THE DISTRIBUTION OF DISCRIMINATION IN IMMIGRANT EARNINGS - EVIDENCE FROM BRITAIN *

Transcription

1 THE DISTRIBUTION OF DISCRIMINATION IN IMMIGRANT EARNINGS - EVIDENCE FROM BRITAIN * Abstract: KEVIN J. DENNY Unversty College Dubln & Insttute for Fscal Studes, London COLM P. HARMON Unversty College Dubln MAURICE J. ROCHE Natonal Unversty Ireland Maynooth Verson 4.11 August 21st 1997 Ths paper uses the General Household Survey data for the UK to study earnngs dscrnaton between natves and grants. The key result s that the an source of dscrnaton s ethncty rather than grant status per se. Ths paper dffers fro the conventonal focus n studes of earnngs dscrnaton, whch focus on ean wage dfferences. In contrast we study the entre dstrbuton of the wage gap, and ncorporate dstrbutonally senstve easures of the wage gap reflectng dfferent levels of averson to dscrnaton. Our results are consstent wth prevous studes for the UK that fnd that non-whte grants are the ost wdely dscrnated n ters of ther labour arket returns. Moreover ths dscrnaton on the bass of colour s also present n the sub-saple of natves. * The authors acknowledge the coents of Patrck Geary, Stephen Jenkns and Dorren McMahon on earler versons of ths paper. The usual dsclaer apples. Materal fro the General Household Survey s used wth persson of the Offce for Populaton Censuses and Surveys and ade avalable through the Data Archve at the Unversty of Essex. The provders bear no responsblty for the calculatons contaned heren nor for any nterpretaton ade by the authors. Correspondng author: Kevn Denny, Departent of Econocs, Unversty College Dubln, Belfeld, Dubln 4, Ireland. Phone (+353 1) Fax (+3531) E-al kevn.denny@ucd.e.

2 INTRODUCTION Ths paper looks at the dstrbuton of wage dfferences between grants and natves n Brtan over the perod Unlke than the conventonal Blnder-Oaxaca decoposton, we consder the coplete dstrbuton of the ean wage gap. In addton we calculate easures of the degree of dscrnaton based on ths dstrbuton and show how these ay depend on alternatve levels of averson to dscrnaton. The lterature s rch n studes explorng the ssue of grant wage dscrnaton n varous fors n the labour arket. Recent contrbutons, largely led by the work of Borjas (1995) have been based on large-scale croeconoc datasets typcally usng US census data. Papers by Blackaby et al (1995) and Bell (1997) explores the ssue for the Unted Kngdo usng the General Household Survey (GHS) pooled over a 2 year perod and fnds that nonwhte grants are the ost wdely dscrnated aganst n ters of ther labour arket returns. Several studes have explored the ssue for the UK usng sngle cross sectons of crodata (typcally GHS) such as Chswck (1983), McNabb and Psacharopoulos (1981) or Stewart (1983). Usng the ore general set of easures proposed by Jenkns (1994) we fnd evdence of dscrnaton, results whch are consstent wth those of Bell (1997). Our fndngs suggest that the dscrnaton s concentrated by ethncty and not necessarly drven by grant status n tself. Secton I outlnes our approach n ters of earnngs dscrnaton ndces. Secton II presents our fndngs based on the GHS pooled fro Secton III dscusses ltatons to our approach. Secton IV concludes. 1

3 I EARNINGS DISCRIMINATION INDICES We exane whether dscrnaton between grants and natves s present by frst estatng standard earnngs regressons of the for: n (1) log( y ) = X n +u, N, (2) log( y ) = X + u, M, where M s the set of grants, N s the set of natves, y s the hourly wage of person, X s a vector of person-specfc explanatory varables, the β s are paraeter vectors and the u s are rando error ters. Evdence of dscrnaton (n ters of the approach of Oaxaca, 1973) s usually based on coparng the predcted wage of one group, grants for exaple, wth ther predcted reference wage. The predcted reference wage of grants (denoted rˆ ) s defned as the wage they would receve f all the grants attrbutes had the sae rate of return as the natve group. Thus, the predcted reference and actual wages are calculated as: (3) rˆ = exp yˆ = exp N ( X βˆ ) N ( X βˆ ) M, N ( β ˆ ) N rˆ = exp X. The conventonal suary easure of wage dscrnaton s the ean log wage gap, expressed n percentage ters, (4) D F 1 1 [ exp( ean[ log( rˆ ) log( yˆ )])] M [ exp( ean[ ( rˆ yˆ )/ yˆ ])] M. 2

