury rate (IR) and injury-rate interaction variables (e.g., IR × AGE, IR × GRADE). Table 1 defines variables in the estimation models and reports sample means and standard deviations. Table 2 reports a complete


Table 1. Variable means and standard deviations

Variable Mean Standard deviation

Weekly wage 444.614 128.933 Male 0.928 0.259 Caucasian 0.939 0.241 Age 40.933 12.588 Included in company pension plan 0.882 0.323 Included in company health plan 0.968 0.176 Head of household 0.890 0.313 Residence in northeastern U.S. 0.166 0.372 Residence in southern U.S. 0.316 0.465 Residence in northcentral U.S. 0.249 0.433 Married 0.765 0.425 Highest grade attended 12.869 2.501 Occupation

• Professional and technical 0.032 0.176 • Managers and administrators 0.147 0.355 • Sales and clerical 0.142 0.349 • Craftsmen 0.361 0.481 • Operators 0.211 0.409 • Laborers 0.024 0.153

Injuries per million worker-hours 30.982 23.898

set of parameter estimates and regression statistics from the linear and log-linear models with no interactions. 5 The statistical results appear to be robust, and the risk variable is always right-signed and statistically different from zero.

The March 1980 CPS does not report workers' union membership status. But union- ization of blue-collar workers in the railroad industry is high. Consequently, including an indicator variable for union membership would not be expected to improve the estimation model because of insufficient variation in union status among blue-collar sample workers, and because occupational indicator variables already identify blue-collar and white-collar workers.

The estimated annual compensating wage differential for a representative railroad worker is equal to the injury risk coefficient (0.0023 from the log-linear hedonic wage equation) multiplied by the average weekly wage ($445), multiplied by the average number of weeks worked per year (50), or about $51.00. To compute the mean value of avoiding a statistical disabling injury, multiply the compensating differential estimate by the average number of workers affected (500). The resulting value of avoiding a statistical disabling injury is about $26,000. Using the injury risk coefficient from the linear hedonic wage equation, the value-of-safety estimate is $21,000, or about 20% lower.

French (1990) ignored state-injury-rate differentials and estimated linear and log- linear hedonic wage equations with nationwide railroad occupational risk measures. The estimated compensating wage differential using these more aggregate risk data ranged from $19,500 to $22,500 per disabling injury. This result lends support to our earlier


Table 2. Estimation results for linear and log-linear equations

Independent variable

Parameter estimates (N = 374) a

Linear equation Log-linear equation

Constant 90.2974 (62.6791) 5.1491 c (0.1365) Married 5,7603 (14.7862) 0.0274 (0.0322) Head of household 36.6359 b (20.4549) 0.0733 b (0.0446) Male 19.9249 (26.3337) 0.0660 (0.0574) Caucasian 22.5447 (24.6696) 0.0275 (0.0537) Included in company pension plan 19.5544 (18.3966) 0.0800 c (0.0401) Included in company health plan - 19.4524 (34.8100) 0.0268 (0.0758) Residence in northeastern U.S. - 7.4933 (18.0775) - 0.0471 (0.0394) Residence in northcentral U.S. 3.2059 (15.3237) 0.0084 (0.0334) Residence in southern U.S, - 15.3530 (16.1186) - 0.0346 (0.0351) Age 0.7829 (0.5102) 0.0011 (0.0011) Highest grade attended 7.8964 c (2.6654) 0.0156 c (0,0058) Professional and technical 184.1271 c (47.5241) 0.4900 c (0.1035) Managers and administrators 246.8101 c (31.3160) 0.5861 c (0.0682) Sales and clerical 104.8533 c (36.5541) 0.3358 c (0.0796) Craftsmen 123.4973 c (27.8544) 0.3443 c (0.0607) Operators 101.3527 c (24.7541) 0.2683 c (0.0539) Laborers 26.0696 (42.2001) 0.0681 (0.0919) Injuries per million worker-hours 0,8370 c (0.3886) 0.0023 c (0.0008) ~2 0,2821 0.2962 Mean value of avoiding a statistical disabling injury $20,926 $25,964

aStandard errors in parentheses. bStatistically significant at the 0.05 confidence level (one-tailed test). cStatistically significant at the 0.01 confidence level (one-tailed test).

hypothesis that using aggregate risk measures in hedonic wage equations will result in lower value-of-safety estimates.

3. Conclusions

The estimated value of safety in the railroad industry is between $21,000 and $26,000 per statistical disabling injury, given that workers' compensation through RUIA and FELA is available. The midpoint, $23,500, is larger than most previous value-of-safety estimates but still well within the range of the distribution ($12,000 to $34,000) from other studies. Because no data are available that show systematic variation in the rate of wage replace- ment for railroad workers injured on the job, we are unable to estimate how the value of safety would change if ex post compensation were not available. In theory, compensating wage differentials should be larger if RUIA and FELA were not available to all railroad workers.


