Training, Technological Changes, and Displacement

1 downloads 0 Views 180KB Size Report
By matching industry/occupation data on training to displaced worker data from the .... showed that the longer training period to become fully qualified for a job ...
Training, Technological Changes, and Displacement

Younghwan Song ∗



Department of Economics, Union College, Schenectady, NY 12308, USA. E-mail: [email protected]. Tel: 518-388-8043. Fax: 518-388-6988.

Abstract By matching industry/occupation data on training to displaced worker data from the Current Population Surveys, this paper analyzes why many older workers were displaced by technological changes in the 1990s, and why these workers incurred large earnings losses. When technological changes depreciate the existing stock of firm-specific human capital, older workers who receive higher wages from the sharing arrangement of the returns to investment in firm-specific human capital are dismissed as firms find it unprofitable to retain them. These displaced workers have higher predisplacement wages with steeper wage-tenure profiles, and hence incur larger earnings losses after displacement than other displaced workers.

Keywords: Displacement, Technological change, Human capital

2

Even as the U.S economy was enjoying an unprecedented expansionary period during the 1990s, the media were constantly reporting stories of older workers being involuntarily separated from well-paying jobs at the height of their careers. Many middle-aged workers were displaced from stable jobs in the name of downsizing and corporate restructuring and consequently incurred large earnings losses. Such unanticipated mass displacement of older workers and the detrimental economic consequence of displacement puzzled policy-makers and researchers, who believed that older workers usually face lower risks of displacement than do younger ones, and prompted them to search for explanations. The explanations provided in the literature are, however, though informative, still incomplete. Exploring the implication of technological change on the probability of displacement by reason and age, Aaronson and Housinger (1999) first showed that the probability of displacement by position abolition, reflecting the wave of corporate downsizing and restructuring, was increased by technological changes. But they found little evidence for the hypothesis that the technology-displacement relationship disproportionately affects older workers.

In contrast,

Rodriguez and Zavodny (2000) found that the pool of older workers at risk of displacement had increased because aging baby boomers had changed the age-distribution of the labor force, and the probability of overall displacement among older workers had indeed increased in the 1990s owing to technological changes. But unlike Aaronson and Housinger, Rodriguez and Zavodny failed to recognize that the mechanisms of displacement differ by cause of displacement. Furthermore, even though it is well known that the probability of job loss is strongly monotonically declining in tenure (Parsons 1972; Mincer and Jovanovic 1981; Farber 1993), both studies had no control for job tenure and confounded the effects of age and tenure in their analyses. Finally, though Farber (1997) confirmed that older workers displaced due to position abolition incurred larger earnings losses than did other displaced workers during the 1990s, no study has provided a consistent explanation for the detrimental economic consequences of displacement for these workers. In this paper I try to fill this gap in the displacement literature by analyzing why many

3

older workers were displaced by technological changes 1 in the 1990s, and why these workers incurred larger earnings losses than did other displaced workers. I explain these phenomena as arising out of the depreciation of firm-specific human capital under technological change. The surplus from the sharing arrangement of the costs of and returns to investment in firm-specific human capital reduces firms’ incentives to dismiss workers who have firm-specific human capital. When technological changes depreciate the existing stock of firm-specific human capital, however, older workers who receive higher wages from investment in firm-specific human capital are dismissed as firms find it unprofitable to retain them. By showing that workers displaced by position abolition have higher predisplacement wages, this paper’s empirical results provide yet another evidence of the wage rigidity derived from high transaction costs in the sharing arrangement of the costs of and returns to investment in firmspecific human capital (Hashimoto 1981; Antel 1985). This paper also presents empirical results that run counter to the argument made by Gibbons and Katz (1991) for the signaling effects of layoffs. Workers displaced by position abolition do have larger earnings losses, not because these laid-off workers are regarded as lemons and receive lower earnings in their new jobs, but rather because they had higher predisplacement earnings.

Firm-Specific Human Capital and Displacement According to human capital theory, investment in firm-specific human capital makes the value of a worker’s marginal product higher within the firm than it is for his outside alternatives. The costs of and returns to such investment are shared between the worker and the firm, so as to reduce the possibility of losing the investment owing to turnover initiated by either side (Becker 1962; Hashimoto 1981). This sharing arrangement sets the worker’s wage level lower than the value of his marginal product but higher than his best alternative wage outside, which reduces both the 1

Similar to Aaronson and Housinger (1999) and Rodriguez and Zavodny (2000), I use technology levels and technological change interchangeably under the assumption that jobs with high levels of technology levels

4

firm’s incentive to dismiss the worker and the worker’s incentive to quit the job. When economic downturns lower the value of the marginal product, the firm will not lay off workers who have firm-specific human capital, not as long as there is a surplus from the sharing of the returns to investment in firm-specific human capital. Given the fact that training is the main medium of investment in firm-specific human capital, therefore, an inverse relationship between firm-specific training and layoff is to be expected. By using data drawn from the Panel Study of Income Dynamics (PSID), Mincer (1988) showed that the longer training period to become fully qualified for a job reduces layoff rates. 2 Because a direct measurable distinction between general and specific human capital investments was not available in the PSID, he assumed a positive correlation between total volumes of training and the amounts of both general and specific training received in firms. Lynch (1991, 1992) succeeded in distinguishing firm-specific training from general training by using detailed training data drawn from the National Longitudinal Survey of Youth (NLSY). She found that company training is highly firm-specific and the receipt of company training reduces the probability of turnover, though she did not separate layoffs from other types of separation from an employer. The job security resulting from the sharing of the costs of and returns to investment in firm-specific human capital is not, however, unchangeable. Technological changes can undermine that job security in the case of older workers (Mincer and Higuchi 1988; Bartel and Sicherman 1993). Higher rates of technological change depreciate the existing stock of firm-specific human capital at a faster rate, and will prompt the need for new investment in firm-specific human capital. Because older workers have a shorter span of working life over which the investment can be recovered, and often have higher opportunity costs owing to their higher wages, firms will be reluctant to train older workers unless the existing stock of human capital is highly complementary will have greater technological change over time. In the previous studies on the effects of firm-specific human capital on layoffs, variables other than training were included as a proxy for firm-specific human capital. Parsons (1972), for example, examined the effects on layoffs of the various factors associated with firm-specific human capital investment. Mincer and

