Volume 29, Issue 2. How do firms interpret a job loss? Evidence from the National Longitudinal Survey of Youth

Size: px
Start display at page:

Download "Volume 29, Issue 2. How do firms interpret a job loss? Evidence from the National Longitudinal Survey of Youth"

Transcription

1 Volume 29, Issue 2 How do frms nterpret a job loss? Evdence from the Natonal Longtudnal Survey of Youth Stephen M. Kosovch Stephen F. Austn State Unversty Abstract Emprcal studes n the job dsplacement lterature have found that workers face sgnfcant earnngs losses on average, when they are permanently dsplaced from jobs. Prevous research also suggests that the costlness of job loss vares wdely. Gbbons and Katz (1991) develop and test a theoretcal model n whch layoffs provde the market wth nformaton concernng the qualty of lad off workers, whle plant and frm closngs do not. Usng data from the Natonal Longtudnal Survey of Youth, ths paper tests a model that descrbes how frms can use addtonal nformaton about job losses to determne worker qualty. The results suggest that workers face the most stgma from very recent and uncommon job losses. Ctaton: Stephen M. Kosovch, (2009) ''How do frms nterpret a job loss? Evdence from the Natonal Longtudnal Survey of Youth'', Economcs Bulletn, Vol. 29 no.2 pp Submtted: Apr Publshed: May 19, 2009.

2 1.) Introducton Theoretcal and emprcal evdence suggests that the condtons under whch a worker loses a job may affect post-dsplacement wages (Gbbons and Katz, 1991). I extend the theory to model frms as usng all avalable nformaton about a worker s job loss to determne wage offers to dsplaced workers. Usng a probt and subsequently a bvarate probt model, frms are modeled as calculatng the probablty of job loss for each worker, and use these estmates to determne the lkelhood of a job loss. For example, frms may use the unemployment rate at the tme of the job loss as proxy for local labor market condtons. As the number of layoffs n the local economy ncreases, so does the expected margnal productvty of a lad-off worker, reducng the stgma assocated wth layoffs. One dstnct advantage of ths approach s that through the constructon of the bvarate probt model, I can examne the mpact of early versus later career job dsplacements. Emprcal fndngs from a sample of the Natonal Longtudnal Survey of Youth 1979 (NLSY79) provde support for the hypothess that frms use nformaton about the crcumstances of a job loss to assess the productvty of dsplaced workers. 2.) Prevous Lterature Most studes n the job dsplacement lterature focus on measurng the wage and earnng losses of dsplaced workers. Several studes have used the Dsplaced Worker Survey (DWS) to quantfy earnngs losses due to job dslocaton. The DWS s a natonally representatve sample of dsplaced workers, conducted bennally n ether January or February snce 1984 as a specal supplement to the Current Populaton Survey (CPS). Farber (1997) s an example of a typcal study usng the DWS. 1 Farber estmates the probablty of job loss by educaton, sex, race, and year dsplaced. Usng data from the seven DWS surveys from 1984 to 1996, Farber estmates that over the perod from , real weekly post-dsplacement earnngs were 13 percent lower than real weekly pre-dsplacement earnngs, controllng for tenure. Ruhm (1991) nvestgates long-term earnngs losses usng data from the Panel Study of Income Dynamcs (PSID) for a sample of 3,813 workers who were dentfed as dsplaced over the perod. The sample s restrcted to workers between the ages of 21 and 65 durng who were n the labor force for part of Excludng dsplacement from temporary and seasonal jobs, workers are defned as permanently dslocated f they were termnated because of plant closngs or layoffs and dd not return to the orgnal employer wthn the end of the second year. Ths classfcaton clearly dffers from that n the DWS. In partcular, workers fred for cause are ncluded n the sample of layoffs. Also, as compared to the DWS, the avalable sample sze s much smaller. More recently, researchers have utlzed addtonal longtudnal data sources. Usng the NLSY79, Kletzer and Farle (2003) estmate the long-term costs of job dsplacement for young adults for the perod , also usng fxed effects. Sgnfcant earnngs losses are found for ths group of workers, although the earnngs losses appeared shorter-lved for ths group of younger than average workers, as compared to the results from the PSID, whch samples workers of all ages. The authors 1 I do not provde an exhaustve revew of all studes n the dsplacement lterature. A good survey of the lterature s Kletzer (1998). 1

