. Journal of Risk and Uncertainty, 18:231 248 1999 [615620]

. Journal of Risk and Uncertainty, 18:231 ]248 1999
Q1999 Kluwer Academic Publishers. Manufactured in The Netherlands.
The Impact of Institutional Change on
Compensating Wage Differentials for Accident
Risk: South Korea, 1984 ]1990
SEUNG-WOOK KIM
Chung-Ang Uni ¤ersity, Seoul, Korea
PRICE V. FISHBACK
Uni¤ersity of Arizona, Tucson, Arizona [anonimizat]
Abstract
Institutional change can lead to substantial changes in the size of compensating differentials for
workplace accident risk. The South Korean labor market experienced two major institutional changes
between 1984 and 1990. First, a relaxation of restrictions on Korean labor unions in 1987 led to a sharp
jump in the extent of strike activity and bargaining rounds which was associated with a reduction in the
size of compensating differences. Second, reform of Korean workers' compensation in 1989 led to a
substantial rise in benefits that also served to reduce the extent of compensating differentials.
Key words: compensating differentials, institutional change, unions, workplace accident risk, workers'
compensation
JEL Classification: 824, 830
Changes in labor market institutions can lead to substantial differences in the size
of compensating differentials for workplace accident risk. Thomas Kniesner and
. John Leeth 1991 found in comparisons of the U.S., Australia, and Japan that
compensating differentials were largest in the unionized Australian setting and
smallest in Japan where labor mobility was restricted most by the institutional
setting. Hedonic-wage studies on United States labor markets suggest that compen-
sating wage differentials are typically higher in union than in nonunion settings.1
. On the other hand, Seung-Wook Kim and Price Fishback 1993 found in the
historical analysis of the U.S. railroad industry that as more railroad jobs were
unionized and government arbitration played a more extensive role in the wage-
setting process, the risk premia for accidents declined. Several studies have shown
that reforms in workers' compensation programs also lead to substantial changes in
 compensating differences see Moore and Viscusi, 1990, p. 24; Gruber and Krueger,
. 1991; Fishback and Kantor, 1995 .

KIM AND FISHBACK 232
The South Korean labor market in the 1980s offers an excellent opportunity to
examine the impact of institutional change on compensating differentials becausethe South Korean government instituted two major labor market reforms. First, thegovernment enhanced the role of labor unions in 1987 by relaxing prior restrictionson inter-firm and inter-industry ties that had forced unions to bargain firm by firmin relative isolation. The relaxation of restrictions led to a sharp jump in the extentof strike activity and greater interaction between the striking unions, factors whichpotentially might have increased or decreased the size of compensating differen-tials. Second, the government reformed workers' compensation and raised benefitsby 30 percent for fatalities and 10 percent for nonfatal accidents. Earlier researchon the impact of such increases in post-accident compensation suggests that the
 risk premiums in wages will fall see summaries in Moore and Viscusi, 1990, p. 24;
. Fishback, 1998 . Using a panel of industry averages from 1984 through 1990 we
examine the impact of these institutional changes on compensating differences inthe South Korean labor market.
1. The South Korean labor market institutions and their predicted impact on
compensating wage differentials
Following a recession at the beginning of the decade, the South Korean economy
expanded rapidly during the 1980s. Real GDP grew at an average annual rate of9.4 percent. Workers shared in this growth, as real wages grew at rates rangingfrom 5.5 percent to 15.4 percent between 1982 and 1990. Despite the rapid growthin Korean wages, Korean workers still earned substantially less than workers inAustralia, Japan, and the United States. The average annual earnings for Koreanworkers in our sample was roughly $5,000 U.S. dollars in 1986, compared withaverage annual earnings of around $15,000 U.S. dollars in Australia, $21,600 inJapan, and $35,500 in the U.S.
2Korean workers also faced more dangerous
working conditions. Fatal accident rates in South Korea during the 1980s werehigher than in all but a handful of countries, typically at levels 3 to 10 times higherthan in the leading industrial nations.
3
The South Korean labor market in many ways is similar to Japan's and is quite
different from the labor market institutions in western countries. Labor customs inKorea are similar to Japanese customs because a large part of the modernizationof Korea was accomplished during the Japanese colonial period from 1910 to 1945.Many large Korean firms have Japanese-style internal labor markets that rewardworkers for long-term attachment. While many companies in the U.S. adopt ajob-based pay system, Japanese and Korean firms tend to implement seniority-basedpay systems which emphasize seniority elements such as the length of the worker's
. service and his age Ahn, 1994, 1996 . The basic pay, which is the main component
of the pay structure of Korean companies, follows the typical pattern of theseniority-based pay system. Accordingly, when a worker joins a company, his pay isdetermined by his academic level. After joining the company, his pay is augmented