2

5

to the new investment. Without retraining, the value of the marginal product from these older workers will fall below the wage, due to the fast depreciation of firm-specific human capital, and the firm will either lay off these older workers or have them take early retirement. 3 However, if both firms and older workers could agree on the decreasing value of the marginal product, firms could provide lower wages and retain their older workers instead of laying them off. But owing to high transaction costs, the post-investment wage level, fixed by the optimal sharing ratio of the expected returns to investment, is not renegotiated and inefficient separations may occur as both parties make unilateral separation decisions (Hashimoto 1981). The prediction of fixed wages owing to high transaction costs is supported by empirical evidence as well. Antel (1985) provided empirical test results that are consistent with the high transactions cost model and the resulting fixed wages. Also, based on the survey of 184 firms, Campbell and Kamlani (1997) provided answers consistent with the explanation of wage rigidity offered by firm-specific human capital theory: firms do not cut wages across the board because they fear the loss of firm-specific human capital when experienced workers quit, even if they lay off less productive workers during recessions. Older workers are laid off in the same manner, without renegotiations or wage cuts, as technological changes depreciate their firm-specific human capital. Then, these dismissed workers have steeper wage-tenure profiles and higher wages than do other workers. First, it is because these workers are laid off from jobs providing firm-specific human capital investment.

As

compared with other jobs that provide no investment in firm-specific human capital, jobs involving such investment provide workers with steeper wage-tenure profiles and higher wages through the Jovanovic (1981) showed an inverse relationship between slopes of tenure-wage profiles and turnover. Mincer and Higuchi (1988) showed that sectors undergoing more rapid technological changes tend to have mandatory retirement rules and an earlier retirement age in Japan. Bartel and Sicherman (1993) also showed that unexpected technological shocks induce the earlier retirement of older workers.

3

6

sharing arrangement of the costs of and returns to investment. 4 Second, it is because when firms need to lay off workers with firm-specific human capital, they seek to minimize their losses by first laying off workers who have a larger share of the returns to investment in firm-specific human capital. Given the same level of investment in firm-specific human capital, a larger share of the returns being held by workers implies steeper wage-tenure profiles and higher wages. The literature on displaced workers has shown that displaced workers often incur large earnings losses. 5 Such large earnings losses of displaced workers, especially those with longer tenures, often are ascribed to the loss of firm-specific human capital (Addison and Portugal 1989; Topel 1990; Kletzer 1996). Given that workers displaced due to technological changes have steeper wage-tenure profiles, such workers have more to lose when displaced. Hence earnings losses will be greater, in the case of workers displaced due to technological changes. Three empirically testable hypotheses thus can be formulated, with respect to the relationship between firm-specific human capital and displacement under technological change. First, the inverse relationship between firm-specific training and displacement could be altered for older workers, when technological changes depreciate firm-specific human capital.

Second,

workers displaced because of the depreciation of firm-specific human capital under technological change have steeper wage-tenure profiles and higher wages, owing to the fact that they have a larger sharing ratio of the returns to investment in firm-specific human capital than do other workers. Third, workers displaced due to the depreciation of firm-specific human capital under technological change incur larger earnings losses than do other displaced workers.

4

In addition to the sharing arrangement of the costs of and returns to investment in firm-specific human capital, there are other explanations for rising wage-tenure profiles and higher wages. To list a few, Salop and Salop (1976) presented an adverse selection model of wage growth, where firms use a wage-tenure profile as a means of screening workers by the probability of quitting their jobs. Lazear (1979) has given us a delayed payment contract model that induces workers not to shirk. In addition, Altonji and Shakotko (1987) regarded the strong cross-sectional relationship between tenure and wages as being a result of heterogeneity bias. Only firm-specific human capital theory, however, is consistent with increased layoffs of older workers owing to technological changes. 5 For a review of the literature on displacement, see Fallick (1996) and Kletzer (1998).

7

Data For empirical analyses, I have used data drawn from three different supplements to the Current Population Survey (CPS). First, the 1996, 1998, and 2000 Displaced Workers Surveys (DWS) provide samples of displaced workers and comparable nondisplaced workers. The DWS has been conducted biennially since 1984. The DWS asks workers older than 20 years of age if, in the prior five years (prior three years, starting with the 1994 DWS), they have lost or left a job because of a plant closing, slack work, a position or shift abolished, or other reasons. 6 With respect to the workers who answered affirmatively, detailed information was collected about the predisplacement job, including tenure, weekly earnings, and full-time/part-time status. Following Farber (1993, 1997), I have used nondisplaced workers employed as of the survey date from the 1996, 1998, and 2000 DWS, for the relevant pool of workers who were at risk of losing a job during the time-period of the prior three years. 7 The data on nondisplaced workers in other years lack detailed job-characteristic information, especially with respect to job tenure. In these three years both the DWS and the Job Mobility and Tenure Supplements were conducted as supplements to the February CPS, and job-tenure information from all employed workers was made available. In my sample, displaced workers are those who were between the ages of 20 and 61 when permanently displaced from a private-sector job because of a plant closing, slack work, or a position or shift being abolished. Workers displaced from agricultural and construction jobs were not included in the sample. 8 The comparable pool of nondisplaced workers consists of workers who were between the ages of 20 and 61 and who were employed in private-sector jobs as of the survey date. It too excludes workers employed in agricultural and construction jobs. 6

Other reasons include “seasonal job completed,” “self-operated business failed,” and “some other reason.” Such cases are counted as nondisplaced workers in the sample. 7 Another data set that includes a nationally representative sample of displaced workers is the PSID, which also contains information on nondisplaced workers (Topel 1990; Stevens 1997). The displaced workers sample from the PSID is very small, however, and includes workers who were fired as well as those who have been laid off. Also, one cannot distinguish workers displaced by insufficient work from those displaced by shift/position abolition in the PSID. 8 The practice of excluding displaced workers from the agricultural and construction industries is a common one in the literature on displaced workers, since it is hard to define displacement in those industries.

