Ethnic Sorting in the Netherlands

Transcription

1 DISCUSSION PAPER SERIES IZA DP No 3155 Ethnc Sortng n the Netherlands Aslan Zorlu Jan Latten November 2007 Forschungsnsttut zur Zukunft der Arbet Insttute for the Study of Labor

2 Ethnc Sortng n the Netherlands Aslan Zorlu AMIDst, AIAS, Unversty of Amsterdam and IZA Jan Latten Statstcs Netherlands Dscusson Paper No 3155 November 2007 IZA PO Box Bonn Germany Phone: Fax: E-mal: za@zaorg Any opnons expressed here are those of the author(s) and not those of the nsttute Research dssemnated by IZA may nclude vews on polcy, but the nsttute tself takes no nsttutonal polcy postons The Insttute for the Study of Labor (IZA) n Bonn s a local and vrtual nternatonal research center and a place of communcaton between scence, poltcs and busness IZA s an ndependent nonproft company supported by Deutsche Post World Net The center s assocated wth the Unversty of Bonn and offers a stmulatng research envronment through ts research networks, research support, and vstors and doctoral programs IZA engages n () orgnal and nternatonally compettve research n all felds of labor economcs, () development of polcy concepts, and () dssemnaton of research results and concepts to the nterested publc IZA Dscusson Papers often represent prelmnary work and are crculated to encourage dscusson Ctaton of such a paper should account for ts provsonal character A revsed verson may be avalable drectly from the author

3 IZA Dscusson Paper No 3155 November 2007 ABSTRACT Ethnc Sortng n the Netherlands * Ths paper examnes the resdental moblty behavour of mgrants and natves n the Netherlands usng a rch admnstratve ndvdual data fle The nclnaton to move and the choce of destnaton neghbourhood are estmated, correctng for the selecton bas of movers Subsequently, the role of preferences n the moblty behavour s mplctly derved from regresson estmates The analyss shows that the percentage of natves n the destnaton neghbourhood s predcted to be about 18 percentage ponts lower for nonwestern mgrants than for natves About 65 percent of the dfferental s explaned by ther observable characterstcs; the remanng part can largely be attrbuted to preferences and dscrmnaton No ndcaton s found of the spatal assmlaton of second-generaton nonwestern mgrants On the other hand, the moblty pattern of the second-generaton western mgrants s smlar to that of natves JEL Classfcaton: J1, J6, R3 Keywords: mgrants, resdental segregaton Correspondng author: Aslan Zorlu AIAS, Amsterdam Insttute for Advanced Labour Studes Unversty of Amsterdam Plantage Mudergracht TV Amsterdam The Netherlands E-mal: AZorlu@uvanl * Ths research s supported by the Netherlands Organsaton for Scentfc Research (NWO), VICI grant no The paper s part of a jont project wth Statstcs Netherlands; we are grateful to the Department of Total Statstcs for allowng us the use of the Socal-Statstcal Database

4 2 1 Introducton Wth the massve nflux of non-western mmgrants from the 1970s onwards, the Netherlands has turned nto an mmgraton country In the past few decades, large numbers of mmgrants from less-developed non-western countres have changed the demographc and socal composton of the populaton at the natonal level The newcomers are concentrated n the economc urban centres n the Western part of the country, manly n older neghbourhoods where houses are cheaper Ths tendency to spatal concentraton n the older urban parts n the country was accelerated by an ongong selectve suburbanzaton of natves As a result, many neghbourhoods n the four largest Dutch ctes of Amsterdam, Rotterdam, The Hague, and Utrecht have become predomnantly nhabted by mgrants Government polces on the natonal and local levels seek to counteract ths tendency Most polcy measures are based on the assumpton that the spatal segregaton of the poor derves from ther ndvdual economc poston If a neghbourhood becomes less popular, those who cannot afford to lve n a more expensve neghbourhood stay and are joned by newcomers who are equally mpovershed Eventually, the selectve resdental moblty leads to more economc segregaton Snce non-western mgrants have a less favourable economc poston, an outcome of more ethnc segregaton s selectve moblty Prevous research has demonstrated that dspartes n ncome and wealth play an mportant part n explanng ethnc resdental segregaton (Bolt and Van Kempen, 2003) Polcy makers tend to focus on the housng market to manpulate resdental moblty, because the hgh densty of socal-rental houses n the large ctes s seen as one of the man sources of ethnc resdental segregaton Correspondngly, mxng housng stocks s vewed as an effectve nstrument to create ntegrated neghbourhoods where varous economc classes and, mplctly, dfferent ethnc groups wll resde (Musterd et al, 2003) However, lttle s known about the drvng forces n the ongong process of ethnc spatal segregaton on a local level Ths study examnes ethnc dfferences n moblty patterns n the Netherlands and ams to assess the role of observed characterstcs and unobserved factors such as preferences and dscrmnaton In addton to dspartes n ndvdual socoeconomc postons, preferences can drve locaton choce accordng to the ethnc composton of the neghbourhood One can argue that the preferences of natves, the domnant group, are more mportant n ntatng the outflow and the transformaton of a neghbourhood, because natves are n a relatvely advantageous poston They have more opportuntes to realze ther preferences to lve n a neghbourhood wth a desred ethnc composton Whatever the reasons from whch they may be derved, natve segregatory taste or ntolerance does not need to be very pronounced Schellng (1971) suggests n hs segregaton model that the ndvdual behavour of the domnant group, even f based on mnor dfferences n preferences, can lead to segregaton n urban areas, even n the absence of other segregatng forces Preferences regardng the ethnc makeup of the neghbourhood are dffcult to dentfy n the absence of qualtatve nformaton on stated preferences The housng demand survey contans some nformaton, but t s unsatsfactory Such data are not avalable n the Netherlands Instead, we have used a rch ndvdual admnstratve data fle from 2002 and 2003 (SSD) housed by Statstcs Netherlands Ths unque database allows ethnc dfferences to be consdered wth respect to many ndvdual demographc and socoeconomc characterstcs, as well as relevant 2