3 do not attempt to estmate any systematc dfferences between layoffs and plant closngs, n the NLSY79. One explanaton for dfferences n the costs of dsplacement by type of job loss s the possblty that frms use the type of job dsplacement as some ndcator of worker productvty. Ths research s motvated by the work of Akerlof (1976) and Spence (1973) n whch asymmetrc nformaton concernng worker productvty can affect outcomes n the labor market. In partcular, layoffs and plant closngs have been dentfed as havng dfferent levels of stgma. Gbbons and Katz (1991) develop a model that assumes a worker s current employer has access to nformaton concernng a worker s productvty that other potental employers do not. A hrng frm may be able to use the way n whch a worker s dscharged from the frm as an nformatve sgnal. The authors argue that workers dsplaced by plant or frm closng wll face lower dsplacement costs than other types of job losers, snce these types of separatons provde no nformaton regardng worker qualty. Conversely, frms are assumed to layoff only lower qualty workers and other frms wll consequently offer these workers lower wages. In a workng paper, Nakamura (2004) develops and tests a theory connectng layoffs and stgma to the busness cycle. In the model, workers are lad off ether because of selecton or bad luck. Durng recessons the proporton of workers lad off because of bad luck ncreases, dmnshng the sgnal assocated wth a layoff. Nakamura (2004) uses a sample from the DWS and fnds emprcal evdence suggestng that durng recessons the sgnal of worker productvty assocated wth a layoff decreases. Nakamura s work dffers from my analyss s several mportant ways. Frst, the measure of labor market condtons s the natonal unemployment rate over the busness cycle, whereas my focus s state unemployment rates. 2 Also, Nakamura (2004) utlzes only a cross-secton of workers usng the DWS; I conduct an emprcal test usng a longtudnal sample. However, the ntuton concernng the lnk between the stgma of layoffs and labor market condtons s essentally the same. When many layoffs occur, the stgma assocated wth layoffs decreases. 3.) Data Descrpton The Natonal Longtudnal Survey of Youth (NLSY79) s a natonally representatve sample of men and women between the ages of 14 and 22 n The survey contans nformaton concernng labor market experence, job hstory and moblty, demographc characterstcs, along wth many other varables of nterest. The NLSY79 was admnstered annually from 1979 to 1994 and has been conducted bennally snce then. Attrton rates have been low; of the 14,574 ndvduals selected n the full sample for ntervew n 1979, 12,686 partcpated. 3 In 1994, the total sample contaned responses from 9,964 ndvduals. Part of the reason for the low attrton rates s that sample members not partcpatng n any gven year are stll consdered members of the sample and may partcpate n later years. Attrton s not generally thought to be 2 An earler verson of ths paper used local area unemployment rates, wth smlar results. I choose to use the state unemployment rates n the year of dsplacement to avod reducng the sample sze to only workers n metropoltan areas. 3 Wenberg and Shp fnd that the retenton rate of other commonly used surveys such as the Panel Study of Income Dynamcs and the Survey of Income and Program Partcpaton are approxmately two-thrds of that of the NLSY after approxmately twelve ntervews. 2

4 random, and those leavng the survey are more lkely to be less educated and not employed than the average respondent. MaCurdy, Mroz and Grtz (1998) examned the effects of attrton n the NLSY79 and fnd no evdence of any systematc bas resultng from non-random attrton. For ths analyss, the sample s restrcted to males durng the years Male workers are agan the focus, snce they are more lkely to have some attachment to the labor force. The sample s also lmted to those workers wth some labor market experence at the begnnng of the perod, snce the NLSY79 s a sample of younger workers and many ndvduals have not yet left school. A male worker s ncluded n the sample f that ndvdual was employed durng the year 1984 at a job at whch the worker reported workng at least 20 hours a week. Subsequent nformaton about jobs durng the years for these workers s ncluded, creatng a balanced panel for the years I exclude workers that report beng enrolled n school full-tme n 1984, and workers who have no labor market experence durng These workers have less exposure to the labor market, whch wll lkely affect the probablty of a job loss over ths perod. Indvduals that report not workng at any one of the fve reported jobs durng a year ether because of a layoff or a plant closng are consdered dsplaced provded that several condtons are met. Frst, temporary job losses are not consdered dsplacements, and those workers who report returnng to ther employers after such a loss are not recorded as dsplaced. Addtonally, ndvduals must have reported workng at least 20 hours per week at the job n order for t to be consdered ether a layoff or plant closng. Real hourly wages n 1994 are used n the second-stage wage equaton, so workers must also have worked n 1994 to be ncluded n the sample. 4.) Emprcal Methods: Sngle-Perod Analyss Frms may have more nformaton regardng the condtons under whch layoffs occur than smply the unemployment rate at the tme of a job loss, and the type of dsplacement. In ths secton, I allow frms to observe each worker s hstory of job loss and also some worker and job specfc attrbutes. Job loss s modeled usng the followng lnear ndex functon: * (1) y ' X where t s assumed that y, the observed value of the job loss s: * 1 f y 0 y * 0 f y 0 Although dsplacements are recorded by year, as a frst step y s set equal to one f worker s ever observed to be dsplaced durng the perod between The ntal specfcaton treats plant closng and layoffs as dentcal, whle later specfcatons 4 Workers dsplaced before 1984 are excluded from the entrety of the sample, snce t s mpossble to determne whch type of job loss they faced. 5 Throughout the sample there are some mssng or mscoded varables for workers. I only exclude a worker f the mssng varable s used n ether the frst-stage probt estmatons, or the second stage wage equatons, or f the varable affects the constructon of each ndvdual s work hstory, snce these varables are used to calculate labor market experence. 6 I do not exclude workers who return to school between the years , but I do exclude workers enrolled n school durng 1994, the year n whch I estmate the wage equaton. 3