8

The reported cause of displacement provides information as to whether a worker was displaced due to shortfalls in demand (plant closing or slack work) or due to other reasons such as downsizing or corporate restructuring (position abolition). 9 Since previous studies have found that the mechanisms of displacement differ by cause of displacement (Farber 1993, 1997; Aaronson and Housinger 1999), I will categorize displaced workers accordingly in the empirical analysis. The displaced workers in the sample were involuntarily separated from their jobs in the period 19931999, when the economy was enjoying the long expansionary period following the recession of the early 1990s. 10 The sample used in the analyses of the probability of displacement contains 96,738 workers, including 89,338 nondisplaced workers and 7,400 displaced workers. When it comes to comparisons of wages and wage-tenure profiles, subsets of the whole data set are used, because earnings data for nondisplaced workers are available only for the outgoing rotation groups of the DWS. In my final analysis of the earnings losses of displaced workers, the sample includes only displaced workers who had been reemployed in wage and salary employment as of the survey date. Second, the 1991 January training supplements to the CPS provide the industry/occupation level training data. 11 In those supplements, two questions regarding the incidence of training were asked: “Did you need specific skills or training to obtain your current (last) job?” and “Since you obtained your present job, did you take any training to improve your skills?” The first question is about job-qualification training, the second about skill-improvement training.

Since job-

qualification training often precedes current employment, and is unlikely to be financed by the current employer, I have used data on skill-improvement training only. Workers answering “yes” to these questions were asked to select from a detailed list the sources of each type of training. The 9

Cappelli (2000), who defined downsizing as reductions in jobs driven by the desire for operating efficiencies, distinguished it from the more typical layoffs associated with shortfalls in demand. 10 During this period the national annual unemployment rate monotonically decreased from 6.91 percent in 1993 to 4.23 percent in 1999. 11 The first training supplement was conducted in 1983. See U.S. Department of Labor (1985, 1992), for a detailed report drawn from the worker training supplements.

9

source list for skill-improvement training includes school, a formal company-training program, informal on-the-job training, and others. Individuals were allowed to pick more than one source. Among the four sources of skill-improvement training, I classify “formal company training program” and “informal on-the-job training” as firm-specific training. While company training is highly firm-specific (Lynch 1991, 1992), some of the informal on-the-job training also is firmspecific (Loewenstein and Spletzer 1999). 12 Other sources of training—school and others—are regarded as general training and also are included in the analyses. The training data show the percentages of workers who received each type of skill-improvement training, broken down by 426 industry/occupation cells: 213 three-digit industries, multiplied by the two basic occupations groups of white-collar and blue-collar workers. 13 The CPS sampling weight has been used for weighting. The sample workers used in the calculation are wage and salary workers holding private jobs and between 20 and 61 years of age at the time of the survey. Finally, as measures of technological change, I have used (1) industry/occupation level computer-use data, drawn from the October 1993 CPS School Enrollment supplement, and (2) the ratio of scientific and engineering employment to total employment at the industry level, calculated from the full-year 1993 CPS Outgoing Rotation Groups. 14 Jorgenson, Ho, and Stiroh (2008) showed that productivity growth slowed during the 1970s, 1980s, and early 1990s, and then increased substantially in the mid 1990s due to productivity growth and massive investment in information technology. Capturing such an expansion in information technology, the computer-use 12

In 1993, the NLSY included questions about the transferability of skills learned in formal and informal training programs. While about 51 percent of individuals who received informal training reported that all or almost all of the skills they learned in the informal training were useful at other employment, the rest replied that some of the skills were not transferable. For details, see Loewenstein and Spletzer (1999). 13 U.S. Department of Labor (1985, 1992) showed a stark difference in proportions of workers who received training, between the white-collar and blue-collar occupations. White-collar occupations include managers and administrators, professional and technical workers, clerical workers, and sales workers. Blue-collar occupations include craft and kindred workers, operatives, laborers, transport operatives, and service workers. 14 Computer-use surveys were conducted in 1984, 1989, and 1993. See Krueger (1993) and Autor, Katz and Krueger (1998) for a detailed description of computer-use survey data and for changes in computer use over the period. Allen (2001) showed that the ratio of scientific and engineering employment to total employment was highly correlated with the R&D intensity measures published by the National Science Foundation.

10

data show the percentages of workers who directly use a computer keyboard at work, and they have been calculated for the 426 industry/occupation cells. The ratio of scientific and engineering employment to total employment has been calculated for the 213 industry cells. The CPS sampling weight has been used for weighting in both calculations. The sample workers used in both calculations are wage and salary workers holding private jobs who at the time of the survey were between 20 and 61 years of age. Both Aaronson and Housinger (1999) and Rodriguez and Zavodny (2000) employed computer use as a proxy for technological change. And the ratio of scientists and engineers to total employment is conceptually similar to the list of high-technology industries based on share of R&D employees used by Aaronson and Housinger (1999). 15

Effects of Firm-Specific Training on the Probability of Displacement The first step of the investigation is to discover whether the inverse relationship between firmspecific training and displacement was in any way altered for older workers under the technological changes of the 1990s. For this purpose I need to verify with my sample that position abolition was indeed the reported reason when older workers were displaced due to technological changes. Table 1 presents the descriptive statistics for the sample of nondisplaced workers and displaced workers categorized by cause of displacement. Workers displaced by position abolition are mostly white-collar, older, and better educated than any other workers inclusive of nondisplaced workers.

They also have longer job tenure than do other displaced workers.

Consistent with Aaronson and Housinger (1999), these workers are displaced from jobs experiencing rapid technological change, measured as both higher levels of computer use and a higher ratio of scientists and engineers to total employment. The fact that these workers were displaced despite their jobs’ having more incidences of both types of firm-specific training— 15

Aarons and Housinger additionally used investment in computer equipment, total factor productivity, and output per hour (labor productivity). These measures of technological change are not included in this paper

11

company training, and informal on-the-job training—strongly suggests a possible depreciation of firm-specific human capital due to technological changes. Table 2 reports the results of a multinomial logit estimation, for a detailed examination of the effects of worker characteristics on the probability of displacement. Since the displaced workers cited one of the three reasons of displacement exclusively, it is necessary to employ the multinomial logit estimation so as to analyze the different probabilities of displacement, by reason. 16

The reported regression includes the following independent variables: three age

dummies, a female dummy, a full-time dummy, three education dummies, four job-tenure dummies, a nonwhite dummy, three region dummies, a SMSA dummy, eight industry dummies, eight occupation dummies, and two year-of-survey dummies. Panel A shows the coefficients, and panel B shows the marginal effects evaluated at the sample mean. In panel A, the excluded base category is nondisplaced workers. The coefficients on the age dummy variables reveal that the probability of displacement in the 1990s was not lower for older workers, as compared to younger workers between the ages of 20 through 29.