4 Whle the approxaton ndcated n (4) provdes an ntutve nterpretaton of the easure as the ean proportonate dfference n the wages of the reference groups, t s nexact to the extent that the varances dffer between groups 1. Dfferences between the reference and actual wages for the grant groupng would ndcate dscrnaton s present. Jenkns (1994) argues that ths suary approach "...ay be consstent wth very dfferent dstrbutons of dscrnaton experence". For exaple, suppose that n 1974 and 1993 the average wage gap s estated to be 5% n favour of natves. In 1974 the average wage gap could be due to all grants gettng pad 5% less than natves, whle n 1993 the average wage gap could be due to half the grants gettng the sae wage and the other half gettng 1% less than natves. The presence of heterogenety n dscrnaton across a dstrbuton of ndvduals sees lkely gven the fndngs of Stewart (1983), who shows that whle the edan dfferental n earnngs by race was about 12% aganst non-whtes n the UK the dfferental for those n the upper quartle of earnngs falls to 3% and at the lower quartle the gap wdens to alost 2%. Jenkns (1994) proposes soe ethods for analyzng the jont dstrbuton of the two groups. The frst of these easures s based on the Lorenz curve and the assocated Gn coeffcent. We order the grants n ascendng order of the observed wage and plot the cuulatve predcted wage per capta aganst cuulatve saple share for each eber of the grant saple. In a slar fashon we suarze the predcted reference wage by plottng the cuulatve predcted reference wage per capta aganst the cuulatve saple share ranked n the sae order as for the Lorenz Curve. In Jenkns (1994) ternology these two plots are the Generalzed Lorenz Curve (GLC) and Generalzed Concentraton Curve (GCC). If there s no dscrnaton the curves wll concde, but there wll be dscrnaton f the predcted reference and predcted actual wages are not equal. In ths nstance the GCC curve wll le everywhere above the GLC curve. An aggregate dscrnaton ndex that s analogous to the 3

5 4 Gn coeffcent (n the sense of usng the area between the curves as a easure of dscrnaton) s gven as: (5) y y - r n 1 - y y - r 2n = C M ˆ ˆ 2 where the bar ters denote eans. Ths easure takes the dfferences n eans as n a tradtonal decoposton easures of dscrnaton but also ncorporates a ter whch s the wage gap weghted by the rank n the predcted earnngs dstrbuton. The second easure of overall dscrnaton suggested by Jenkns (1994) s: (6) ( ), d 1- d = J - n = 1 M - α α α ω = ω 1 where > r y - r = 1+ d y n y = α ω ˆ ˆ ˆ Here ω s a grants wage share and d s a noralzed wage gap (the gap for the grant ndvdual relatve to the ean of the reference dstrbuton). If the wage gap equals 1 (n the no dscrnaton case where the predcted reference and actual wage are equal) the J α s equal to zero. Ths easure allows for the wage gaps to be aggregated n dfferent ways, wth the paraeter α nterpreted as the degree of dscrnaton averson, wth hgher values for α correspondng to greater averson (Jenkns, 1994). Wth ths nterpretaton the averson paraeter s the ncrease n wages requred to copensate an ndvdual for a sall ncrease n