5 dstngush between layoffs and plant closngs. In these later emprcal specfcatons, I only nclude those lad-off workers that were never dsplaced by a plant closng, and those workers affected by a plant closng that were never lad-off. 7 The sample contans 2225 workers, of whch 795 are dsplaced at least once durng the perod. The vector X ncludes worker-specfc trats such as years of educaton, race, tenure, experence, annual area unemployment rate, geographc regon, and ndustry and occupaton specfc characterstcs. For smplcty, these varables are measured durng 1984, the begnnng of the perod. 8 It s therefore assumed that frms, n 1994, have access to these worker and job-specfc trats. Descrptve statstcs for the varables used n the estmatons are reported n Table 1. For layoffs, there s a partcularly clear nterpretaton of the ndex functon. If a worker s margnal productvty s below some threshold, the worker s lad-off; otherwse the worker s retaned. The vector X ncludes varables that affect ths threshold, and therefore change the nformaton contaned n a layoff. For plant closngs, the nterpretaton s less straghtforward, snce plant closngs have been modeled as ndependent of any ndvdual worker s productvty. However, one mght expect that workers affected by plant closngs could have lower post-dsplacement wages due to a loss of human captal or a loss of work experence. In fact, one mght expect that the those wage losses would be largest for those workers affected by plant closngs and who are predcted to be dsplaced by these closngs. I frst assume that a frm estmates the probablty of a job loss for each worker n the sample usng a probt model. Ths requres the assumpton that Prob(y =1)= and that ~N(0,1) where and j contnuous normal probablty dstrbuton functon. ' x ( t) dt ( j) are ndependent, and where (t) represents the Ths probablty of a job loss s done va maxmum lkelhood estmaton, and ths value s then used n a second stage wage regresson. 9 Specfcally, the followng equaton s estmated: (2) ln W (1994) ' Z ' D u where W(1994) s the hourly wage of a worker n 1994, Z s a vector of worker and jobspecfc trats for the worker n 1994, and D s a set of dummy varables based on both observed dsplacement and predcted dsplacement, for each worker. The vector D contans three dummes: a varable equal to one f a worker s dsplaced durng the perod and the predcted probablty s estmated above the mean of predcted probabltes, a second varable equal to one f a worker s dsplaced and has an estmated probablty below the mean, and a thrd dummy equal to one f the worker s not observed to be dsplaced but has an estmated probablty of dsplacement above the mean. A 7 Ths restrcton does elmnate some workers from the sample, snce multple types of job losses are possble for one partcular worker. However, multple layoffs are much more common that multple plant closngs. Only 87 workers are dropped from the sample because of ths restrcton. 8 Industry and occupatonal characterstcs are recorded for the job at whch the ndvdual worked the most hours at durng a year for the probt model, and for the job at whch the ndvdual worked the most durng 1994 for the second-stage wage equaton. 9 The predcted probablty for each ndvdual s smply: p ( ' x ), where ( ) represents the cumulatve normal dstrbuton. 4