The probabilities of displacement through plant closing or slack work do not

significantly differ by worker’s age. In the case of displacement by position abolition, however, older workers have significantly higher probabilities of displacement than do younger ones. This tells us that the increased probability of overall displacement of older workers in the 1990s, found in Rodriguez and Zavodny (2000), is mostly due to position abolition, which reflects downsizing and corporate restructuring. The negative and significant coefficients on the tenure dummy variables, except in the case of the two-through-four years of the tenure dummy variable, make it clear that senior workers with longer tenure still have lower risks of displacement. By examining the coefficients on both the age because they are not available for all industries.

12

and the tenure dummy variables, we are able to learn that it is older workers, not senior workers, who in the 1990s faced higher risks of displacement by position abolition. Because of a shorter span of working life over which the investment can be recovered, firms are reluctant to retrain and retain older workers under conditions of rapid technological change. Panel B shows that when evaluated at the sample mean, a 50-year-old worker with 20 years of tenure faces about the same probability of displacement by position abolition as does a 20year-old worker with tenure of less than one year (.007 - .007 = 0). Senior workers, however, still are kept in the firm, as long as they are not too old. Thus a 40-year-old worker with 10 years of tenure faces a lower risk of displacement by position abolition than does a 20-year-old worker with tenure of less than one year (.005 - .007 = - .002). Finally, the education dummy variables show that the more-educated workers are less likely to be displaced through plant closing or slack work, and yet more likely to be displaced through position abolition. This seems to be the case because high-tech industries have more educated workers, and position abolition is the method of displacement in these industries, as the following analysis shows. Table 3 reports the effects of the two measures of technological change on the probability of displacement. Each row of the table is from a separate regression, with each measure of technological change being added individually to the multinomial logit model presented in table 2. Table 3’s estimation results vary, depending on the technology proxy employed, for displacement by plant closing or slack work.

While computer use significantly lowers the probability of

displacement by plant closing, the ratio of scientists and engineers to total employment has only insignificant positive effects on it. The probability of displacement by slack work is insignificantly reduced by high computer use, whereas it is significantly increased when more scientists and 16

For ease of interpretation of the estimates, Farber (1997) and Aaronson and Housinger (1999) both used separate Probit analyses of the probability of displacement, by reason. When it comes however to analyzing the structural differences in the probability of displacement for three different causes, the multinomial logit estimation is more appropriate.

13

engineers are employed. When it comes however to displacement by position abolition, we see, in table 3’s last column, that the results are more robust. Both measures of technological change significantly increase the probability of displacement by position abolition, even when other variables are being controlled for. These results are consistent with the finding by Aaronson and Housinger (1999) that position abolition is more likely to occur in the high-tech industries. But the results in table 2 that older workers have significantly higher probabilities of displacement by position abolition are different from Aaronson and Housinger. There may be a few reasons for this. First, the effects of age on the probability of displacement might have changed between 1980s and 1990s. While this paper uses the data of displaced workers drawn from the 1994, 1996, and 1998 DWS, the data used by Aaronson and Housinger included workers drawn from the DWS of 1986, 1988, 1990, 1992, 1996, and 1998. Second, job tenure and age are positively correlated, but Aaronson and Housinger (also Rodriguez and Zavodny) have failed to control for job tenure. Thus the effects of age are negatively biased in their analyses of the probability of displacement. Furthermore, even though many displaced workers do not have particularly long job tenure (Kletzer 1998), 17 Aaronson and Housinger used a selected sample of workers who have more than five years of tenure. As a next step I investigate whether the inverse relationship between firm-specific training and displacement is changed in the case of displacement by position abolition. The multinomial logit regression results, estimating the effects of firm-specific training on the probability of displacement, are reported in table 4. The coefficients reported in the first two columns of panel A indicate that both types of firm-specific training lower the probability of displacement through a plant closing or slack work.

Company training significantly reduces the probability of

displacement by plant closing and slack work. Because informal on-the-job training is not as firm specific as company training, the coefficients on informal on-the-job training are insignificant, 17

For example, in the sample used in this paper 68% of the displaced workers have less than 5 years of tenure and the average tenure of displaced workers is 5.1 years.

14

though still negative. In the column for displacement by position abolition, the coefficients on both types of firm-specific training are positive, and the coefficient on company training is even significant at the 10 percent level. Given that position abolition is the displacement category in which older workers face a higher risk of displacement and technological changes increase that risk, the positive coefficients on both types of firm-specific training support the notion that firm-specific human capital decreases the job security of older workers in conditions of technological change. If wages were costlessly renegotiable the firm could retain these older workers at lower wages, when the value of their marginal product falls under technological change. Therefore the fact that these workers are displaced strongly implies that high transaction costs have barred renegotiations. When displaced, these workers have steeper wage-tenure profiles and higher wages that were set as a result of the sharing arrangement of the costs of and returns to investment in firmspecific human capital.

Comparison of Earnings between the Displaced and the Nondisplaced In order to make possible a comparison of the predisplacement wage-tenure profiles and wage levels of workers displaced due to position abolition with those of other workers, I have used subsets of the whole sample used in the previous section. The nondisplaced workers have been drawn from the outgoing rotation groups of the DWS. Once observations lacking reported earnings and union status have been dropped, the sample has 29,025 workers, including 6,426 displaced workers. Because the available predisplacement earnings variable in the DWS represents weekly earnings without the number of hours worked, I have estimated two separate earnings regressions: one for the whole sample including some part-time workers, and another for full-time workers only. The size of the full-time-worker sample is 24,569, including 5,552 displaced workers. Weekly earnings, the dependent variable, are expressed in logs after being deflated to 1996 dollars