SOUTH KOREA, 1984 ]1990 233
in proportion to increases in his age or years of service. Wage increases are
awarded at predefined rates; therefore, a worker's capability, performance, andcontribution are not explicitly taken into account in the determination of pay.Similarly, bonuses are given to workers on a uniform and fixed rate, regardless ofindividual performance. Merit ratings in Korean companies are exclusively used asa standard for promotion, and not as criteria for increases in wages or bonuses
.  .Ahn, 1994, pp. 55 ]56 . Hee-Tak Ahn 1996 argues that the wage system in Korea
was intended more to stabilize the monetary security of workers rather than as ameans of improving productivity or motivating workers to work harder.
4
The presence of rigid structures in internal labor markets that focus on age,
seniority, and academic origin, as in South Korea and Japan, might have led tolower compensating differentials than in markets characterized by more fluidity.After all worker mobility is the key factor underlying Adam Smith's original insightthat competitive labor markets lead to compensating differentials. The rigidemphasis on academic origin and seniority within firms seems to imply thatintrafirm compensating differentials would be diminished in South Korean labormarkets. If the firms are focused in one industry, we might still see substantialcompensating differences across industries. To the extent that the internal labormarkets span a variety of industries within the same firm, then inter-industrycompensating differentials potentially could be reduced as well. Such theoretical
. predictions are supported by Kniesner and Leeth 1991 findings that compensating
differentials were substantially lower in Japan than in the United States andAustralia.
Two masters projects on the Korean labor market prior to the substantial change
in the treatment of unions in 1987 show that the rigidities of Korean labor markets
. had not eliminated compensating differentials. Sung-Joong Kim 1985 found
positive compensating wage differentials for fatal injury risk and negative differen-tials for non-fatal risk for the manufacturing sectors for the year 1982. Hisestimates for fatal injury risk suggested that the value of a life implicit in the
.
5wage-risk premia was between $550,000 and $1.1 million dollars pp. 55 ]58 . In
. another masters thesis at Seoul University, Bong-Jeup Kim 1988 examined
compensating differentials in 1986 for and found that blue-collar workers receivedpositive compensating wage differentials for fatal injury risk but not for nonfatalinjury risk.
6
The 1987 change in the legal regime for unions
In 1987 the South Korean government changed their treatment of labor unions.
The Trade Union Law and the Labor Dispute Adjustment Law of December 1980had severely weakened the labor movement and collective bargaining. The lawstransferred the locus of organizing and bargaining authority from the Federation ofKorean Trade Unions to firm-level unions, which lacked the resources and exper-tise of the broader-based unions. These laws treated leaders of inter-firm and

KIM AND FISHBACK 234
inter-industry unions as „third parties'' in firm-level bargaining, such that they
could be charged with criminal offenses should they cooperate in firm-levelbargaining or striking. The laws made it more difficult to secure local-level unioncertification by asking for a „showing of interest'' that revealed the names andaddresses of union members and establishing a „cooling'' period of 10 days todetermine whether the certification application was complete. The employer there-fore had a period in which he could negotiate with or bribe enough workers to voidthe application for certification. Union leaders who started strikes during thecooling-off period could be sent to jail and would therefore lose their union officebecause the law prevented individuals with prison records from holding a unionoffice. The laws made it difficult to conduct legal strikes. In consequences, thenumber of strikes in Korea fell from over 400 in 1980 to 88 in 1982 before slowlyrising above 200 by 1986. Meanwhile, the number of union members as a percent-age of the nonagricultural workforce fell from 23 percent in 1979 to 15.1 percent in
 1986 Bognanno et al., 1994, pp. 356 ]357.
The treatment of unions changed dramatically in June 1987. Presidential candi-
date Tae-Woo Rho promised political and industrial democracy in his „6.29Minzoowha Suneun,'' or Declaration of Democracy. The declaration amended theTrade Union Law and Labor Dispute Adjustment Law in favor of labor. TheFederation of Korean Trade Unions and industrial unions were no longer treatedas third parties in firm-level negotiations and thus inter-firm and inter-industrybargaining was allowed once again. The minimum „showing of interest'' provisionwas repealed and the ten-day cooling period was reduced to three days. The„prison record'' and related tests for eligibility to hold union office were deletedand the provision allowing the state to dissolve a labor union and require unionofficials to be re-elected was dropped from the statute. Finally, labor unions weregiven greater freedom to strike. Cooling-off periods were shortened; the state'srole in determining the lawfulness of a strike notification report was eliminated;and laws authorizing the state to mandate mediation and compulsory arbitration
. were severely restricted Bognanno et al., 1994, pp. 358 ]359 .
The reforms led to sharp increases in both unionization and strikes. Union
membership rose sharply to 23.4 percent of the nonagricultural population by 1989and then began to drop off again. Meanwhile the number of strikes jumped from289 in 1986 to 3,749 in 1987 and then fell to 1,873 in 1988, 1,616 in 1989, andaround 300 in 1990. The democratic reforms in 1987 significantly changed thenature of collective bargaining in Korean industry. They allowed for much moreinteraction across firms and industries than had been allowed in the early 1980s. Atthe same time the changes had the potential to dramatically change the wagestructure, as greater strike activity and changes in bargaining strength becamemore important to the wage-setting process.
The increased role of labor unions could have led to either an increase or
decrease in the size of compensating differentials in Korea. Unions can increasethe size of compensating differentials in situations where the union providesworkers with better information about accident risks, counteracts an employers'

SOUTH KOREA, 1984 ]1990 235
monopsony, or more effectively represents the interests of inframarginal workers
who seek higher accident payments for accepting accident risk. Compensatingdifferentials potentially will be reduced when the union seeks to level wages forequity reasons or establishes rigid pay scales for jobs based on seniority rather than
. risk. David Fairris 1989 suggests in his political model of union behavior that the
 union's ultimate effect on compensating differentials depends upon among other
. factors the existing level of compensating payments, the strength of the union, the
extent of multiemployer bargaining, and the degree of conscious concern the uniongives to working conditions in contract negotiations. The latter aspect is oftendiminished when there are multiple unions, each representing different industriesor different jobs within the same industry, negotiating with the employers. Withmultiple unions there is no one central body negotiating an overall job evaluationformula that gives extra weight to dangerous jobs or industries. Wage differencesacross jobs are more subject to differences in the bargaining strength of the unionsrepresenting each industry or job category. The timing of wage adjustments in eachindustry may also be affected by the timing of negotiating rounds and strikeactivity.
Empirical work on the impact of unionization on compensating differentials
suggests that greater unionization is often associated with larger compensating
. differentials. Michael Moore and W. Kip Viscusi 1990, pp. 111 ]120 find that
unionized workers tend to receive higher compensating differentials in the United
. States. In international comparisons Kniesner and Leeth 1991 found that Aus-
tralian wages contained a larger risk premium for fatal accident risk than did wagespaid to U.S. workers and Japanese workers. They suggest that the higher Aus-tralian risk premium was possibly due to the major role played by unions and thegovernment in the wage-setting process across industries. Yet there are no guaran-tees that an increased role for unionization and government activity will always
. lead to larger compensating differentials. Kim and Fishback 1993 found in the
U.S. railroad industry that as more railroad jobs were unionized and governmentarbitration played a more extensive role in the wage-setting process, the riskpremia for accidents declined. The differences between Australia in the mid-1980sand U.S. railroads in the early 1900s point out why it is extremely important toexamine the institutions for wage-setting. In Australia in the mid-1980s, there wasan explicit channel for establishing salary supplements, „dirt pay,'' for exposure toworkplace accidents. In the U.S. railroad industry there were a wide range ofunions negotiating separately, and there was no explicit government arbitrationpolicy to take into account accident risk in choosing the final solution.
The 1989 workers' compensation reform in Korea
A key component of the accident compensation for workers in most nations is
explicit post-accident payments through workers' compensation programs. SouthKorea instituted a major institutional change in workers' compensation in 1989