6 fourth dummy, equal to one f the worker s not dsplaced and has a predcted probablty below the mean, s excluded. The vector Z ncludes all the varables n X, excludng martal status and number of chldren, although these varables are measured durng the year Results for the frst stage estmatons are presented n Table 2. Column (1) ncludes estmated coeffcents for all dsplacements, whle (2) and (3) separates layoffs and plant closngs. Job tenure, experence, and educaton at the begnnng of the perod appear to be mportant statstcally n determnng whether a worker s dsplaced durng that perod. Not surprsngly, observable factors are less successful at predctng plant closngs; ths result provdes more evdence that plant closngs are ndependent of worker attrbutes. Broad occupatonal and ndustry factors at the begnnng of the perod are also surprsngly neffectve n predctng plant closngs. Second stage wage regressons are presented n Table 3. For the full sample of all dsplacements, workers that are dsplaced durng the perod but are not predcted to be dsplaced are estmated to have statstcally sgnfcantly (wth 95 percent confdence) lower real wages at the end of the perod. Specfcally, these workers are estmated to have 7.4 percent lower real wages, as compared to non-dsplaced workers, that are not predcted to be dsplaced. For the sample of lad-off workers (and non-dsplaced), ths dfference s estmated to be approxmately 8.6 percent. No such fndngs exst for the sample of plant closng workers. Interestngly, those workers affected by plant closng and that are predcted to be dsplaced are estmated to have lower real wages than the group of non-dsplaced workers, non-predcted workers. 10 Ths result perhaps suggests that workers affected by plant closngs do have some estmated earnngs losses, but that they are the result of predctable job or ndustry attrbutes, not any nformaton contaned n a plant closng. Also, workers that are not dsplaced, but have the characterstcs of dsplaced workers, are estmated to have lower real wages than those non-dsplaced, not predcted dsplaced workers. Although ths estmate s not statstcally sgnfcant at any usual level of sgnfcance, ts sgn may be more evdence that a worker s attrbutes can affect later wages. Perhaps frms have alternatve sgnals of these workers productvtes, other than layoffs. 4.2.) Two-perod Analyss As stated earler, one advantage of ths two-stage analyss s that t s possble to examne the mpact of early versus late job losses. The sample s now dvded nto two fve year perods: and , and the dsplacements n perods are modeled as determned for perod one by: *1 1 1 (3) perod1) ' y ( X and for perod two: *2 (4) perod 2) ' y 2 2 ( X 10 The estmated predcted dfference s not statstcally sgnfcant at any normal level of sgnfcance. Also, the sample of plant-closngs s small at only 134 workers. 5

7 * t 1 f y 0 t It s further assumed that y * t for t=1,2, and where 1 and 2 are 0 f y dstrbuted bvarate normal, [0,0,1,1,: ], and where X and X are measured at the begnnng of each perod. 11 The parameter measures the correlaton between the errors of equatons (3) and (4). Ths specfcaton has several advantages over a sngle perod analyss. Frst, the effects of job losses are allowed to dffer based on whether the loss occurred durng two dfferent perods n a worker s career, whch wll allow for a comparson of early versus later job losses. 12 Second, nformaton regardng worker and job-specfc characterstcs durng two ponts n a worker s career can be used to calculate probabltes of job loss. Fnally, ths modelng approach allows correlaton between unobserved factors affectng the probabltes of dsplacement n each of the two perods. For layoffs t s lkely that unobserved worker-specfc attrbutes wll affect the probablty of a layoff n both perods, makng postve. Results of the bvarate probt specfcatons are presented n Tables 4, 5, and 6 for all dsplacements, only layoffs, and only plant closngs, respectvely. As n the sngleperod probt, job tenure, experence, and educaton are statstcally mportant n determnng the probablty of job loss n both perods for all dsplacement and only layoffs. Agan, observable factors are less successful at predctng plant closngs. As prevously mentoned, measures the correlaton n the two errors of the estmated probablty of job loss n the two perods. A Wald test s conducted to determne f s statstcally dfferent that zero. For all dsplacements, s estmated to be approxmately 0.061, and s statstcally dfferent that zero at the 10 percent level. The correlaton n errors s stronger when plant closngs are excluded, as s estmated at 0.110, and s sgnfcant at the 5 percent level. Intutvely, the unobserved attrbutes that make a worker more lkely to be lad-off n perod one also ncrease the lkelhood of a layoff n the second perod. No such correlaton s found for plant closngs, whch s more evdence suggestng that plant closngs are unaffected by worker-specfc trats. As n the sngle-perod analyss, the frst stage estmatons of the probablty of job loss are used n a second-stage wage regresson: 1' 1 2' 2 (5) ln W (1994) ' Z D D u In ths specfcaton there are two sets of three dummy varables, one set from each perod of job loss. These dummes are agan calculated usng the mean predcted probablty as a cutoff value. The vector Z s agan a set of worker and job-specfc attrbuted for each ndvdual, measured durng Results for these estmates are presented n Table 7. For the sample of layoffs only, a worker lad-off n the frst perod but not predcted to be lad-off, s estmated to 11 The bvarate normal cdf s 2 ( z1, z2, ) dz1dz 2. x2 x2 12 Ths approach cannot account for the effects of multple job losses that occur durng the same fve-year perod. However, t s unclear how much addtonal nformaton a frm may gan from observng a workers loss of a job mmedately followng a prevous job loss. 6