15

using the GDP deflator. Table 5 shows the coefficients on the causes of displacement, and their interaction terms with tenure and tenure square in regressions. Included among the independent variables are tenure and its square, age and its square, a marriage dummy, three education dummies, a nonwhite dummy, a female dummy, three region dummies, a SMSA dummy, eight industry dummies, eight occupation dummies, a union dummy, and year dummies. Columns 1 and 2 also include a full-time dummy. The causes of displacement are included as dummy variables in columns 1 and 3. The coefficients on the position abolition dummy variable, in columns 1 and 3, confirm that workers displaced by position abolition earned significantly more in their predisplacement jobs: approximately 8 percent more than comparable nondisplaced workers, even after controlling for other worker characteristics. The coefficients on other causes of displacement are negative, though insignificant. In contrast to the results obtained by de la Rica (1992) and Stevens (1997), the earnings of workers displaced due to plant closures are found to be insignificantly lower than those of nondisplaced workers. Given that de la Rica (1992) used the 1986 DWS, and Stevens (1997) used the 1968-1988 waves of the PSID, this result may indicate that wage reductions prior to displacement for those displaced by plant closures disappeared in the 1990s. Columns 2 and 4 of table 5 include, in addition to the causes of displacement specified as dummy variables, the interaction terms of causes of displacement and predisplacement tenure and its square. This has been done to see if workers displaced by position abolition have steeper predisplacement wage-tenure profiles as well. The positive and significant coefficients on the interaction terms of tenure and position abolition, in both samples, reveal that workers displaced by position abolition also have steeper wage-tenure profiles than do nondisplaced workers. In sum, the results in table 5 support the notion that workers displaced by position abolition have higher wages and steeper wage-tenure profiles. This is the case because workers who have steeper wagetenure profiles and higher wages derived from their investment in firm-specific human capital are laid off as their wages become too high when technological changes depreciate the stock of firm-

16

specific human capital and decrease the value of the marginal product. One possible interpretation also consistent with the result that workers displaced by position abolition have higher wages and steeper wage-tenure profiles, involves the layoffs of senior workers after hostile takeovers or under sectoral decline. Shleifer and Summers (1988) argued that hostile takeovers can entail the abrogation of long-term implicit contracts with employees, leading to layoffs of workers who have extramarginal wage payments. Gokhale, Groshen and Neumark (1995) have empirically shown that hostile takeovers reduce extramarginal wage payments to more tenured workers, by reducing the employment of more senior workers and by flattening wage-tenure profiles. What is missing, however, from this interpretation of the use of layoffs following hostile takeovers to reduce extramarginal wage payments, is any explanation of the changing effects of firm-specific training under technological change. Idson and Valletta (1996) examined the possibility of increased layoffs of senior workers under technological change, as they sought to explain the decreasing retention rate of senior workers under sectoral decline. Ultimately they rejected the effect of technological change on the value of firm-specific investments, in favor of firm default on delayed payment contracts. These interpretations both imply layoffs of senior workers with more tenure, but not necessarily layoffs of older workers. My empirical results in table 2 show that it is older workers, not senior workers with longer tenure, who face a higher probability of displacement through position abolition. Those results certainly are more consistent with my interpretation that when technological changes depreciate the stock of firm-specific human capital, older workers are laid off since firms find it unprofitable to retrain them. Another interpretation consistent with the higher wages of workers displaced by position abolition is the wage premium associated with computer use (Krueger 1993; Dinardo and Pischke 1997). Although it still remains a matter of controversy whether computer use itself increases the productivity, or whether computer use by individual workers is associated with individual heterogeneity which increases productivity, it is unquestionably true that those who use computers

17

earn higher wages. Since workers displaced by position abolition are displaced from industries that have a higher proportion of workers using computers, their higher wages may be due to the premium associated with computer use. This interpretation seems rather supplementary to firmspecific human capital theory, however, since the use of computers often requires training. If the computer training is firm-specific, then this is exactly what firm-specific human capital theory explains. If on the other hand the computer training is general, there is no reason why those displaced workers should incur larger earnings losses, as is shown in the ensuing section. Finally, because no wage premium was observed among the workers displaced by other reasons, the higher wages of workers displaced by position abolition do not seem to be compensating wage differentials observed in the jobs that have higher anticipated probabilities of unemployment (Abowd and Ashenfelter 1981). Thus in conclusion, the explanation provided by human capital theory is the only one capable of consistently explaining why workers displaced by position abolition have steeper wagetenure profiles and higher wages. Such workers are displaced because their wages become too high as their stock of firm-specific human capital depreciates in a time of technological change.

Comparison of Earnings Changes among Displaced Workers The large earnings losses experienced by displaced workers are often ascribed to the loss of firmspecific capital (Addison and Portugal 1989; Topel 1990; Kletzer 1996). Since workers displaced by position abolition have higher wages derived from their investment in firm-specific human capital, they have more to lose from the moment of displacement, as opposed to other displaced workers. Hence, earnings losses will be greater for workers displaced by position abolition. In order to empirically test this interpretation, I have compared the earnings changes of displaced workers, broken down by reason of displacement. It is for this purpose that the sample includes those displaced workers who as of the survey date had been reemployed in wage and salary employment. I also have estimated two separate earnings regressions: one for the whole

18

sample (4,718 workers), and another for those displaced from full-time jobs only (4,136 workers). Table 6 shows the coefficients on the position-abolition dummy variable in the regressions of earnings changes. The dependent variable is the log of the difference in current weekly earnings and predisplacement weekly earnings, both deflated to 1996 dollars using the GDP deflators. The reference group consists of workers who have been displaced by plant closings or slack work. In addition to the independent variables used in the earnings-level regressions of the previous section, a written-notice dummy and years since displacement are included to control for the differences in job-search behaviors. The negative and significant coefficients on position-abolition dummy, in both samples, show that workers displaced by position abolition incur larger earnings losses than do other displaced workers. 18 This result also is what Farber (1997) found, and supports the notion that workers displaced by position abolition are losing more because they have higher wages and more firm-specific human capital in their predisplacement jobs than do other displaced workers. The larger earnings losses of workers displaced by position abolition, as compared with those of other displaced workers, also seem to be consistent with the signaling effects of layoffs, as Gibbons and Katz (1991) have argued. They developed an asymmetric information model in which a layoff event reveals the low ability of a worker to the market when firms have discretion over whom to lay off. The model predicts that as a result, while the predisplacement wages of observationally equivalent workers do not differ by cause of displacement, workers displaced by layoff, which includes position abolition and slack work, will have lower postdisplacement wages and thus incur larger wage losses than those displaced by a plant closing. Using data drawn from the 1984 and 1986 DWS, Gibbons and Katz were able to present empirical results consistent with the predictions of their model. 18

The fact that some of the displaced workers are excluded from the sample because they are not reemployed as of the survey date could produce sample selection bias in the estimates. Thus I have reestimated the models presented in table 6 using the Heckman two-stage sample-selection bias-correction method. Due to lack of a variable in the DWS that affects the probability of reemployment but does not affect earnings, the selection correction models are identified only by the nonlinearity of the sample-selection correction term. The estimation results, though not reported here, are almost identical to those shown in table 6.

