These circumstances are similar to those facing displaced workers. Both injury and displacement involve a period off work related to an exogenous event, both
involve a loss of human capital and consequently in wages, and both present the possibility that employers will discriminate in decisions about reemployment.
2
A range of studies has found that displaced women with similar characteristics lose a
greater proportion of pre-injury earnings than do men Ruhm 1987; Podgursky and Swaim 1987; Jacobson, LaLonde, and Sullivan 1993; Crossley, Jones, and Kuhn
1994. The parallel between displacement and injury and the initial findings of gender differences in losses from workplace injuries lead us to pursue this issue.
We begin by estimating losses for men and women separately, using a difference- in-differences approach: Given the characteristics of our data, we calculate differ-
ences between post-injury earnings of a comparison group and of injured workers. Next, we attempt to determine which factors can explain the observed differences.
We examine the extent to which observed personal, job, employer, and injury charac- teristics account for gender disparities. To do this, we apply an extension of the
Oaxaca-Blinder decomposition Oaxaca 1973; Blinder 1973, as refined by Neumark 1988 and Oaxaca and Ransom 1994. We use this to evaluate the difference be-
tween expected male and female injury-induced changes in earnings and to calculate ‘‘nondiscriminatory’’ changes in earnings. We can measure not only gender differ-
ences in losses but also the extent to which men appear to gain from favoritism ‘‘nepotism’’, and women appear to lose from discrimination.
After accounting for gender disparities in observed covariates, we use additional information to see whether the differences that remain can be explained by hypothe-
ses other than discrimination. We first estimate the impact of workplace injuries on the probability of being employed having positive earnings in a given post-injury
quarter and examine the impact of nonemployment on earnings disparities. We then consider alternate factors that may contribute to women’s losses, using both our
primary data set and additional data from a survey of a stratified random sample of 1,461 workers with back injuries from the same population. Here, we look for evi-
dence that differential injury severity, reduced hours of work, withdrawal from the labor force, greater loss of job-specific human capital, loss of compensating wage
differentials, more reinjury, and longer recovery times contribute to the observed gender differentials.
II. Data
Our study uses matched administrative data from three sources: Workers’ Compensation records, unemployment compensation wage records, and
employment security employer data covering workers employed in Wisconsin ex- cluding the self-employed and Federal government employees. We began with ad-
ministrative data for workers with lost-time injuries in the State of Wisconsin be- tween April 1, 1989, and September 30, 1990. Using individual identifiers, we
2. Still, workplace injuries are clearly not the same as displacement, since some injured workers suffer permanent productivity losses because of their injury-related work limitations. Also, unlike displaced work-
ers, most injured workers return to the initial employer.
matched more than 97 percent of the injured workers to their unemployment insur- ance wage records. From these wage records, we extracted employer identifiers and
quarterly earnings from the beginning of 1988 through the end of 1993. We imputed zero quarterly earnings in quarters for which no employer reports earnings for a
person in our sample. We created a data set with 24 quarterly records for each worker, including quarterly earnings and employer as well as personal, injury, and employer
characteristics. Each worker has quarterly earnings data for five to ten quarters before and 13 to 18 quarters after the injury date. Using the employer identifier, we then
matched to quarterly unemployment insurance QUI employer data, which allowed us to determine the industry 1987 SIC and employment size of each quarter’s pri-
mary employer. The primary employer is the employer who paid the most wages in a quarter.
Our final data set consists of 24 quarterly observations for each worker. Each observation includes the calendar quarter of the observation, the relative time to the
quarter of injury, and the worker’s earnings in the quarter which we express in constant 1994 dollars. For each worker, we also have data on the coefficient of
variation of quarterly earnings during the pre-injury period, the pre-injury frequency of change of employer the number of actual changes divided by the number of
possible changes, the industry and employment size of the employer at injury 1987 SIC, the tenure with that employer at the date of injury, the worker’s occupation
at injury 1980 Census Occupation Code, the worker’s gender, the part of body injured, and type of workers’ compensation claim.
We estimate earnings for injured workers who received workers’ compensation permanent partial disability PPD benefits or temporary disability benefits lasting
at least eight days, for a total of 47,910 men and 22,467 women. As we discuss below, workers with only temporary disability benefits lasting eight to ten days are
chosen as a comparison group. We exclude from the estimates of earnings 110 work- ers who received permanent total disability payments or who suffered fatal injuries.
We further exclude 4,138 workers otherwise eligible for the comparison group but who had more than one injury in the observed period. This leaves 44,899 men and
21,340 women with temporary disability at least eight days or PPD benefits. Of these, 16 percent of men’s and 19 percent of women’s observations have missing
values—most frequently on the tenure or age of the injured workers. We also remove 158 outliers from our estimates of earnings: men with a residual greater than 30,000
in quarterly earnings and women with a residual greater than 20,000, leaving 36,283 men and 18,026 women.
The strengths of these data include the richness of the information about workers, the relatively long earnings series, and the large number of observations. The rich
information on covariates allows us to use regression adjustment to account for dif- ferences in the chosen comparison group and the injured group. The long earnings
series enables us to examine the pre-injury earnings stream of the comparison and injured-worker groups to test the quality of the adjusted match. In addition, it allows
us to see the extent to which losses shrink or remain relatively stable over time. The large number of observations is important because we can measure the impact of
injuries with sufficient precision to address our hypotheses.
On the other hand, the data have some limitations. They lack information on some determinants of earnings, notably educational attainment and race although race is
relatively unimportant to determining average effects in Wisconsin, an overwhelm- ingly White state. Also, the data include earnings but not separate information on
wages and hours. As a consequence, we cannot identify separately the impact of injuries on wages and hours. Finally, because we use Wisconsin earnings data, we
cannot observe out-of-state earnings. Collateral evidence suggests that this is a minor shortcoming. Among a surveyed subpopulation of the data we analyze, an earlier
analysis found that only 4.2 percent held an out-of-state job in the five to six years between their injuries and the interviews Galizzi, Boden, and Liu 1998. This per-
centage was equal for the comparison group and the injured groups studied.