8 have approxmately a 9.6 percent lower real wage, as compared to a non-dsplaced worker not predcted to be dsplaced. For layoffs n the second perod, ths percentage dfference s estmated to be 11.3 percent. 13 Ths result provdes addtonal evdence that a lad-off worker wth attrbutes not consstent wth layoffs ncur a wage penalty. It also demonstrates that layoffs occurrng both early and later n a worker s career have an estmated mpact on post-dsplacement wages. Those workers that are lad-off n the frst perod but have attrbutes consstent wth layoffs also are estmated to have lower real hourly wages n As n the sngle perod analyss, the magntude of ths coeffcent s larger for plant closngs. Ths result agan suggests that perhaps there are some costs of job loss that are the result of predctable worker or job-specfc attrbutes that do not have to do wth the nformaton contaned n a layoff. 5.) Concludng Remarks Ths paper provdes evdence that frms nterpret job dsplacements usng all avalable nformaton regardng workers and ther job hstores. In partcular, I allow frms to construct an estmated probablty of job loss for each worker, based upon worker and job-specfc characterstcs. Frms then use these probabltes to adjust ther wage offers. My results suggest that real wages are lower for lad-off workers wth below average predcted probabltes of dsplacement. Frms understand that these workers were less lkely to experence a layoff, and lost ther job because of ther low margnal productvty. One dstnct advantage of ths approach s that t s easly extended to examne the ssue of multple dsplacements, through technques such as the bvarate probt model. The bvarate probt approach s partcularly attractve snce t provdes an explct measure of the relatedness of job loss over tme. Estmates from these models are consstent wth the dea that layoffs provde a stronger sgnal of worker qualty, when workers are less lkely to lose ther jobs. Ths nsght s consstent wth the stylzed fact of the job dsplacement lterature that the costlness of job loss vares wdely. My results suggest that some of ths varaton s due to the fact that frms understand that the value of a layoff as a sgnal of worker productvty depends upon the crcumstances surroundng the job loss. 13 The frst estmated coeffcent s statstcally sgnfcant at the 5 percent level, the second at the 10 percent level. 7

9 TABLE 1: Descrptve Statstcs (Means) for Selected Varables (1984, 1994) All Non-Dsplaced Workers All Dsplaced Lad-off Workers Plant Closng Workers 1984 Varables Unemployment rate (annual average, 1984) (3.13) (3.31) (3.37) (3.15) Race (nonwhte=1) (0.44) (0.47) (0.46) (0.47) Hghest Grade Completed (1984) Labor Market Experence (weeks, 1984) (1.94) (1.79) (1.73) (2.14) (105.73) (100.19) (97.64) (104.86) Job Tenure (weeks, 1984) (96.26) (85.22) (83.09) (95.21) Age (years, 1984) (2.25) (2.23) (2.26) (2.12) Marred (yes=1, 1984) (0.45) (0.42) (2.27) (0.42) Chldren (yes=1, 1984) Varables (0.45) (0.45) (0.45) (0.42) Hghest Grade Completed (1989) Labor Market Experence (weeks, 1989) (2.38) (1.98) (1.94) (2.23) (157.23) (179.38) (166.50) (106.53) Job Tenure (weeks, 1989) (169.60) (112.80) (113.21) (124.53) Marred (yes=1, 1989) (0.49) (0.50) (0.49) (0.49) Chldren (yes=1, 1989) (0.50) (0.50) (0.50) (0.50) 8

10 1994 Varables Real Hourly Wage (1994, n 1992 dollars) Labor Market Experence (weeks, 1994) (11.10) (10.49) (6.92) (15.88) (183.86) (152.65) (151.73) (141.34) Job Tenure (weeks, 1994) (243.74) (157.43) (160.73) (152.10) Number of Observatons Standard devatons are n parentheses. All means are for those workers n the sample, as descrbed n the text of ths paper. 9