KIM AND FISHBACK 236
that led to a substantial rise in benefit payments. As in many other countries,
South Korean workers harmed in workplace accidents arising out of or in the
course of employment receive benefits that replace lost wages as well as some form
of medical coverage. The wage replacement rates in Korea are similar to those
seen in other countries. Throughout most of the 1980s, the families of Korean
workplace accident victims received death benefits equal to up to 1000 times the
workers' daily wage. Such a payment would be about 3.57 times average annual
earnings, which is in the same ballpark as the payments of 2.67 times average
annual earnings in Australia in 1984 ]85, 3.93 times annual earnings in the United
States in 1978, and 3.57 times average annual earnings in Japan in 1986.7For
nonfatal injuries Korean workers' compensation programs paid out 62 percent of
the workers' weekly wage. In April of 1989 the South Korean government reformed
the workers' compensation program. The death benefit was raised by 30 percent
from 1000 times the worker's daily wage to 1300 times the daily wage, or roughly4.6 times annual earnings. Meanwhile, the percentage of the wage paid to workers
 with nonfatal injuries rose from 62 percent to 70 percent Williams, 1991, pp.
. 174]178; Yoon et al., 1996, pp. 12, 106 ]108 .
Earlier empirical work on the impact of changes in workers' compensation
benefits suggests that the rise in benefits in Korea in 1989 was likely to be
associated with a decline in compensating wage differentials. Moore and Viscusi
.1990, p. 24 summarized a series of studies for the United States that nearly all
show a reduction in wages or in compensating differentials when workers' compen-
. sation benefits were improved. Fishback and Kantor 1995 find similar results
when examining the introduction of workers' compensation in the early 1910s.
. Meanwhile, Kim and Fishback 1993 also found that increases in the extent of
liability imposed on railroads for workplace accidents reduced the size of compen-
sating differentials.
The estimation procedures and results
To examine the impact of the institutional changes on compensating differentials
we developed a panel data set for 1984 through 1990 from a series of sources that
describe the South Korean labor market. We obtained information on hourly
earnings, the extent of overtime pay, employee turnover, the average age and job
tenure of workers, and average firm size from the ministry of Labor of Korea,
which publishes information for each of 50 industries each year in the December
. issue of the Report on Monthly Labor Sur
¤eyEnglish translation is included , and
in the Sur¤ey on Basic Statistics for the Wage Structures . Both surveys use the Korea
Standard Industrial Code for industry classification, so the information from the
two publications can be merged easily. We obtained accident data for 61 industries
from the Ministry of Labor's Analysis for Industrial Accident . The industrial
classifications in this publication differ from those in the Korea Standard Industrial

SOUTH KOREA, 1984 ]1990 237
Code; therefore, we went through and matched the KSIC industries to the accident
industries. The matching choices we made are reported in Appendix Table A1.
To estimate compensating differentials for the South Korean labor market, we
follow the standard procedure of using a single-equation reduced form estimate.For some purposes it may be useful to examine the employers' cost curves or
. workers' indifference curves see Viscusi, 1993 , but our goal here is to measure
the final relationship between wages and accident risk after the interactions ofemployers' and workers' desires are complete. There is legitimate reason to believethat the factors influencing employers' demands for labor and workers' supply oflabor given the presence of workplace disamenities would influence the wage-setting process under all institutional settings. The differences we might expect tosee would be differences in the responsiveness of the institutions to these variousfactors, such that workplace disamenities or experience, would receive differingweights under differing institutional regimes.
We ran a fixed-effects analysis of hourly earnings as a function of the annual
fatal accident rate, the annualized nonfatal accident rate, and several controlvariables. The equation is estimated in semi-log form, using White's method forestimating standard errors in the presence of heteroskedasticity:
Log Male Hourly Earnings
.
sfFAR, NFAR, Control Variables, Industry Effects, Year Effects ..
The dependent variable in the regression is the log of hourly earnings for males
in each industry.8Hourly earnings for males are calculated by dividing total
 monthly cash earnings including regular and overtime payments and special cash
. payments by total hours worked including regular hours worked and overtime
. hours worked . The special cash payments account for roughly 17 percent of the
earnings of workers and might have been used to reward workers when the firmwas more profitable or for special holidays. The hourly earnings are deflated using
.
9the Korean Consumer Price Index 1985 s1.
Our measures of accident risk include three-year moving averages of the fatal
accident rate, FAR, which is the number of workers killed by job-related accidentsduring the year per thousand workers and the non-fatal accident rate, NFAR,which is the number of workers injured, disabled or who caught industrial diseasesduring the year per thousand workers. We chose three-year moving averagesbecause accidents, particularly fatal ones, are not very common random events. Webelieve a three-year moving average better represents the workers' and employers'prediction of the expected accident rate than just the evidence for one year.
10
The control variables include the average age of the male workers, their average
tenure in the industry, average firm size, and an estimate of turnover, the ratio ofworkers placed in jobs to the average number of workers.
11In the results reported
in Table 1 only the coefficient of the job placement ratio is statistically significant,and it implies that high rates of turnover in an industry were associated with lower