6 the wage gap (or dscrnaton) large values of α suggest a large averson to dscrnaton. It s soewhat analogous to the falar Atknson Averson to Inequalty paraeter. Ths ndex (unlke the Lorenz Curve based C Index) has the desrable property that t s addtvely decoposable: that s one can wrte the aggregate dscrnaton aganst grants (say) nto a weghted su of the dscrnaton ndces pertanng to two sub groups: black grants and whte grants. 2 The J ndex s a concave functon of proportonate wage gaps: a gven percentage wage gap receves a lower weght the hgher t occurs n the wage dstrbuton reflectng the vew that one cares less about dscrnaton aganst the relatvely well off. One could, of course, argue for a convex functon on the bass that a gven wage gap corresponds to a greater absolute aount of dscrnaton the hgher t occurs n the wage dstrbuton. Wth soe farly plausble assuptons about the underlyng Socal Welfare Functon (see Jenkns, 1991) one can wrte welfare for the group of nterest (.e. those for who we are easurng dscrnaton aganst) as (7) W = y w ( 1 J ) α whch akes explct the welfare cost of dscrnaton n ters of average wages of the group. II. RESULTS Usng the pooled saple We estate a sple earnngs functon usng GHS data for the perod usng the log real wage (n 1974 prces) where the GHS nonal wage s dvded by the consuer prce ndex. Our set of characterstcs nclude age, age squared and years of educaton, as well as sets of duy varables to control for regonal, occupatonal and ndustry effects as well as year dues fro (1974 s the reference year). 3 5

7 Our total saple sze s 98,839, of whch 92,726 are natves and 6,133 are grants defned n the GHS by country of brth. The classfcaton of non-whte for the purposes of ths paper s defned explctly wthn the GHS questonnare. The ntervewer ust observe the respondent and classfy accordngly. Thus non-whte s not defned arbtrarly by vrtue of specfc countres of orgn. Mgrant status s therefore separate to whte/non-whte status. Whle the GHS s a representatve survey t s possble that recent grants are underrepresented snce they are less lkely to be n the saplng frae 4. We confne our analyss to eployed ales. In addton to analyzng the dscrnaton between total natves and grants we can dvde the total nto whte and nonwhte sub groups. In Table 1 we present suary statstcs for the grant wage dstrbuton for the total, whte and non-whte groups. The raw wage gap between predcted reference and actual wages s always postve for each group and s relatvely large for the non-whte group, supportng the conclusons of Bell (1997). Table 1 Suary statstcs for grant wage dstrbutons Statstc Total Whte Non-whte Predcted reference wages Mean Medan Predcted wages Mean Medan Raw wage gaps (r M -y M ) Mean Medan It s also nforatve to llustrate the coplete dstrbuton of dscrnaton usng the Generalzed Lorenz and Generalzed Concentraton Curves. An appealng feature of the dagraatc approach n the context of ths paper s that t llustrates the coplete dstrbuton of the degree of dscrnaton. In Fgure 1A we present the GLC (lght lne) and GCC (heavy 6

8 lne), estated usng the total saple of natves and grants. The GCC les everywhere above the GLC, whch ndcates wage dscrnaton, s present over the whole dstrbuton. Ths s not as evdent when we exane the case of whte grants versus whte natves (see Fgure 1B), but s very evdent when we exane the case of non-whte grants versus nonwhte natves (see Fgure 1C). Fro ths t appears that t s the non-whte grant group that are beng dscrnated aganst and ths s reflected n the results for the total saple. Fgure 1A Dscrnaton between Mgrants and Natves - Total Saple 1.2 (Cuulatve Saple Share) x (Mean Earnngs of Reference Group) GCC GLC Populaton Share 1 Fgure 1B Dscrnaton between Mgrants and Natves - Whte Saple 7

9 1.2 (Cuulatve Saple Share) x (Mean Earnngs of Reference Group) GCC GLC Populaton Share Fgure 1C Dscrnaton between Mgrants and Natves - NonWhte Saple 1.25 (Cuulatve Saple Share) x (Mean Earnngs of Reference Group) GCC GLC Populaton Share 8

10 In order to throw further lght on these fndngs we repeat our analyss but ths te we exane the dstrbuton of dscrnaton experences between whte and non-whte groups. We fnd that there s soe degree of colour dscrnaton n the total saple (see Fgure 1D) whch s not due to dscrnaton n the natve saple (see Fgure 1E) but reflects dscrnaton n the grant saple (see Fgure 1F). It s also noteworthy that whereas n ost of the dagras the curves grow further apart as we ove to the rght, n Fgure 1E, the two converge. Ths suggests that there s lttle dscrnaton n the natve saple between blacks and whtes at the upper end of the dstrbuton. 9