11 TABLE 2: Probt Estmates, for Selected Sample of Male Workers ( ) (1) (2) (3) Varables (Dependent Varable=1 f Dsplacement) All Dsplaced Layoffs Plant close Urate (state unemployment rate, 1984) (1.30) (1.46) (0.44) Race (nonwhte=1) (1.46) (0.49) (1.37) Hghest grade Completed (as of 1984) (5.84)*** (5.61)*** (2.31)** Marred (as of 1984) (2.99)*** (2.98)*** (0.06) Any Chldren (as of 1984) (2.83)*** (2.75)*** (0.21) Tenure wth Employer (as of 1984) (4.30)*** (4.50)*** (1.10) Tenure Squared (2.45)** (2.66)*** (0.79) Experence (as of 1984) (1.74)* (2.34)** (0.30) Experence (as of 1984) (2.00)** (2.70)*** (0.39) Occupatonal dummy, clercal (as of 1984) (0.00) (0.23) (0.68) Occupatonal dummy, laborer (as of 1984) (2.79)*** (3.04)*** (0.49) Occupatonal dummy, servce (as of 1984) (0.84) (1.01) (0.59) Industry dummy, agg. and mn. (as of 1984) (3.35)*** (3.72)*** (0.16) Industry dummy, const. (as of 1984) (4.36)*** (4.93)*** (1.31) Industry dummy, other manuf. (as of 1984) (3.52)*** (3.12)*** (2.04)** Industry dummy, other servce (as of 1984) (0.99) (0.84) (0.23) South Regon (as of 1984) (0.85) (0.68) (0.76) West Regon (as of 1984) (1.29) (0.89) (0.68) North Central Regon (as of 1984) (0.39) (0.39) (0.11) Constant (1.97)** (1.06) (1.42) Number of Observatons Absolute value of robust clustered t statstcs are n parentheses, * sgnfcant at 10 % level, ** sgnfcant at 5% level, *** sgnfcant at the 1% level. 10

12 TABLE 3: Hourly Wage Regressons, for Male Workers (1994), Usng Frst-Stage Probt (1) (2) (3) All Dsplaced Layoffs Plant close Varables (Dependent Varable s ln(hourly wage) ln(wage) ln(wage) ln(wage) Observed Dsplaced, Predcted Dsplaced (1.57) (0.96) (1.60) Observed Dsplaced, Not Predcted Dsplaced (2.19)** (2.24)** (1.10) Not Observed Dsplaced, Predcted Dsplaced (1.01) (1.14) (1.02) Urate (state unemployment rate, 1984) (2.07)** (1.91) (1.62) Race (nonwhte=1) (3.86)*** (3.75)*** (1.63) Hghest grade Completed (as of 1994) (9.63)*** (9.30)*** (8.39)*** Tenure wth Employer (as of 1994) (7.18)*** (6.58)*** (5.27)*** Tenure Squared (5.61)*** (5.25)*** (4.26)*** Experence (as of 1994) (3.77)*** (3.85)*** (3.18)*** Experence (as of 1994) (2.19)** (2.20)** (1.92) Occupatonal dummy, clercal (as of 1994) (4.10)*** (3.90)*** (3.11)*** Occupatonal dummy, laborer (as of 1994) (6.01)*** (5.47)*** (5.06)*** Occupatonal dummy, servce (as of 1994) (7.10)*** (6.53)*** (6.51)*** Industry dummy, agg. and mn. (as of 1994) (1.96) (2.38)** (1.28) Industry dummy, const. (as of 1994) (4.09)*** (3.64)*** (1.20) Industry dummy, other manuf. (as of 1994) (0.40) (0.50) (0.38) Industry dummy, other servce (as of 1994) (2.59)*** (2.24)** (1.01) South Regon (as of 1994) (0.78) (0.98) (0.61) West Regon (as of 1994) (4.69)*** (4.28)*** (3.74)*** North Central Regon (as of 1994) (7.19)*** (6.82)*** (6.11)*** Constant (17.15)*** (16.17)*** (12.98)*** Number of Observatons Absolute value of robust clustered t statstcs are n parentheses, * sgnfcant at 10 % level, ** sgnfcant at 5% level, *** sgnfcant at the 1% level. 11