KIM AND FISHBACK 238
. Table 1. Results of fixed effects analysis for log hourly earnings for South Korea, 1984 ]1990
Specification 1 Specification 2 Specification 3 Specification 4
Std.
Mean dev. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat.
Intercept 7.508 92.58 7.477 98.75 7.458 89.45 7.506 85.34
Fatal accidents per 1000 0.485 0.837 0.094 4.78 0.087 3.95 0.154 2.67 0.204 2.96
workers, 3-year movingaverage
Nonfatal accidents per 30.648 25.031 y0.0021 y2.73 y0.0013 y1.63 y0.0012 y1.37 y0.0029 y3.46
1000 workers, 3-yearmoving average
Fatal accident rate times y0.033 y2.23 y0.032 y2.18
union-rule dummy
Nonfatal accident rate 0.0010 2.28 0.0010 2.40
times union-rule dummy
Fatal accident rate times y0.042 y2.60 y0.041 y2.46
workers' compensationchange
Nonfatal accident rate 0.0021 4.14 0.0021 3.97
times workers'compensation change
Fatal accident rate times y0.072 y1.19 y0.127 y1.76
union dummy
Nonfatal accident rate 0.0002 0.19 0.0015 1.56
times union dummy
Average number of 95.336 120.671 y0.0002 y0.79 y0.0003 y1.31 y0.0003 y1.58 y0.0002 y1.06
workers per firm
Average age 32.793 3.730 y0.00003 y0.03 y0.0005 y0.41 y0.0005 y0.36 y0.0002 y0.16
Average job tenure 4.097 1.735 0.004 0.34 0.005 0.45 0.0057 0.50 0.0053 0.45Ratio of overtime pay to 0.161 0.072 y0.040 y0.18 y0.049 y0.21 y0.043 y0.19 y0.039 y0.17
total earnings

SOUTH KOREA, 1984 ]1990 239. Table 1. Continued
Specification 1 Specification 2 Specification 3 Specification 4
Std.
Mean dev. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat. Coeff. t-stat.
Ratio of workers placed to 0.207 0.252 y0.255 y6.94 y0.221 y5.78 y0.217 y5.71 y0.241 y6.58
average number ofworkers
Year dummies
1985 0.140 0.348 0.024 1.81 0.028 2.21 0.027 2.18 0.024 1.871986 0.140 0.348 0.065 4.46 0.070 4.97 0.069 4.88 0.064 4.431987 0.140 0.348 0.039 1.51 0.044 1.65 0.046 1.72 0.047 1.841988 0.146 0.354 0.120 4.82 0.129 5.08 0.132 5.17 0.127 5.091989 0.146 0.354 0.269 9.65 0.246 8.78 0.249 8.84 0.276 9.941990 0.146 0.354 0.364 12.66 0.353 12.57 0.357 12.61 0.370 12.96
Industry dummies included included included includedN 321 321R-square 0.973 0.9755 0.9757 0.973Rbar-square 0.967 0.9696 0.9696 0.967
Data sources: See text.
Notes : The mean for the log wage is 7.57. he mean average hourly earnings is 2136 won per hour and the exchange rate we use for conversions for
the entire period in the text is 850 won per U.S. dollar. A White test rejected the hypothesis of homoskedasticity in the error terms; therefore, weused White-corrected standard errors.

KIM AND FISHBACK 240
wages. The qualitative results for the accident risk coefficients are basically the
same whether the control variables are included in the analysis or not. The industryeffects capture the impact of differences across industries that do not change overtime, while the year effects capture the impact of differences across years that arecommon to all industries in that year. F-tests reject the hypothesis that the
coefficients of the industry and year effects are equal to zero.
The fixed-effect results in specification 1 of Table 1 suggest the presence of a
positive compensating differential in South Korea over the period from 1984through 1990. The coefficient of the fatal accident rate is statistically significantand implies a 9.4 percent increase in hourly earnings for an increase in the fatalaccident rate by 1 per 1000 workers.
12Such an increase is well within the bounds of
the sample, since the standard deviation of the fatal accident rate is 0.84. The wageelasticity with respect to changes in the fatal accident rate is 4.56, which impliesthat a 1 percent increase in the accident rate raised the wage by 4.56 percent.
13For
comparisons with earlier research on Korea we can put this measure into value oflife terms. The average hourly wage in the entire sample was around 2136 won,which converted to $2.51 per hour at the average exchange rate for that period.The risk premium implies a value of life of $529,125 or approximately 94 times theaverage annual earnings of the period of roughly $5,629 per year. Since we areusing industry aggregates, the estimate of the risk premium might be biased
. downward see Viscusi, 1993 . However, our estimate is very close to the lower end
 of the range of estimates from individual-level data found by Sung-Joon Kim 1985,
. pp. 55 ]58 who found a range for the value of life of between $550,000 and $1.1
million.
On the other hand, the Japanese rKorean-style labor markets do not reward
workers for accepting nonfatal accident risk. The coefficient on the non-fatalaccident rate for specification 1 in Table 1 is negative. This is consistent with the
. estimates for nonfatal injury risk found by Sung-Joon Kim 1985 and Bong-Jeup
. . Kim 1988 for Korea, and Kniesner and Leeth's 1991 for Japan. In fact, in
. Kniesner and Leeth's 1991 estimates for Australia and for aggregated data in the
U.S. there were no statistically significant compensating differences for nonfatalaccident risk. The one exception in the Kniesner and Leeth study was that theyfound compensating differences for nonfatal accident risk for individual-level datain the United States. It is possible that the lack of compensating differences hereand in Kniesner and Leeth's study of Japan and Australia are caused by the use of
. . aggregated data. However, Kim 1985 and Kim 1988 found no risk premia for
nonfatal accident risk with individual-level data in Korea. Another possible expla-nation for the difference in the extent of compensating differentials for fatal andnonfatal accident risk is that workers' compensation benefits cover more of thewage loss for nonfatal accidents than for fatal accidents. In Korea the wagereplacement rate for nonfatal injuries was at least .62 throughout the period, whilethe fatal accident benefit was at most 5 times annual earnings. If workers have adiscount rate of 5 to 10 percent, the present value of the lost earnings from aworkers' death would range from 20 times to 10 times annual earnings. Another