11 Fgure 1D Dscrnaton between Non Whtes and Whtes - Total Saple 1.25 (Cuulatve Saple Share) x (Mean Earnngs of Reference Group) GCC GLC Populaton Share Fgure 1E Dscrnaton between Non Whtes and Whtes - Natve Saple 1.25 (Cuulatve Saple Share) x (Mean Earnngs of Reference Group) GCC GLC Populaton Share 1

12 Fgure 1F Dscrnaton between Non Whtes and Whtes Mgrant Saple (Cuulatve Saple Share) x (Mean Earnngs of Reference Group GCC GLC Populaton Share These results are also evdent when we calculated dstrbutonal dscrnaton ndex estates. The standard D f and Jenkns (1994) C and varous J α ndces are calculated for natve/grant (see Table 2) and whte/non-whte (see Table 3) wage dscrnaton. The degree of grant dscrnaton usng any of these ndces s always hgher for the non-whte group. The degree of non-whte dscrnaton usng any of these ndces s always hgher for the grant group. Thus over the perod non-whte grants appeared to be subject to real wage dscrnaton. The frst colun n Table 2 copares all grants aganst all natves. The second colun copares grants who are whte aganst natves who are whte and so on. The frst row gves the conventonal ean wage gap and the second s the Gn based easure. We can use equaton (7) to nterpret the varous J ndces. Table 2 ples that for α =.25, the welfare cost to grants of dscrnaton s equvalent to the welfare loss caused by reducng ther average wages by about 2.4%. For purposes of coparson we nclude estates of ale/feale dscrnaton 11

13 fro Jenkns (1994) whch suggest that our estates of dscrnaton aganst non-whtes are of slar agntude to that experenced by feales n the labour arket. The fndng that colour s the key dstncton contrasts wth the results of Heath & McMahon (1997b) who fnd n a study of occupatonal attanent that frst generaton Irsh grants to Brtan the bulk of the whte grants- experence an ethnc penalty n uch the sae anner as other grant groups. Table 2 Dstrbutonal Dscrnaton Index: Natves versus Mgrants Total Whte Non-whte Males vs Feales (Jenkns) Df C J α α= J α α= J α α= J α α= J α α= J α α= Table 3 Dstrbutonal Dscrnaton Index: Whte and Non-whte Groups Total Natves Mgrants Df C J α α= J α α= J α α= J α α= J α α= J α α=

14 The te pattern of dscrnaton In ths secton we brefly consder the evoluton of dscrnaton ndces over the perod. We estate the earnngs functons for each year of the GHS saple for each year and graph the J ndex for α = 1 n Fgure 2A coparng total natves wth total grants as well as akng the sae coparson for the whtes and non-whtes. Consstent wth Table 2 and Table 3 we fnd postve levels of dscrnaton wth a uch hgher level aganst nonwhtes n all but one year. Fgure 2B plots the sae dscrnaton ndex for non-whtes versus whtes wthn the whole saple. Ths backs up the story n Fgure 2A: when we dstngush between natves and grants there s very lttle dfference, the three lnes ove together, except for the years In general t s dffcult to detect a te pattern n the dscrnaton ndces, apart fro a slght upward tendency over the latter part of the perod. Ths also holds f we plot the D f or C ndces over te where the ncrease s ore pronounced. Indeed the stablty of the easures s rather strkng consderng the changes n the Brtsh labour arket throughout the 198 s. One ght expect the large fluctuatons n uneployent and an ncreasngly de-regulated and less unonsed labour arket to have had soe nfluence on the level of dscrnaton, although econoc theory provdes few, f any, predctons as to the effect on dscrnaton. Moreover there has been an ncrease n ncoe nequalty over the 198 s so one ght expect that to have had soe pact. 6 If so, t s not apparent. To test ths ore forally we regressed our te seres easures of dscrnaton aganst a nuber of ndcators of aggregate actvty ncludng uneployent, GDP and nflaton 7. In general no robust results eerge although there s weak evdence of a negatve effect of uneployent on soe of the dscrnaton easures. 13