13 TABLE 4: Bvarate Probt Estmates, All Dsplacements, Male Workers ( ) (1) (2) Varable (Depended Varable =1 f Dsplacement) Perod 1 ( ) Perod 2 ( ) Urate (state unemployment rate, 1984) (0.21) (0.90) Race (nonwhte=1) (0.00) (1.32) Hghest grade Completed (start of perod) (4.66)*** (2.61)*** Marred (start of perod) (3.33)*** (1.29) Any Chldren (start of perod) (3.87)*** (1.03) Tenure wth Employer (start of perod) (4.19)*** (4.36)*** Tenure Squared (start of perod) (1.91)* (2.25)** Experence (start of perod) (2.20)** (0.00) Experence (start of perod) (2.05)** (0.46) Occupatonal dummy, clercal (start of perod) (0.39) (0.82) Occupatonal dummy, laborer (start of perod) (3.26)*** (2.93)*** Occupatonal dummy, servce (start of perod) (1.33) (1.24) Industry dummy, agg. and mn. (start of perod) (3.34)*** (1.65)* Industry dummy, const. (start of perod) (4.48)*** (3.85)*** Industry dummy, other manuf. (start of perod) (2.93)*** (1.16) Industry dummy, other servce (start of perod) (2.10)** (0.37) South Regon (start of perod) (0.33) (0.36) West Regon (start of perod) (1.57) (1.47) North Central Regon (start of perod) (0.56) (2.03)** Constant (0.09) (1.15) Rho Wald test of Rho=0 (Ch squared [1]) * Number of Observatons Absolute value of robust clustered t statstcs are n parentheses, * sgnfcant at 10 % level, ** sgnfcant at 5% level, *** sgnfcant at the 1% level. 12

14 TABLE 5: Bvarate Probt Estmates, Only Layoffs, Male Workers ( ) (1) (2) Varable (Depended Varable =1 f Dsplacement) Perod 1 ( ) Perod 2 ( ) Urate (state unemployment rate, 1984) (0.68) (0.90) Race (nonwhte=1) (0.88) (0.85) Hghest grade Completed (start of perod) (4.49)*** (2.76)*** Marred (start of perod) (3.14)*** (2.07)** Any Chldren (start of perod) (3.88)*** (0.73) Tenure wth Employer (start of perod) (4.19)*** (3.32)*** Tenure Squared (start of perod) (2.37)** (1.36) Experence (start of perod) (2.50)** (0.63) Experence (start of perod) (2.63)*** (1.19) Occupatonal dummy, clercal (start of perod) (0.16) (0.77) Occupatonal dummy, laborer (start of perod) (3.11)*** (3.12)*** Occupatonal dummy, servce (start of perod) (1.43) (0.61) Industry dummy, agg. and mn. (start of perod) (3.33)*** (1.13) Industry dummy, const. (start of perod) (4.47)*** (4.08)*** Industry dummy, other manuf. (start of perod) (2.25)** (0.80) Industry dummy, other servce (start of perod) (1.60) (0.05) South Regon (start of perod) (1.10) (0.59) West Regon (start of perod) (2.19)** (1.24) North Central Regon (start of perod) (1.13) (1.89)* Constant (0.89) (0.77) Rho Wald test of Rho=0 (Ch squared [1]) 4.08** Number of Observatons Absolute value of robust clustered t statstcs are n parentheses, * sgnfcant at 10 % level, ** sgnfcant at 5% level, *** sgnfcant at the 1% level. 13

15 TABLE 6: Bvarate Probt Estmates, Only Plant Closngs, for Selected Sample of Male Workers ( ) (1) (2) Varable (Depended Varable =1 f Dsplacement) Perod 1 ( ) Perod 2 ( ) Urate (state unemployment rate, 1984) (0.91) (0.69) Race (nonwhte=1) (0.75) (1.74)* Hghest grade Completed (start of perod) (1.98)** (0.70) Marred (start of perod) (0.13) (2.57)** Any Chldren (start of perod) (0.50) (0.82) Tenure wth Employer (start of perod) (1.57) (3.16)*** Tenure Squared (start of perod) (0.48) (2.31)** Experence (start of perod) (1.29) (0.65) Experence (start of perod) (1.05) (0.73) Occupatonal dummy, clercal (start of perod) (0.13) (0.60) Occupatonal dummy, laborer (start of perod) (0.11) (0.22) Occupatonal dummy, servce (start of perod) (0.37) (0.28) Industry dummy, agg. and mn. (start of perod) (1.24) (0.35) Industry dummy, const. (start of perod) (0.15) (0.36) Industry dummy, other manuf. (start of perod) (2.66)*** (1.63) Industry dummy, other servce (start of perod) (1.20) (0.29) South Regon (start of perod) (1.37) (0.31) West Regon (start of perod) (0.10) (0.96) North Central Regon (start of perod) (0.03) (1.03) Constant (1.92)* (2.32)** Rho Wald test of Rho=0 (Ch squared [1]) Number of Observatons Absolute value of robust clustered t statstcs are n parentheses, * sgnfcant at 10 % level, ** sgnfcant at 5% level, *** sgnfcant at the 1% level. 14