19

Thus the larger wage losses of workers displaced by position abolition in my sample seem to be consistent with the predictions of Gibbons and Katz’s model.

And yet the higher

predisplacement wages of workers displaced by position abolition, the results shown in this paper, are not consistent with the predictions of their model that predisplacement wages will not differ by cause of displacement. Therefore, in order to investigate whether the larger earnings losses of workers displaced by position abolition still are due to their lower postdisplacement earnings, as predicted by the model, I estimated another set of regressions using postdisplacement wages as the dependent variable. 19 The reference group here is workers displaced by plant closings. Separate dummy variables for position abolition and slack work are included in the estimated equations. Columns 1 and 2 of table 7 present the coefficients on the position-abolition and slack-work dummy variables in the estimations, for the whole sample and for the full-time-worker sample, respectively.

There is no evidence that workers displaced by position abolition have lower

postdisplacement earnings than do those displaced by plant closings.

If anything, workers

displaced by position abolition or slack work have larger postdisplacement earnings than do those displaced by plant closings, though the difference is not significant. Therefore, workers displaced by position abolition incurred larger earnings losses not because of the signaling effect, as argued by Gibbons and Katz, but rather owing to the loss of higher predisplacement earnings. 20

Conclusions In this paper I have explained why many older workers were displaced by technological changes in the 1990s, and why workers displaced by technological changes incurred larger earning losses than did other displaced workers. When technological changes depreciate the stock of existing firmspecific human capital, the investment in firm-specific human capital no longer provides job 19

Unlike the 1984 and 1986 DWS used by Gibbons and Katz, an individual-level union-status variable is available in the 1996, 1998, and 2000 DWS, and is included in my estimation. 20 Song (2007) also showed that the lemons effect of layoffs found by Gibbons and Katz partly stems from recall bias in the 1984 and 1986 DWS.

20

security to older workers. Older workers receiving higher returns from the investment in firmspecific human capital are laid off, as firms find it unprofitable to retain them in a time of technological change. These workers have higher predisplacement earnings derived from their investment in firm-specific human capital, and thus incur larger earnings losses than do other displaced workers. The empirical results, using data drawn from the CPS, are consistent with this explanation. The empirical result that workers displaced by position abolition have higher predisplacement wages provides us with yet another evidence of the wage rigidity derived from high transaction costs in the sharing arrangement of the costs of and returns to investment in firmspecific human capital. The larger earning losses of workers displaced by position abolition, losses due to the loss of their higher predisplacement wages but not due to lower postdisplacement wages, also provide us with empirical evidence refuting the stigma effects of layoffs. To a great extent, it was technological change that brought on the unprecedented expansionary economy of the 1990s.

It was the same technological change, however, that

prompted the mass layoffs of older workers. Since technological change boosts the economy’s level of productivity, everyone would benefit if a way could just be found to compensate workers displaced due to technological changes. The search for a proper mechanism of compensation will doubtlessly be at the core of much future research.

21

Acknowledgements I would like to thank Todd Idson, Brendan O’Flaherty, Steve Cameron, Lena Edlund, Nachum Sicherman and seminar participants at Columbia University, University of Pennsylvania, IZA, and Union College for their valuable comments.

22

References Aaronson D, Housinger K (1999) The impact of technology on displacement and reemployment. Fed Reserve Bank Chicago Econ Perspect 23:14-30 Abowd JM, Ashenfelter O (1981) Anticipated unemployment, temporary layoffs, and compensating wage differentials. In: Rosen S (ed) Studies in labor markets. University of Chicago Press, Chicago Addison JT, Portugal P (1989) On the costs of worker displacement: the case of dissipated firmspecific training investments. Southern Econ J 56:166-182 Allen SG (2001) Technology and the wage structure. J Labor Econ 19:440-483 Altonji JG, Shakotko RA (1987) Do wages rise with job seniority? Rev Econ Stud 54:437-459 Antel JJ (1985) Costly employment contract renegotiation and the labor mobility of young men. Am Econ Rev 75:976-991 Autor D, Katz L, Krueger A (1998) Computing inequality: have computers changed the labor market? Q J Econ 113:1169-1214 Bartel A, Sicherman N (1993) Technological change and retirement decisions of older workers. J Labor Econ 11:162-183 Becker GS (1962) Investment in human capital: a theoretical analysis. J Polit Econ 70:9-49 Campbell CM, Kamlani KS (1997) The reasons for wage rigidity: evidence from a survey of firms. Q J Econ 112:759-789 Cappelli P (2000) Examining the incidence of downsizing and its effect on establishment performance. NBER Working Paper 7742 de la Rica S (1992) Displaced workers in mass layoffs: pre-displacement earnings losses and the union effect. Working Paper 303, Princeton University, Industrial Relations Section, Princeton, NJ DiNardo JE, Pischke J (1997) The returns to computer use revisited: have pencils changed the wage structure too? Q J Econ 112:291-303 Fallick BC (1996) A review of the recent empirical literature on displaced workers. Ind Labor Relat Rev 50:5-16 Farber HS (1993) The incidence and costs of job loss: 1982-91. Brook Pap Econ Act: Microecon 1993:73-119 Farber HS (1997) The changing face of job loss in the United States: 1981-1995. Brook Pap Econ Act: Microecon 1997:55-128 Gibbons R, Katz L (1991) Layoffs and lemons. J Labor Econ 9:351-380