Estimates of the value of reduced risk of job injury that we report in this study may be more precise than many previous estimate for two reasons. First, previous studies use industrywide injury rates averaged across many occupations and firms--or occupation- wide rates averaged across many industries and firms--to measure the risk variable. This is the first hedonic wage study to use injury risk measures for a single industry. Using industry-specific risk data should produce more precise estimates for that industry than risk data averaged across industries or occupations, ceteris paribus. Second, this study matches injury risk to sample workers by state of residence and railroad occupation, where state of residence is designed to be a proxy for firm of employment, Earlier studies were confined by data limitations to matching injury risk to sample workers by industry only or by occupation only. Less aggregate measures of injury risk matched to sample workers should produce more precise estimates of compensating wage differentials.

Comparing value-of-safety estimates for the railroad industry with other estimates that are averages across industries or across occupations is difficult. Estimates in this study are higher than most estimates from other studies. Several factors may explain this difference. Estimates from other studies that use aggregate risk data could be lower because of lower variation in average risk than individual risk. Job injury risk perceived by workers in the railroad industry could be generally higher than in most industries. Or workers in the railroad industry could be more risk averse than workers in other industries (although this hypothesis contradicts jobs search theory).

Without additional industry-specific studies and better matching of risk data to individual workers, answers will remain unclear. This study is a step toward producing industry-specific information about the value of job safety. Additional studies for other industries would be useful to expand our understanding of the value of job safety.

Appendix A: 1980 man-hours, disabling injuries, and injury rates by occupation

Occupation Annual million man-hours Annual disabling injuries Disabling injury rate

Executives/officers 12.279 9 0.73 Assistant officers 24.972 26 1.04 Professionals 21.023 33 1.57 Supervisors 18.901 33 1.75 Messengers 0.911 18 19.76 Clerical specialists 13.983 51 3.65 Clerks 78.498 527 6.71 Office machine operators 7.307 13 1.78 Secretaries 13.174 18 1.37 Buyers/agents 2.943 16 5.44 Telephone/station agents 29.532 150 5.08 Freight agents/investigators 2.101 13 6.19 Truckers 11.031 275 24.93 Policemen/guards 6.817 289 42.39 Conductors 37.967 1608 42.35

(continued)


Occupation Annual million man-hours Annual disabling injuries Disabling injury rate

Boilermakers 3.061 143 46.72

Machine operators 26.875 605 22.51 Carpenters 8.079 372 46.05

Electrical workers 21.353 630 29.50 Apprentices 6.737 209 31.02

Linemen/groundmen 4.352 152 34.93 Foremen 56.820 740 13.02

Inspectors 3.036 27 8.89 Engineers 60.807 1212 19.93

Firemen 15.498 273 17.62 Machinist 33.211 1128 33.96

Carmen 72.912 2473 33.92 Plumbers/masons 6.323 211 33.37 Sheet metal workers 9.417 283 30.05 Station engineers 0.581 3 5.16 Station firemen 0.335 8 23.88 Brakemen/flagmen 114.349 7590 66.38

Laborers 86.288 5638 65.34 Coach cleaners 39.45 227 57.54

Building attendants 3.330 83 24.92 Train attendants 2.084 236 113.24

Appendix B: 1980 man-hours, disabling injuries, and injury rates by state

State Annual million man-hours Annual disabling injuries Disabling injury rate

Alabama 14.877 386 25.97 Arizona 8.237 180 21.80 Arkansas 12.894 394 30.52 California 53.685 1514 28.21 Colorado 14.079 308 21.88 Connecticut 5.858 208 35.46 Delaware 3.030 119 39.27 Dist. of Columbia 2.943 122 41.45

Florida 17.089 476 27.87 Georgia 27.539 717 26.02 Idaho 5.687 200 35.08 Illinois 72.718 2222 30.56

Indiana 29.344 903 30.78 Iowa 16.452 466 28.31 Kansas 24.193 606 25.05 Kentucky 22.366 605 27.05 Louisiana 13.902 445 32.03 Maine 4.332 206 47.46 Maryland 15.272 362 23.70

(continued)


State Annual million man-hours Annual disabling injuries Disabling injury rate

Massachusetts 9.572 287 30.16

Michigan 26.871 812 30.20 Minnesota 28.705 909 31.67 Mississippi 8.756 188 21.44 Missouri 29.263 844 28.85

Montana 12.442 324 26.01 Nebraska 36.641 1143 31.18 Nevada 1.836 65 35.14

New Hampshire 0.707 19 26.72 New Jersey 15.001 502 33.48 New Mexico 5.677 108 19.06 New York 47.986 2340 48.76

North Carolina 12.664 299 23.58 North Dakota 6.900 209 30.27

Ohio 51.672 1470 28.45 Oklahoma 10.807 281 25.96 Oregon 14.654 465 31.70 Pennsylvania 71.382 2254 31.58

Rhode Island 0.883 36 40.61 South Carolina 8.506 205 24.16 South Dakota 1.858 68 36.53 Tennessee 20.167 499 24.72 Texas 57.920 1610 27.79 Utah 7.848 236 30.12

Vermont 0.853 12 14.55 Virginia 27.612 767 27.80

Washington 15.689 433 27.59 West Virginia 15.883 395 24.86 Wisconsin 20.441 860 42.08 Wyoming 10.243 330 32.19

Notes

1. See, for example, Thaler and Rosen (1976), R. S. Smith (1976), Viscusi (1978, 1983), Brown (1980), Olson (1981), Arnould and Nichols (1983), and Viscusi and Moore (1987).