SOUTH KOREA, 1984 ]1990 241
possible reason for the lack of a compensating difference for nonfatal accident risk
is problems with measurement error in the data. The severity of nonfatal accidentsranges from permanent paralysis to sprains that keep workers out of work for twoweeks. In the analysis the lack of more specific data means that the coefficient isthe same for both types of accidents. Our impression is that studies that use lostwork days tend to find compensating differences more often than studies relying onthe type of raw injury rates that we are forced to use here.
The panel allows us to examine the impact of the two major institutional changes
in the Korean economy during the 1980s. In 1987 the Korean government dramati-cally changed the rules under which labor unions operated. The rule changeimmediately led the number of strikes to rise from 276 in 1986 to 3,749 in 1987,1873 in 1988, and 1,616 in 1989. Membership in Korean labor unions surged from15.5 percent of the nonagricultural labor force in 1986 to 23.4 percent in 1989before falling to 21.7 percent in 1990. In 1989 the Korean government establisheda workers' compensation reform that increased the fatal accident benefits byapproximately 30 percent and raised the replacement rate paid for nonfatalaccident rates from 62 percent to 70 percent. The effects of these institutionalchanges on the intercept for the wage are captured by the coefficients for the yeareffects in the regressions, but their impact on the compensating differential are notcaptured.
We examine the effect of these institutional changes by interacting the fatal and
nonfatal accident rates with two dummy variables. The union-rules change iscaptured by a dummy with value zero from 1984 through 1986 and then one from1987 forward. The workers' compensation rule change is captured by a dummy witha value of zero from 1984 through 1988 and a value of one thereafter. We wereforced to use dummy variables to measure these changes because both institutionalchanges were national in scope and affected all workers, at least a priori, in thesame way.
14
The results in specification 2 in Table 1 show that the institutional changes had
dramatic effects on the compensating differentials for fatal and for nonfatalaccidents. The coefficient for fatal accidents shows the compensating differentialfor years prior to 1987, which implies a wage elasticity with respect to changes infatal accident risk of 4.21. The union rule change had a strong negative impact onthe risk premia for fatal accidents, lowering the wage elasticity with respect tochanges in fatal accident risk to 3.1. Since most studies of the impact of unioniza-tion on fatal risk premia suggest higher premia for unionized workers than fornonunionized workers, the fact that strengthening the unions' bargaining positionlowered the risk premia seems somewhat unusual. However, the rule changesunleashed a large number of strikes and bargaining activity in industries. Differ-ences in the bargaining strength of industries as well as differences in the timing ofthe impact of labor strife, could have led to substantial changes in wages in someindustries that were not matched in other industries. Real wages may have beenrising across industries in ways that imply that fatal accident risk was not a centralissue determining the wage changes across industry. In addition, the temporary

KIM AND FISHBACK 242
wage structure established by the union upheaval in 1987 and 1988 influenced the
relative wages across industries that followed, and it does not appear that there wasa reversion to the pre-1987 status quo in 1989 or 1990.
In fact, the rise in workers' compensation benefits compounded the decline in
the compensating differential for fatal accidents. The fatal accident risk premiumfor the period 1989 and 1990, when both rule changes were in force, is the sum ofthe coefficients of the fatal accident risk, the interaction with the union-rulechange, and the interaction with the workers' compensation rule change. Thus forthe years 1989 and 1990 the fatal accident risk premium falls to 0.012, implying awage elasticity with respect to changes in the fatal accident rate of 0.59. This is arelatively large drop given the change in workers' compensation benefits. Theimplicit value of life in the labor market falls by roughly $200,000. This is asubstantial drop in response to a change in workers' compensation benefits value,particularly because the workers' compensation death benefit for the averageworker rose only from 3.57 to 4.64 times annual income. Such a drop in the wagepremia is large enough that it appears that other factors may have been contribut-ing to the dropoff. It may be that the workers' compensation change is associatedwith other factors that contributed to the decline in the compensating differentialfor fatal accidents.
It should be noted that workers were experiencing substantial wage growth
during this period. The changes in 1987 and 1989 were associated with dramaticchanges in the intercept term of the logwage regression. In 1987 when the unionrepression declined and strike activity exploded, the 1987 year coefficient suggeststhat real wages were 3.9 percent above their 1984 level, holding other factorsconstant, implying a slight decline in wages from 1986. However, in 1988, 1989, and1990, workers experienced substantial growth in the wage intercept term to levels12 percent, 26.9 percent, and 36.4 percent above the 1984 level, respectively. Thus,the reduction in the compensating difference for fatal accident risk was offset by adramatic general rise in real hourly earnings in Korea.
When compared with their impact on the fatal accident coefficients, both the
union-rule change and the workers' compensation changes have precisely theopposite effects on the coefficients for nonfatal accident risk. The union rulechange raised the nonfatal accident coefficient from y0.0013 to y0.0003, although
this still implies no compensating differential for nonfatal accident risk. The 1989rise in workers' compensation benefits was associated with another increase of0.0021 to a level of 0.0024, which implies an implicit value of an injury of $132,445,which seems unusually large.
Many scholars have tested for the impact of unionization on compensating
differentials. Given the large impact of the change in the treatment of unions in1987, it is important to investigate the impact of unionization on compensatingdifferentials. We matched up information of the number of workers in the Korean
 unions with the KSIC industry codes Korea Employers' Association, various
. years . Given the uncertainties of matching the union information with the wage
information, we were hesitant to do more than create a zero-one union dummy for