23

Gokhale J, Groshen EL, Neumark D (1995) Do hostile takeovers reduce extramarginal wage payments? Rev Econ Stat 77:470-485 Hashimoto M (1981) Firm-specific human capital as a shared investment. Am Econ Rev 71:475482 Idson TL, Valletta RG (1996) Seniority, sectoral decline, and employee retention: an analysis of layoff unemployment spells. J Labor Econ 14:654-676 Jorgenson DW, Ho MS, Stiroh KJ (2008) A retrospective look at the U.S. productivity growth resurgence. J Econ Perspect 22:3-24 Kletzer LG (1996) The role of sector-specific skills in post-displacement earnings. Ind Relat 35:473-490 Kletzer LG (1998) Job displacement. J Econ Perspect 12:115-136 Krueger AB (1993) How computers have changed the wage structure: evidence from microdata, 1984-1989. Q J Econ 108:33-60 Lazear EP (1979) Why is there mandatory retirement? J Polit Econ 87:1261-1284 Loewenstein MA, Spletzer JR (1999) General and specific training: evidence and implications. J Hum Resour 34:710-733 Lynch LM (1991) The role of off-the-job vs. on-the-job training for the mobility of women workers. Am Econ Rev 81:151-156 Lynch LM (1992) Private-sector training and the earnings of young workers. Am Econ Rev 82:299-312 Mincer J (1988) Job training, wage growth, and labor turnover. NBER Working Paper 2690 Mincer J, Higuchi Y (1988) Wage structures and labor turnover in the United States and Japan. J Jpn Inter Econ 2:97-133 Mincer J, Jovanovic B (1981) Labor mobility and wages. In: Rosen S (ed) Studies in labor markets. University of Chicago Press, Chicago Parsons DO (1972) Specific human capital: an application to quit rates and layoff rates. J Polit Econ 80:1120-1143 Rodriguez D, Zavodny M (2000) Explaining changes in the age distribution of displaced workers. Fed Reserve Bank Atlanta Working Paper 2000-1 Salop J, Salop S (1976) Self-selection and turnover in the labor market. Q J Econ 90:619-627 Shleifer A, Summers LH (1988) Breach of trust in hostile takeovers. In: Auerbach AJ (ed) Corporate takeovers: causes and consequences. University of Chicago Press, Chicago Song Y (2007) Recall bias in the displaced workers survey: are layoffs really lemons? Labour Econ 14:335-345

24

Stevens AH (1997) Persistent effects of job displacement: the importance of multiple job losses. J Labor Econ 15:165-188 Topel R (1990) Specific capital and unemployment: measuring the costs and consequences of job loss. Carnegie-Rochester Conf Ser Public Policy 33:181-214 U.S. Department of Labor (1985) How workers get their training. Bureau of Labor Statistics Bulletin 2226. Government Printing Office, Washington, DC U.S. Department of Labor (1992). How workers get their training: a 1991 update. Bureau of Labor Statistics Bulletin 2407. Government Printing Office, Washington, DC

25

Table 1 Descriptive Statistics of the Whole Sample

Variable

Nondisplaced Workers

Age Tenure in years Female White-collar Full-time Less than high school High school Some college More than college Computer use Ratio of scientists & engineers Company training Informal on-the-job training DWS 1996 DWS 1998 DWS 2000 Percentage among displaced workers

38.29 6.68 0.50 0.61 0.85 0.09 0.33 0.31 0.26 0.51 0.031 0.17 0.16 0.31 0.34 0.35

Number of observations

89,338

Displaced Workers Plant Closure

Slack Work

Position Abolition

37.73 5.86 0.52 0.58 0.88 0.11 0.36 0.32 0.20 0.48 0.030 0.16 0.15 0.33 0.33 0.33

35.60 3.06 0.45 0.47 0.81 0.15 0.36 0.31 0.18 0.42 0.037 0.14 0.15 0.41 0.31 0.28

38.91 6.09 0.52 0.75 0.90 0.05 0.26 0.33 0.37 0.61 0.042 0.20 0.18 0.37 0.34 0.29

0.42

0.30

0.28

3,072

2,242

2,086

NOTE: All observations have been drawn from the 1996, 1998, and 2000 DWS. Displaced workers are workers who were between the ages of 20 and 61 when permanently displaced from a private-sector job, excluding the agriculture and construction industries, within three years prior to the survey date because of a plant closing, slack work, or a position or shift being abolished. Nondisplaced workers are workers between the ages of 20 and 61, employed as of the survey date in private-sector jobs, excluding the agriculture and construction industries. Computer-use data are the author’s calculations based on the 1993 October CPS. Ratio of scientists and engineers data are the author’s calculations based on the 1993 CPS outgoing rotation groups. Training data are the author’s calculations based on the 1991 January CPS.

26

Table 2 Probability of Displacement by Cause of Displacement (Multinomial Logit Estimation) Panel A: Coefficients (Excluded Category is Nondisplaced) Variable Age 30 – 39 Age 40 – 49 Age 50 – 61 Tenure 2-4 Tenure 5-9 Tenure 10-19 Tenure more than 20 Female Full-time High school Some college More than college

Plant Closing

Slack Work

Position Abolition

.072 (.051) .084 (.054) .043 (.063) .266** (.047) -.184** (.057) -.342** (.064) -.277** (.084) .278** (.042) .335** (.060) -.081 (.064) -.033 (.067) -.189** (.077)

.035 (.055) .087 (.060) -.029 (.076) -.321** (.051) -1.17** (.074) -.1.61** (.095) -1.99** (.154) -.090* (.050) -.145** (.060) -.164** (.069) -.194** (.073) -.383** (.088)

.179** (.064) .266** (.067) .315** (.076) .060 (.058) -.311** (.067) -.462** (.074) -.441** (.099) .083* (.050) .496** (.078) .266** (.111) .508** (.112) .668** (.117)

-32,802.39 96,738

Log likelihood Number of observations

Panel B: Marginal Effects at the Sample Mean Variable Age 30 – 39 Age 40 – 49 Age 50 – 61 Tenure 2-4 Tenure 5-9 Tenure 10-19 Tenure more than 20 Female Full-time High school Some college More than college

Nondisplaced

Plant Closing

Slack Work

Position Abolition

-.006 -.009 -.007 -.004 .024 .032 .030 -.007 -.013 -.000 -.006 -.004

.002 .002 .001 .008 -.004 -.008 -.006 .008 .008 -.002 -.001 -.005

.000 .001 -.001 -.005 -.014 -.017 -.017 -.002 -.003 -.003 -.003 -.006

.003 .005 .007 .001 -.005 -.007 -.007 .001 .008 .005 .010 .015

NOTE: The reported regression also includes the following independent variables: a nonwhite dummy, three region dummies, a SMSA dummy, eight industry dummies, eight occupation dummies, and two year-ofsurvey dummies. Standard errors are in parentheses. * Statistically significant at the .10 level. ** Statistically significant at the .05 level.