15 Fgure 2A Natve-Mgrant J Dscrnaton Index (α = 1).45.4 Total Whte Non-Whte.35.3 Index Value Year Fgure 2B Whte-NonWhte J Dscrnaton Index (α = 1).45.4 Total Natve Mgrant.35.3 Index Value Year 14

16 III DISCUSSION There are any other aspects to the study of dscrnaton, whch we have not explored. Frstly, race s lterally and fguratvely not a black and whte dstncton. We have for exaple eschewed dstngushng between Asan, Afrcan and Carbbean grants although there s consderable research n the socology lterature whch suggests that ths dstncton s portant. These dfferences arse not just because there ay be ntrnsc dfferences between the ethnc groups but also because they wll have arrved at dfferent tes reflectng changes n both the econoc crcustances of the recevng and sender country and changes n graton polcy 8. We have also assued that the dstncton between grant and natve s clear cut: however the process of asslaton ay ean that long ter grants have ore n coon wth natves than wth recently arrved grants. The role of asslaton has been wdely studed by socologsts but less so by econosts 9. As ponted out by Stewart (1983) research such as ths consders only dscrnaton wthn the labour arket; n partcular t takes characterstcs of ndvduals such as educaton as exogenous and dentfes dscrnaton as arsng for dfferences n the returns to these characterstcs. If, for exaple, the return to schoolng for a partcular group s low then we ght expect the to respond by partcpatng relatvely less. There s evdence that schoolng s endogenous and that controllng for ths usng nstruental varables procedures affects the estated return. 1 However fndng a sutable nstruent for both natve and grant groupngs n ths nstance s probleatc. So we do not study dscrnaton whch ay nfluence access to educaton or eployent status tself. The nterpretaton of dfferences n returns across deographc groups as dscrnaton s, of course, debatable. The estates of the returns are condtonal on the specfcaton of the equatons and holdng other thngs equal. We have controlled for educaton n the standard way 15

17 (years of schoolng) but t s possble that the qualty of schoolng dffers between grants or natves or that eployers perceve ths to be the case. There s soe evdence that the huan captal of grants s less portable 11. It ay be possble to get further evdence on the extent to whch dscrnaton s colour based by dstngushng between frst generaton grants (those who actually grate) and second generaton grants (ther off-sprng, born after graton occurred) snce any dsadvantage arsng fro grants status per se should not apply to the second generaton. Fnally we have not attepted to deal wth saple selecton ssues whch could arse f the nature of graton to Brtan was non rando. It s dffcult to see how one could pleent any correcton for ths potental proble gven that we are unable to odel structurally the graton process, as no nforaton as to the reasons for graton s n the dataset. However evdence for the Unted States shows that varatons n the decson to grate are attrbutable to econoc and poltcal crcustances n the hoe countres and that ths affects the labour arket qualty of grants. 12 IV CONCLUSIONS Ths paper apples the ethods of Jenkns (1994) to exane the ssue of dscrnaton aganst grants n the UK labour arket. The nnovaton of ths approach s that t changes the focus away fro sply lookng at the average level of dscrnaton and consders the entre dstrbuton of wage dscrnaton. Usng GHS data pooled over the years our fndngs support recent research on the UK labour arket experences of grant workers and suggest that the dscrnaton present ay be racal rather than due to grant status alone. Across the broad grant/natve splt the grant saple appear dscrnated aganst, but coparng whte natve to whte grant does not support any strong presence of dscrnaton practces. However 16

18 exanaton of the data for non-whte grants shows clear adverse dfferences for ths saple aganst whte natves and even aganst non-whte natves. 17