SOUTH KOREA, 1984 ]1990 243
the presence of unions in the KSIC industry.15Essentially, the union dummy is
very close to a linear combination of the industry dummies. Therefore, the industrydummies are capturing the intercept effect of unionization on the wage. Inspecification 3 we include the impact of unionization on the compensating differen-tials by interacting our union dummy with the accident rates. The only substantialqualitative change in the results is that the coefficient for fatal accident risk, whichis focused on nonunion workers in the period prior to 1987 rises to 0.154, implyinga value of life that is 154 times annual income. There is a negative coefficient forthe union interaction with the fatal accident risk variable; however, we cannotreject the hypothesis of no difference between union and nonunion industries. Thecoefficient of the union interaction with the nonfatal accident rate is small and notstatistically significant. Essentially, the inclusion of the union interaction termsdoes not change the qualitative story with respect to the compensating differencesfor either fatal or nonfatal accident risk. In statistical terms, the inclusion of theunion interaction terms may not add any information. The results of an F-test
suggest that we cannot reject the hypothesis that the coefficients of the unioninteraction terms are all zero. We have also experimented with including interac-tions between the union variable and the rule-change interaction terms. Again theF-tests cannot reject the hypothesis that all of the coefficients of the union
interaction terms are zero.
In specification 4 in Table 1 we have also estimated the union effects while
eliminating the rule-change-accident-rate interaction terms. Again, the impact ofunionization on the compensating differential for fatal accidents is negative,although statistically significant only at the 90 percent level. The union compensat-ing difference for nonfatal accidents is higher but not in a way that is statisticallysignificant. When comparing specification 4 and specification 1, we again find thatthe results of an F-test suggest that we cannot reject the hypothesis that the
coefficients of the union interaction terms are all zero.
Conclusions
Changes in labor market institutions can lead to quite different rewards to workers
for their acceptance of accident risk. During the 1980s South Korean labor marketswent through two major institutional changes: a change in 1987 that gave laborunions more favorable treatment, and a reform of workers' compensation legisla-tion in 1989. In many settings unionization has been associated with increasedcompensating differentials. In Korea, however, the reduction in the repression oflabor union activities in 1987 unleashed a large number of strikes and bargainingrounds in the Korean economy. The resulting upheaval in the wage-setting processwas associated with a substantial decline in the compensating differences for fatalaccident risk. The 1989 increase in workers' compensation benefits was associatedwith an additional reduction in the size of Korean compensating differentials,which is a result that is similar to the findings of other empirical studies.

KIM AND FISHBACK 244
Appendix
Table A1. Matching classifications for wage, accident, and union data
Industrial
Union Federation of Union
KSIC Industry Name IACA Industry Name Class Union dummy
210 Coal Mining 100 Coal Mining 1 Mining 1230 Metal Ore Mining 101 Metal Ore Mining 1 Mining 1290 Other Mining 102 Quarrying 1 Mining 1311 Food Manufacturing 200 Food 2 0313 Beverage Industries 200 Food 3 0314 Tobacco 201 Tobacco 4 Gov't Monopoly 1
. Manufacturing Tobacco and Jinseng
321 Manufacturing 202 Textiles 5 Textiles 1
Textiles
322 Wearing Apparel 202 Textiles 5 Textiles 1
except Footwear
323 Leather and Products 202 Textiles 5 Textiles 1
of Leather
324 Manufacturing 202 Textiles 5 Textiles 1
Footwear
331 Manufacturing 203 Sawmill 6 0
Wood andCork Products
332 Furniture and 204 Wood Products 7 0
Fixtures
341 Manufacturing 205 Paper Products 8 0
Paper andPaper Products
342 Printing and 206 Newspaper and 9 Publication 1
Publishing Publishing
351 Manufacturing 209 Chemical Products 10 Chemistry Union 1
IndustrialChemicals
352 Manufacturing Other 210 Medical Supplies 10 Chemistry Union 1
Chemicals
353 Petroleum Refineries 11 0
Deleted
354 Manufacturing 211 Coal Products 12 0
MiscellaneousProducts ofPetroleum
355 Rubber Products 212 Rubber Products 13 Rubber Products 1356 Plastic Products 209 Chemicals 14 1361 Pottery, China and 213 China 15 0
Earthen-ware
362 Glass and Glass 214 Glass Products 16 0
Products

SOUTH KOREA, 1984 ]1990 245
. Table A1. Continued
Industrial
Union Federation of Union
KSIC Industry Name IACA Industry Name Class Union dummy
369 Other Non-metallic, 216 Cement Products 17 0
Mineral Products
371 Iron and Steel Basic 219 Metal Refining 18 Metal Union 1
Industries
372 Non-ferrous Metal 219 Metal Refining 18 Metal Union 1
Basic Industries
381 Fabricated Metal 221 Metal Products 18 Metal Union 1
Products
382 Machinery except 223 Machinery 19
Electrical
383 Electrical and 224 Electronic 20 0
Electronic MachineryMachinery
384 Transport 226 Shipbuilding 21 Automobile 1
Equipment
385 Medical, Optical 228 Measuring, Optical 22 0
and Measuring and ControllingMachinery Equipment
MusicInstrument
390 Other Manufacturing 229 Handmade 23 0
Industries Products
410 Electricity, Gas and 300 Electricity and Gas 24 Electricity 1
Steam
510 Construction General 400 General 25 0
Contractors Construction
520 Construction Special 400 General 26 0
Trade Contractors Construction
610 Wholesale Trade 900 Commission Sale of 27 0
Agricultural andFishery Products
620 Retail Trade 905 Other Service 28 0630 Restaurants and 905 Other Service 29 Tourism 1
Hotels
711 Land Transport 500 Railroads 30 Railroad 1712 Water Transport 504 Water 31 Water Transportation 1
Transportation
713 Air Transport 506 Air Transportation 32 0719 Service Allied to 505 Transportation 33 Transportation Service 1
Transport Service in in Harbor Loading
Harbor Loading and Unloadingand Unloading
720 Communication 510 Communication 34 Communication and1
Postage
Deleted 600 Logging 35 0
810 Financial Institutions 905 Other Service 36 Banking and Financing 1
820 Insurance 905 Other Service 37 Insurance 1830 Real Estate 905 Other Service 38 0