16 TABLE 7: Hourly Wage Regressons, for Male Workers (1994), Usng Frst-Stage Bvarate Probt (1) (2) (3) All Dsplaced Layoffs Plant close Varables (Dependent Varable s ln(hourly wage) ln(wage) ln(wage) ln(wage) Observed Dsplaced, Predcted Dsplaced (Perod 1) (2.52)** (2.27)** (2.38)** Observed Dsplaced, Not Predcted Dsplaced (Perod 1) (2.40)** (2.27)** (1.62) Not Observed Dsplaced, Predcted Dsplaced (Perod 1) (0.91) (0.94) (1.12) Observed Dsplaced, Predcted Dsplaced (Perod 2) (0.92) (1.45) (0.56) Observed Dsplaced, Not Predcted Dsplaced (Perod 2) (1.49) (1.90)* (0.27) Not Observed Dsplaced, Predcted Dsplaced (Perod 2) (1.83)* (1.98)** (1.05) Urate (state unemployment rate, 1984) (2.13)** (1.96)** (1.82)* Race (nonwhte=1) (3.79)*** (3.68)*** (2.17)** Hghest grade Completed (as of 1994) (9.58)*** (9.12)*** (9.51)*** Tenure wth Employer (as of 1994) (7.10)*** (6.50)*** (5.41)*** Tenure Squared (5.69)*** (5.35)*** (4.35)*** Experence (as of 1994) (3.84)*** (3.87)*** (3.46)*** Experence (as of 1994) (2.21)** (2.21)** (2.10)** Occupatonal dummy, clercal (as of 1994) (4.01)*** (3.77)*** (3.02)*** Occupatonal dummy, laborer (as of 1994) (5.92)*** (5.35)*** (4.99)*** Occupatonal dummy, servce (as of 1994) (6.84)*** (6.17)*** (6.47)*** Industry dummy, agg. and mn. (as of 1994) (1.91)* (2.32)** (1.27) Industry dummy, const. (as of 1994) (4.19)*** (3.90)*** (1.17) Industry dummy, other manuf. (as of 1994) (0.41) (0.40) (0.02) Industry dummy, other servce (as of 1994) (2.42)** (2.14)** (0.98) South Regon (as of 1994) (0.78) (1.03) (0.85) West Regon (as of 1994) (4.79)*** (4.36)*** (3.73)*** North Central Regon (as of 1994) (7.16)*** (6.76)*** (5.94)*** Number of Observatons Absolute value of robust clustered t statstcs are n parentheses, * sgnfcant at 10 % level, ** sgnfcant at 5% level, *** sgnfcant at the 1% level. A constant has also been ncluded n estmated equatons. 15

17 REFERENCES Akerlof, George (1976). The Economcs of Caste and the Rat Race and Other Woeful Tales. Quarterly Journal of Economcs, 90, Farber, Henry S (1997). "The Changng Face of Job Loss n the Unted States, " Brookng Papers on Economc Actvty: Mcroeconomcs, Gbbons, Robert, and Lawrence Katz (1991). Layoffs and Lemons. Journal of Labor Economcs, 9, Jacobson, Lous S., Robert J. LaLonde, and Danel G. Sullvan (1993). Earnngs Losses of Dsplaced Workers. Amercan Economc Revew, 83 (4), Kletzer, Lor G. (1998). Job Dsplacement. Journal of Economc Perspectves, 12(1), Kletzter, Lor G., and Robert W. Farle (2003). The Long-Term Costs of Job Dsplacement for Young Adult Workers. Industral and Labor Relatons Revew, 56 (4), Krashnsky, Harry (2002). Evdence on Adverse Selecton and Establshment Sze n the Labor Market. Industral and Labor Relatons Revew, 56 (1), MaCurdy, Thomas, Thomas Mroz, and R. Mark Grtz (1998). An Evaluaton of the Natonal Longtudnal Survey of Youth. Journal of Human Resources, 33(2), Nakamura, Em (2004). Layoffs and Lemons over the Busness Cycle. Workng Paper, Harvard Unversty. Ruhm, Chrstopher (1991). Are Workers Permanently Scarred by Job Dsplacements? Amercan Economc Revew, 81(1), Spence, Mchael (1973). Job Market Sgnalng. Quarterly Journal of Economcs, 87 (3), Stevens, Ann H. (1997). Persstent Effects of Job Dsplacement: The Importance of Multple Job Losses. Journal of Labor Economcs, 15 (1),