Table 1 presents descriptive statistics for the data analyzed in this study, including
Table 1 Summary Statistics for Wisconsin 1989–1990 Injuries with One or More Weeks
of Lost Time
All Claims Men
Women
Individual characteristics Age in years
36.71 36.32
37.24 11.66
11.51 11.91
Tenure in years 5.87
6.29 5.01
7.72 8.25
6.41 Median
2 2
2 Occupation type proportion
Managerial or professional 0.05
0.03 0.09
0.21 0.16
0.28 Clerical
0.07 0.04
0.14 0.26
0.19 0.35
Service 0.15
0.07 0.29
0.35 0.26
0.45 Skilled blue collar
0.22 0.29
0.08 0.41
0.46 0.26
Unskilled blue collar 0.49
0.54 0.39
0.50 0.50
0.49 Agricultural, military, or other
0.02 0.02
0.01 0.13
0.15 0.09
Industry type proportion Agricultural, domestic service, or
0.06 0.07
0.04 other
0.23 0.25
0.20 Mining or construction
0.10 0.15
0.01 0.30
0.35 0.07
Durable manufacturing 0.28
0.31 0.23
0.45 0.46
0.42 Nondurable manufacturing
0.14 0.13
0.17 0.35
0.33 0.37
Transportation, communication, or 0.07
0.09 0.02
utilities 0.25
0.28 0.15
Wholesale trade or retail sales 0.17
0.16 0.17
0.37 0.37
0.38
Table 1 continued
All Claims Men
Women
Finance, insurance, real estate, or 0.18
0.10 0.36
other services 0.39
0.30 0.48
Employer characteristics Number of employees
1,169 989
1,531 2,324
2,105 2,675
Median 237
162 421
Proportion of employees in firms 0.24
0.30 0.12
with 50 or fewer employees 0.43
0.46 0.32
Proportion in public sector 0.10
0.08 0.12
0.29 0.28
0.33 Nature of injury
Objective 0.28
0.33 0.18
0.45 0.47
0.39 Subjective
0.60 0.54
0.70 0.49
0.50 0.46
Other 0.12
0.13 0.11
0.33 0.33
0.32 Part of body injured
Head, neck, or back 0.32
0.32 0.32
0.46 0.46
0.47 Back only
0.29 0.29
0.29 0.45
0.45 0.45
Upper extremities 0.28
0.25 0.33
0.45 0.43
0.47 Carpal tunnel syndrome
0.04 0.02
0.08 0.20
0.15 0.27
Trunk, multiple, or different injuries 0.23
0.23 0.23
0.42 0.42
0.42 Lower extremities
0.18 0.21
0.13 0.38
0.40 0.33
Claim characteristics Proportion with permanent partial
0.18 0.19
0.17 disability
0.39 0.39
0.37 Proportion with only temporary total
0.78 0.78
0.79 disability
0.41 0.42
0.41 Proportion of claims compromised
0.04 0.03
0.05 0.19
0.18 0.21
Earnings and employment Pretax earnings one quarter before
5,499 6,179
4,129 injury
3,420 3,547
2,666 Median
5,112 6,015
3,736 Frequency of pre-injury employer
0.09 0.09
0.08 change
0.15 0.16
0.15 Proportion changing employer after
0.17 0.18
0.16 injury
0.44 0.45
0.43 Total number of observations
54,309 36,283
18,026
Note: Standard deviations are in parentheses. Statistical analysis is based on these data.
Figure 1 Changes in Quarterly Earnings Relative to Comparison Group
median values where distributions are skewed. On average, the workers we study were 36 years old and had six years of tenure with the pre-injury employer. Almost
half were unskilled blue-collar workers. Only 24 percent were employed in firms with fewer than 50 employees and 10 percent were employed in the public sector.
The most frequent injuries were back injuries 29 percent, incurred equally often by men and women. Men were slightly younger, had longer tenure and higher earn-
ings, were more often blue-collar, and were more likely to be in mining, construction or durable manufacturing industries and in smaller firms. Women were more likely
to have service, clerical, and managerial or professional jobs, be in the service sector or nondurable manufacturing and have upper-extremity injuries and, in specific, car-
pal tunnel syndrome.
Figure 1 presents our first look at average differences between quarterly changes in earnings of injured workers
3
and in the comparison group. It represents difference- in-differences calculations of earnings, based on our raw data and therefore unad-
justed for covariates.
4
We impute zero quarterly earnings in quarters for which no employer reports earnings for a person in our sample. The horizontal line at zero is
the baseline, that is, the changes in earnings of the comparison group relative to the pre-injury quarter. In fact, we assume, for the moment, that the eight-to-ten day
group experienced no losses.
3. For simplicity of exposition, we will refer to workers not in the comparison group as ‘‘injured workers,’’ although those in the comparison group had short-term injuries.
4. To create Figure 1, we generate changes in earnings as: [Y
It
⫺ Y
10
⫺ Y
Ct
⫺ Y
C
]Y
10
], where y is wages, I is the injured group, C is the comparison group, t is the time period, and 0 is the quarter before
injury.
Figure 1 shows that, in the first two quarters after injury, men experience an aver- age percentage loss in earnings relative to the comparison group that is about the
same as for women. Yet, the unadjusted means suggest that over the next 3 12 years, men lost an average of 8.3 percent of pre-injury earnings while women lost
10.1 percent—about 14 higher than men. Figure 1 motivated this study and also helped us to specify the statistical model.
III. Methods