KIM AND FISHBACK 246
. Table A1. Continued
Industrial
Union Federation of Union
KSIC Industry Name IACA Industry Name Class Union dummy
840 Business Services 901 Building 39 0
Management
920 Sanitary and Similar 902 Health Service 40 0
Services
930 Social and Related 905 Other Service 41 0
CommunityServices
940 Recreational and 905 Other Service 42 0
Cultural Services
950 Personal and 905 Other Service 43 0
HouseholdServices
Notes
. . 1. For a summary of modern studies see Moore and Viscusi 1990 and Viscusi 1993 . For a summary
. of historical results, see Fishback 1998 .
2. The average annual earnings are rough estimates based on samples for Australia, Japan, and the
. United States in Kniesner and Leeth 1991 . Information on Korea is from the sample in Table 1.
Kneissner and Leeth reported Japanese information on wages and exchange rates for 1986. They
reported earnings information for 1984 ]1985 in Australia and 1978 in the U.S. We raised the
 earnings to their 1986 levels using indexes on wages from International Monetary Fund 1998, pp.
. 208]213, 522 ]525, 540 ]545, 892 ]897 . The monetary values for Korea and Australia were con-
verted to U.S. dollars in 1986 by first converting to 1986 in the home currency and then using 1986
exchange rates with the U.S. dollar. Adjustments were made based on CPI information for Australia
and the United States and exchange rate information for all countries from International Monetary
. Fund 1998, pp. 208 ]213, 522 ]525, 540 ]545, 892 ]897 .
3. The fatal accident rate in Korea during the 1980s ranged between 0.26 and 0.38, compared with 0.01
to 0.03 in Japan, 0.02 to 0.04 in the U.S., and 0.076 in New Zealand. See International Labor
. Organization 1991, Table 30, 1995, Table 30 .
. 4. Hee-Tak Ahn 1996 finds that Korean firms in the mid-1990s began to introduce an annual salary
system in small number of companies. The annual salary system has been restricted to managerial,
professional, and technical staff.
. 5. Kim 1985 matched individual-level information on wages from the Korean Sur¤ey for Wages by Job
in 1982 with aggregate accident data in Analysis for Industrial Accident . After excluding data for
. white-collar jobs managerial and administrative jobs he estimated wage coefficients for fatal and
nonfatal accidents after controlling for age, education, years of service, and sex.
. 6. Kim 1988 matched 6,770 individual observations with aggregate accident rates for 30. He
estimated separate equations for white-collar and blue-collar jobs, while controlling for sex, marital
. status, education, experience age minus education minus 6 , years of service, career year, monthly
working days, monthly overtime working hours, region, firm-size, and the fatal and nonfatal accident
rates.
7. The workers' compensation ratios for Australia and U.S. are based on mean benefits from Kniesner

SOUTH KOREA, 1984 ]1990 247
. and Leeth's 1991 sample. Rough estimates of Japanese and Korean workers' compensation death
. benefits are based on rules reported in Williams 1991, pp. 169 ]170, 177 .
8. When we use female wages or we confine the sample to production workers, we see results that are
qualitatively similar to those reported in Table 1.
9. Since the special cash payments vary substantially from year to year, it is possible that risk averse
workers would demand that the compensating differential for accident risk be incorporated into the
regular wage and not into the special cash payment. To examine this possibility, we have created an
alternative wage variable that excludes the special cash payment from the hourly earnings calcula-
tion. The compensating differences for the earnings that exclude the special cash payment are
similar to the ones reported in Table 1.
10. The moving average is the mean of the value for the current year, the previous year, and two years
before. We performed the analysis with the current year accident rate, and the compensating
differences are smaller by about one-third. We believe the lower risk premia estimated is caused by
annual variations in the accident rate that would influence the workers or employers' estimates of
the true accident rate only as part of a Bayesian updating of past experience. Thus, annual
fluctuations in the accident rate would overstate the fluctuations in the accident risk perceptions of
the participants in the labor market. For a discussion of how Bayesian updating influences the
. accident perceptions of workers and employers, see Viscusi 1992, pp. 114 ]123 .
11. We also have estimates of the number of workers displaced from jobs. However, the placement and
displacement ratios had a correlation of 0.98 and their means in the sample were almost identical;
therefore we included only the placement measure. Inclusion of both measures had very little
impact on the results reported in Table 1.
12. The coefficient of the fatal accident variable when the log wage is run against just the accident risk
variables implies a 22 percent increase in the wage for each increase of one death per thousand
workers. When we add the fixed-effects to the regression, the coefficient implies a 10 percent
increase.
13. The 4.56 wage elasticity with respect to changes in the fatal accident rate is relatively high. Kniesner
and Leeth found elasticities of 2.97 in Australia in 1984 ]1985, 2.27 in the U.S. in 1978, and 1.42 in
Japan in 1986. It should be noted however, that a high elasticity does not necessarily imply a high
value of life estimate for Korea because Korean wages were only about one-fifth the level of U.S.
wages in 1986 and the mean accident rates in Korea were substantially higher than in the other
. three countries. Michael Moore and W. Kip Viscusi 1990, p. 14 find implicit values of human lives
in 1987 dollars ranging from $700,000 to $6.2 million in a series of 13 studies using individual-level
data in the U.S. See also a longer list of studies using 1990 as the base year for comparisons in
. Viscusi 1993, pp. 1925 ]1926 .
14. We have been unable to obtain industry-level strike data.15. We have experimented with alternative measures of the impact of unions, including the growth rate
in union membership and a rough estimate of the percentage of workers in each industry that were
unionized, and the results are essentially the same as reported here.
References
. Ahn, Hee-Tak. 1994 . Current Status of Merit Rating in Korean Companies , Korea Employers Federation.
. Ahn, Hee-Tak. 1994 . „A Study on the Ability-Based Wage System in Korean Enterprises,'' Insawanriy –
. onku 18, 183 ]216 in Korean .
. Ahn, Hee-Tak. 1996 . „The Transition from a Seniority-based Pay System to an Annual Salary System
in Korean Enterprises,'' Keio Business Re ¤iew33, 1 ]11.
. Bognanno, Mario F., John W. Budd, and Young-Myon Lee. 1994 . „Institutional Turmoil and Strike
. Activity in Korea,'' Journal of Industrial Relations Sept. , 353 ]369.