2. See Viscusi (1978a, 1978b, 1979, 1981), V. K. Smith (1983), Arnould and Nichols (1983), Viscusi and O'Conner (1983), Viscusi and Moore (1987), and Viscusi and Evans (1990).

3. In 1980, injuries per million man-hours ranged from a low of 2.41 at Florida East Coast Railroad to a high of 89.91 at Long Island Railroad. Appendix A shows a wide range of injury rates across railroad occupations.

4. A disabling injury is a job injury resulting in at least one lost workday. The disabling injury rate is the number of nonfatal job injuries resulting in at least one lost workday per million man-hours. Job fatality rates are an alternative measure of risk. But cross-sectional fatality rates for a single year may be a poor measure of job risk because many finns and occupations experience no on-the-job fatalities in particular years.

5. We do not report parameter estimates from the interaction models because the interaction variables do not always have a statistically significant effect on wage; moreover, estimated compensating differentials fall within the range of those from model specifications with no interaction terms.


6. Arnould and Nichols (1983) and Viscusi and Moore (1987) estimated hedonic wage models with a workers' compensation variable. Each study found that wage compensation is higher if workers' compensation is unavailable,

References

Arnould, R. J. and L. M. Nichols. (1983). "Wage-Risk Premiums and Workers' Compensation: A Refinement of Estimates of Compensating Wage Differential," Journal of Political Economy 91(2), 332-340.

Bacharach, M. (1965). "Estimating Matrices from Marginal Data," International Economic Review 6(3). Brown, C. (1980). "Equalizing Differences in the Labor Market," Quarter~ Journal of Economics 94(1), 113-134. Dillingham, A. E. (1985). "The Influence of Risk Variable Definition on Value of Life Estimates," Economic

Inquiry 24, 227-294. Federal Employers' Liability Act, 45 United States Code, §§51-60 (1982). French, M. T. (1990). "Estimating the Full Cost of Workplace Injuries," American Journal of Public Health

80(9), 1118-1119. Interstate Commerce Commission. (1981). Transport Statistics in the United States Railroads: 1980, Part 1.

Washington, DC: U.S. Government Printing Office. Olson, C. A. (1981). "An Analysis of Wage Differentials Received by Workers on Dangerous Jobs," Journal of

Human Resources 16(2), 167-185. Rosen, S, (1988). "The Value of Changes in Life Expectancy," Journal of Risk and Uncertainty 1(3), 285-304. Smith, R, S. (1976). The Occupational Safety and Health Act. Washington, DC: American Enterprise Institute

for Public Policy Research. Smith, V. K. (1983). "The Role of Site and Job Characteristics in Hedonic Wage Models," Journal of Urban

Economics 13, 296-321. Stone, R. and J. Brown. (1962). A Computable Model of Economic Growth, A Programme for Growth, Mono-

graph Series. London: Chapman and Hall. Thaler, R. and S. Rosen. (1976). "The Value of Saving a Life: Evidence from the Labor Market." In N.

Terleckyz (ed.), Household Production and Consumption, NBER Studies in Income and Wealth, No. 40. New York: Columbia University Press.

U.S. Bureau of the Census. (1981). 1980 Census of Population Survey. Washington: U.S. Government Printing Office, March.

U.S. Department of Labor, Bureau of Labor Statistics. (1990). Occupational Injuries and Illnesses in the United States by Industry. Washington, DC: Government Printing Office.

U.S. Department of Transportation, Federal Railroad Administration, Office of Safety. (1981). 1980Accident~ Incident Bulletin (149), June. Washington, DC: .

Viscusi, W. K. (1978a). "Labor Market Valuations of Life and Limb: Empirical Estimates and Policy Implica- tions," Public Policy (26), 359-386.

Viscusi, W. K. (1978b). "Wealth Effects and Earnings Premiums for Job Hazards," Review of Economics and Statistics 60(3), 408-416.

Viscusi, W. K. (1979). Employment Hazards: An Investigation of Labor Market Performance. Cambridge: Harvard University Press.

Viscusi, W. K. (1981). "Occupational Safety and Health Regulation: Its Impact and Policy Alternatives." In J. Crecine (ed.), Research in Public Policy Ana~sis and Management, vol. 2. Greenwich, CT: JAI Press, pp. 281-289.

Viscusi, W. K. (1983). Risk by Choice: Regulating Health and Safety in the Workplace. Cambridge: Harvard University Press.

Viscusi, W. K. and O'Conner. (1984). "Adaptive Responses to Chemical Workers: Are Workers Bayesian Decision Makers," American Economic Review 74(5), 942-956.

Viscusi, W. K. and M. J. Moore. (1987). "Workers Compensation: Wage Effects, Benefit Inadequacies, and the Value of Health Loss," The Review of Economics and Statistics 249-261.

Viscusi, W. K. and W. Evans. "Utility Functions That Depend on Health Status: Estimates and Economic Implications,"American Economic Review 80(3), 353-374.

Documents

The value of job safety for railroad workers