19 REFERENCES Baron, Jaes N.; Grusky Davd B. and Trean Donald (eds) Socal Dfferentaton and Socal Inequalty, Boulder, Colorado. Westvew Press, Bell, Bran D. The Perforance of Igrants n the Unted Kngdo: Evdence fro the GHS. Econoc Journal, 17(2), pp , March Blackaby, Davd; Clark, Ken; Lesle, Derek and Murphy, Phlp. Black-Whte Male Earnngs and Eployent Prospects n the 197 s and 198 s: Evdence for Brtan. Econoc Letters, 46, pp , Blanchflower, Davd and Oswald, Andrew. The Wage Curve. Cabrdge: MIT Press, Borjas, George J. Self-Selecton and the Earnngs of Igrants. Aercan Econoc Revew, 77(4) pp , Borjas, George J. Asslaton and Changes n Cohort Qualty Revsted; What Happened to Igrant Earnngs n the 198 s. Journal of Labor Econocs, 13(2), pp , Aprl Chswck, Barry R. The Earnngs of Whte and Coloured Male Igrants n Brtan. Econoca, 47(1), pp.81-87, February Fredberg, Rachel M. You can t take t wth you? Igrant asslaton and the portablty of huan captal. NBER Workng paper 5837, Goodan Alssa and Webb Steve. For Rcher, For Poorer the changng dstrbuton of Incoe n the Unted Kngdo, Fscal Studes, 15(4), pp.29-62, Noveber Goyder, John. The Slent Mnorty: Non-Respondents on Saple Surveys. Boulder, Colorado: Westvew Press, Farley, Reynolds Blacks, Hspancs and Whte Ethnc Groups: Are Blacks Unquely Dsadvantaged? Aercan Econoc Revew, Papers and Proceedngs, 8(2), pp , 199. Halsey A.H.; Lauder, Hugh; Brown, Phlp and Wells, Ay Stuart (eds) Educaton: Culture, Econoy and Socety, Oxford, Oxford Unversty Press, Haron, Col and Walker, Ian. Estates of the econoc return to schoolng for the UK Aercan Econoc Revew, 85(5), pp ,

20 Heath, Anthony and McMahon, Dorren Educaton and Occupatonal Attanents: the pact of ethnc orgns n Halsey et al (1997a). Heath, Anthony and McMahon, Dorren Ethnc Dfferences n the Labour Market: uneployent, self-eployent and salared eployent, forthcong, Revue du Socologe (1997b). Jenkns, Stephen P. Aggregaton Issues n Earnngs Dscrnaton Measureent, Econocs Dscusson paper no. 1/91 (Unversty of Bath, Bath) Jenkns, Stephen P. Earnngs dscrnaton easureent: A dstrbutonal approach. Journal of Econoetrcs, 61(1), pp , McNabb, R. and Psacharopoulos, George. Racal Earnngs Dfferentals n the UK. Oxford Econoc Papers, 33, pp , Nolan, Bran, Macroeconoc condtons and the sze dstrbuton of ncoe: evdence fro the Unted Kngdo, Journal of Post-Keynesan Econocs 11, Oaxaca, Ronald. Male-Feale Wage Dfferentals n Urban Labour Markets. Internatonal Econoc Revew, 14, pp , Stewart, Mark B. Racal Dscrnaton and Occupatonal Attanent n Brtan. Econoc Journal, 93, pp , Septeber Trean, Donald J and Lee Hye-Kyung. Incoe dfference Aong 31 Ethnc Groups n Los Angeles n Baron et al (1996). 19

21 Notes If log( y) = X + u, u ~ N(, σ ) then E ( y) = exp( X + σ 2). Our thanks to Steve Jenkns for ths pont. 2 Ths decoposton s not pursued here n the present paper. 3 Full suary statstcs and estates are avalable fro the authors on request. See studes such as Blanchflower and Oswald (1995) for ore detaled descrptons of the GHS dataset. 4 See Goyder(1987). Ths happens because they are lkely to be ore footloose. 5 Both graphs are 3 year ovng averages. 6 See, for exaple, Goodan and Webb(1994). 7 Ths ethod was used by Nolan(1988/89) to nvestgate acroeconoc patterns n aggregate nequalty n the UK. 8 For exaple Heath and McMahon(1997a, 1997b) for the UK, Farley(199) or Trean and Lee(1996) for Aercan evdence. 9 But see Borjas(1995) for the US and Bell(1997) for Brtan. 1 See, for exaple, Haron and Walker (1996) for consderaton of the ssue of endogenous schoolng n the GHS data. 11 See Fredberg(1996). 12 See Borjas(1987). 2