KIM AND FISHBACK 248
. Fairris, David. 1989 . „Compensating Wage Differentials in the Union and Nonunion Sectors,''
. Industrial Relations 28 Fall , 356 ]372.
. Fishback, Price V. 1998 . „Operations of „Unfettered'' Labor Markets: Exit and Voice in American
. Labor Markets at the Turn of the Century,'' Journal of Economic Literature 36 June , 722 ]765.
. Fishback, Price V., and Shawn Everett Kantor. 1995 . „Did Workers Gain from the Passage of
. Workers' Compensation Laws?'' Quarterly Journal of Economics 110 Aug. , 713 ]742.
. Gruber, Jonathan, and Alan Krueger. 1991 . „The Incidence of Mandated Employer-Provided Insur-
. ance: Lessons from Workers' Compensation Insurance.'' In D. Bradford ed. , Tax Policy and the
Economy , Cambridge, MA: MIT Press and NBER, pp. 111 ]143.
. Hatton, Timothy J., and Jeffrey G. Williamson. 1991 . „Unemployment, Employment Contracts, and
. Compensating Wage Differentials: Michigan in 1890s,'' Journal of Economic History 51 Sept. ,
605]632.
. International Labor Organization. 1991, 1995 . Yearbook of Labor Statistics , Geneva: ILO, 1991 and
1995, Table 30.
. International Monetary Fund. 1998 . International Financial Statistics Yearbook ,1988 . Washington, DC:
Publication Services IMF.
. Kim, Bong-Jeup. 1988 . „An Analysis of Compensating Wage Differentials for Job Hazard,'' A Thesis
 for the Degree of Master of Science in Seoul National University, January in Korean with abstract in
. English .
. Kim, Seung-Wook, and Price Fishback. 1993 . „Institutional Change, Compensating Differentials, and
. Accident Risk in American Railroading, 1892 ]1945,'' Journal of Economic History 53 Dec. , 796 ]823.
. Kim, Sung-Joong. 1985 . „Compensating Wage Differentials for Job Hazards in Korea,'' A Thesis for
the Degree of Master of Science in Cornell University, June.
. Kniesner, Thomas J., and John D. Leeth. 1991 . „Compensating Wage Differentials for Fatal Injury
Risk in Australia, Japan, and the United States,'' Journal of Risk and Uncertainty ,7 5]90.
. Korea Employers Federation. various years . Yearbook of Labor Statistics .
. Lanoie, Paul. 1992 . „Government Intervention in Occupational Safety: Lessons from the American
and Canadian Experience,'' Canadian Public Policy 18, 62 ]75.
. Lewchuk, Wayne. 1991 . „Industrialization and Occupational Mortality in France prior to 1914,''
Explorations in Economic History , 28, 344 ]366.
.  Ministry of Labor, Republic of Korea. various years . Analysis for Industrial Accident , each year in
. Korean .
.  Ministry of Labor, Republic of Korea. various years . Korean Standard Industrial Classification , 1984 in
. Korean .
. Ministry of Labor, Republic of Korea. various years . Report on Monthly Labor Sur ¤ey, Dec. issue of
each year.
. Ministry of Labor, Republic of Korea various years . Year Book of Industrial Accident Insurance Affairs ,
. 1991 in Korean .
. Moore, Michael J., and W. Kip Viscusi. 1990 . Compensation Mechanisms for Job Risks :Wages ,Workers '
Compensation ,and Product Liability . Princeton, NJ: Princeton University Press.
. Viscusi, W. Kip. 1993 . „The Value of Risks to Life and Health,'' Journal of Economic Literature
.31 Dec. , 1912 ]1946.
. Viscusi, W. Kip. 1992 . Fatal Tradeoffs :Public and Pri ¤ate Responsibilities for Risk . New York: Oxford
University Press.
. White, Halbert. 1980 . „A Heteroscedasticity-Consistent Covariance Matrix Estimator and a Direct
Test for Heteroscedasticity,'' Econometrica 48, 817 ]838.
. Williams, C. Arthur, Jr. 1991 . An International Comparison of Workers 'Compensation . Boston: Kluwer
Academic Publishers.
Yang, Byong-Moo. „Trends, Problems and Directions for Improvement of Pay System in Korea Firms,''
unpublished mimeograph.
. Yoon, Joduk, et al. 1996 . Research on the Payment System of Industrial Accident Compensation and
. Insurance in Korean , Korea Labor Institution.