Hedonic damages revisited: some empirical findings.

• Author: Harmon, Oskar Ragnar

THE area of hedonic damages has received a great deal of attention in recent years in both personal injury and wrongful death cases. There are several conceptual assumptions that form the foundation for an economist's analysis of hedonic damages that can be challenged effectively by defense counsel.

These issues were discussed in an earlier article in this journal.(1) This article serves as a follow-up to the earlier study and provides some empirical support to the claim that the correlation between wages and the risk of death on the job--the heart of the value of life formulation--is highly sensitive to model specification at best and spurious at worst.

A recent survey of the value of life in the literature finds that the best estimates are in the wide range from $1.6 million to $8.5 million.(2) It is our hypothesis that even a slight change in specification is capable of producing these dramatic changes in the empirical results. In our study, we re-estimate one of the more recent value of life studies found in the literature and test the sensitivity of that study's results to a slight change in the statistical model.

In the next section, some estimation problems in the "value of human life" determination are discussed. A description of the data we analyze follows along with a discussion of our findings. Some further suggestions for cross-examination are provided in the concluding section

ESTIMATION PROBLEMS

It is often the case that the "value of life" estimates are determined from a single regression coefficient. The human capital model may take the following form:

W = [B.sub.0] + [B.sub.1]F + [B.sub.2]X + e

Where W is the hourly wage rate, net of

taxes;

F is the fatality rate the worker

faces on the job;

X represents all of the other

observable determinants of

wage rates; and

e is the random error term.

The coefficient [B.sub.1] determines the amount by which hourly wages increase for an increase of one death per, say, 100,000 workers. Multiplying this coefficient by 2,000 hours to convert it to its annual equivalent and then again by 100,000 produces the value that 100,000 workers are placing on one life. For example, Moore and Viscusi(3) report a value for the coefficient [B.sub.1] of .022, which implies a value of life estimate of $4.4 million ($4.4 million = .022 x 2,000 x 100,000).

It is well known that a single regression co-efficient can be subject to many biases that can change both its magnitude and statistical significance. Therefore, of particular concern to the value of life estimation is the accuracy of the measure of job fatality risk. With few exception, all the risk-based measures of the value of life have been obtained using job fatality risk data from the Bureau of Labor Statistics.(4)

In 1990, because of measurement difficulties, some of which are discussed below, the BLS determined that these data did not accurately measure on-the-job fatality risk and temporarily suspended the survey until these measurement difficulties could be resolved.(5)

In 1988, a new job fatality risk variable obtained from the National Traumatic Occupational Fatality (NTOF) data set was made available to researchers. The NTOF data are considered superior to those of the BLS because the NTOF is a census of death certificates on which the cause of death is noted as work related. The BLS data are from a sample survey of establishment reported work-related fatalities.

The estimates of work-related fatalities of the NTOF are roughly double those of the BLS data. In 1990, the BLS estimated 2,900 such fatalities, while NTOF estimated 6,400 such fatalities in 1985, the latest year for which these data are available. In the only published study using this data, Moore and Viscusi report that this new data set produces a "value of life" estimate twice the amount they obtained using the BLS data.(6)

Since a great deal rides on the magnitude of the coefficient on the job fatality risk variable, it is important that this variable be an accurate reflection of the risk of work-related fatalities. On close examination of the NTOF data, however, it is clear that there are at least three reasons to suspect these data are contaminated by a great deal of measurement error.

First, the NTOF data, as the BLS data, do not reflect fatalities arising from work-related diseases. The NTOF data are restricted to the International...