27

Table 3 Effects of Technological Changes on the Probability of Displacement, by Cause of Displacement (Multinomial Logit Estimation) Panel A: Coefficients (Excluded Category is Nondisplaced)

Variable Computer use Ratio of scientists & engineers

Plant Closing

Slack Work

Position Abolition

-.354** (.135) .138 (.379)

-.225 (.158) 1.826** (.361)

.676** (.167) 1.057** (.354)

96,738

Number of observations

Panel B: Marginal Effects at the Sample Mean Variable

Nondisplaced

Plant Closing

Slack Work

Position Abolition

.001 -.051

-.010 .002

-.004 .029

.013 .019

Computer use Ratio of scientists & engineers

NOTE: Each row represents results from a separate regression. The reported regressions also include the following independent variables: three age dummies, a female dummy, a full-time dummy, three education dummies, four tenure dummies, a nonwhite dummy, three region dummies, a SMSA dummy, eight industry dummies, eight occupation dummies, school training, other training, and two year-of-survey dummies. Standard errors are in parentheses. * Statistically significant at the .10 level. ** Statistically significant at the .05 level.

28

Table 4 Effects of Firm-Specific Training on the Probability of Displacement, by Cause of Displacement (Multinomial Logit Estimation) Panel A: Coefficients (Excluded Category is Nondisplaced) Variable Company training Informal on-the-job Training

Plant Closing

Slack Work

Position Abolition

-.699** (.317) -.399 (.350)

-.566* (.310) -.330 (.406)

.513* (.287) .323 (.418)

-32,776.27 96,738

Log likelihood Number of observations

Panel B: Marginal Effects at the Sample Mean Variable Company training Informal on-the-job Training

Nondisplaced

Plant Closing

Slack Work

Position Abolition

.018 .010

-.019 -.011

-.009 -.005

.010 .006

NOTE: The reported regression also includes the following independent variables: three age dummies, a female dummy, a full-time dummy, three education dummies, four tenure dummies, a nonwhite dummy, three region dummies, a SMSA dummy, eight industry dummies, eight occupation dummies, school training, other training, and two year-of-survey dummies. Standard errors are in parentheses. * Statistically significant at the .10 level. ** Statistically significant at the .05 level.

29

Table 5 Coefficients on the Cause-of-Displacement Dummy Variables in Earnings Regressions Dependent variable: log (weekly earnings). Whole Sample Variable Position abolition Slack work Plant closure

(1)

(2)

(3)

(4)

.081** (.018) -.006 (.019) -.007 (.017)

.032 (.025) -.017 (.023) .005 (.022) .014** (.005) .001 (.007) -.007* (.004) -.0004** (.0002) -.0002 (.0003) .0003** (.0002)

.076** (.018) -.022 (.019) -.005 (.017)

.021 (.024) -.048** (.024) -.002 (.022) .014** (.005) .006 (.006) -.005 (.004) -.0004** (.0002) -.0001 (.0003) .0003* (.0001)

.5412

.4238

Tenure × Position abolition Tenure × Slack work Tenure × Plant closure Tenure square × Position abolition Tenure square × Slack work Tenure square × Plant closure Adjusted R2 Number of observations

Full-time Workers

.5410 29,025

.4243 24,569

NOTE: The reported regressions also include tenure and its square, age and its square, a marriage dummy, a female dummy, three education dummies, a nonwhite dummy, three region dummies, a SMSA dummy, eight industry dummies, eight occupation dummies, a union dummy, and year dummies. Columns 1 and 2 also include a full-time dummy. Standard errors are in parentheses. Weekly earnings are in 1996 current dollars, deflated by the GDP deflator. * Statistically significant at the .10 level. ** Statistically significant at the .05 level.

30

Table 6 Coefficients on the Position Abolition Dummy Variable in Earnings Change Regressions (Displaced Workers Reemployed as of Survey Date) Dependent variable: log (current weekly earnings/previous weekly earnings).

Variable Position abolition Adjusted R2 Number of observations

Whole Sample (1)

Full-time Workers (2)

-.062** (.023)

-.048** (.023)

.0789 4,718

.0474 4,136

NOTE: The reported regressions also include previous tenure and its square, age and its square, a marriage dummy, a female dummy, a written-notice dummy, three education dummies, a nonwhite dummy, three region dummies, a SMSA dummy, eight previous-industry dummies, eight previous-occupation dummies, a previous-union dummy, years since displacement and year dummies. Column 1 also includes a full-time dummy. Standard errors are in parentheses. Weekly earnings are in 1996 current dollars, deflated by the GDP deflator. * Statistically significant at the .10 level. ** Statistically significant at the .05 level.

Table 7 Coefficients on the Position Abolition and Slack Work Dummy Variables in Postdisplacement Earnings Regressions (Displaced Workers Reemployed as of Survey Date) Dependent variable: log (current weekly earnings).

Position abolition Slack work Adjusted R2 Number of observations

Whole Sample (1)

Full-time Workers (2)

.021 (.024) .006 (.025)

.040 (.025) .010 (.026)

.2865 4,718

.2594 4,136

NOTE: The reported regressions also include previous tenure and its square, age and its square, a marriage dummy, a female dummy, a written-notice dummy, three education dummies, a nonwhite dummy, three region dummies, a SMSA dummy, eight previous-industry dummies, eight previous-occupation dummies, a previous-union dummy, years since displacement and year dummies. Column 1 also includes a full-time dummy. Standard errors are in parentheses. Weekly earnings are in 1996 current dollars, deflated by the GDP deflator. * Statistically significant at the .10 level. ** Statistically significant at the .05 level.

31