5 3 neghbourhood characterstcs and housng market restrctons Addtonally, the attachment to resdence locaton s approxmated by the presence of parents n the same muncpalty, bearng n mnd recent evdence of the relatvely great adverse effect of famly tes on moblty behavour among blacks and low-ncome groups (Splmbergo and Ubeda, 2004; Dawkns, 2006) The data allow ndvdual geographcal moblty to be tracked, but do not nclude drect nformaton on stated preferences Therefore, to uncover the role of preferences and dscrmnaton n ethnc segregaton process, an mplct strategy has been appled: the Oaxaca-Blnder decomposton technque Ths dentfes the contrbuton of observed relevant factors n data to the moblty behavour The remanng unexplaned part of moblty behavour may be attrbuted to preferences and dscrmnaton, whch we refer to here as preferences The role of more explct dscrmnaton n the Dutch housng market s lkely to be more lmted and more embedded n preferences than n the Unted States However, the unexplaned part could also be the result of some other factors that are not n our data Nevertheless, ths study has the potental to dstngush the preference component of ethnc resdental sortng n the Netherlands, because most relevant factors are taken adequately nto account The paper s organzed as follows: the next secton gves a bref hstory of mmgraton to the Netherlands and ethnc spatal concentraton Secton 3 conssts of a descrpton of the data fle used In Secton 4, we dscuss the analytcal framework to estmate the geographcal moblty patterns by ethnc background In Secton 4, the moblty dfferentals are decomposed nto explaned and unexplaned components Secton 5 concludes 2 Trends n spatal concentraton The recrutment of guest workers from Medterranean countres, Turkey and Morocco n partcular, ntated a contnuous growth of the share of non-western mgrants n the Dutch populaton Many of the guest workers dd not return home and formed the bass of mmgrant communtes that grew rapdly through famly reunfcaton n the 1970s and 1980s and famly formaton n more recent years In the md-1970s, mmgraton flows from the former Netherlands colones of Surnam and the Antlles gave an addtonal mpetus to the growng share of mmgrant populaton In addton, mmgraton from developed countres has comprsed almost half the mmgraton flows and has followed busness cycles Durng the 1990s, asylum seekers domnated mmgraton flows Between the early 1970s and 2007, the new non-western populaton has grown tenfold from 160,000 to 17 mllon, ncludng the second generaton, whle the number of western mgrants has reached 14 mllon The Netherlands populaton has always been concentrated n the western part of the country where economc actvtes are concentrated, and today ths tendency s even more marked for the newcomers At the regonal level, the newcomers are concentrated n the cheaper areas n the urban zones, manly n the older neghbourhoods wth cheaper houses In contrast wth other mgrants, the Surnamese and Antllean populaton have also settled n some newly-bult housng stock Asylum seekers n the 1990s spread out on arrval, but later also tended to move to the ctes (Zorlu and Mulder, forthcomng) The regonal and urban concentraton tendences on arrval n the Netherlands have concded wth the already establshed resdental suburbanzaton process of natves, whch accelerated n the 1960s Snce the 1970s, ths suburbanzaton has resulted n a sharp declne of the at that tme manly natve urban populaton At the same tme, a manly non-western foregn-born 3

6 4 populaton settled n these ctes and partly compensated for the natve Dutch resdental outflow, so that the total number of nhabtants of Amsterdam, Rotterdam, The Hague, and Utrecht has been stable snce the md-1980s Ths quanttatve stablty n populaton sze, however, hdes the ongong socal demographc dynamcs n populaton composton, especally regardng ethncty Ths ethnc selectve resdental moblty has correlated wth the socoeconomc status of movers (Musterd et al, 2007) Next to the ethncty-specfc mgraton patterns, natural growth fgures dfferentate smlarly accordng to ethnc background The natve cty populaton shows a clear excess of deaths over brths, whle the natural ncrease of the non-western foregn populaton stmulates the overall natural ncrease n the large ctes The combnaton of mgraton and natural growth trends explans the very fast changes n ethnc composton of the four largest Dutch ctes (Bontje and Latten, 2005) Among mgrants, resdental segregaton s stronger for those of non-western orgn In 2004, 10 percent of the Dutch populaton was non-western, but n the four largest ctes ther share was one thrd Wthn these ctes ther concentraton s even stronger In some neghbourhoods n The Hague, Amsterdam or Rotterdam, the share of non-western foregners has reached levels above 70 percent and even 80 percent, n contrast wth the natonal share of 10 percent The structure of the housng market s probably an mportant factor n the Netherlands Local housng markets n large ctes are domnated by socal rental housng and an extenson of the housng stock s subject to physcal lmtatons The presence of a socal housng sector wth a non-compettve rental polcy below market prces together wth tght regulatons generates a formdable ratonng regme Choces wthn the same cty are lmted for those who want to mprove ther housng qualty by movng to a larger and more comfortable home The suburbs are a logcal opton for short-dstance moblty nduced by economc and lfe cycle factors In ths study, housng market condtons generatng ths type of moblty have been carefully controlled for Whle the suburbansaton trend of natves contnues, the non-western populaton s also suburbansng At frst glance one can welcome ths tendency as proof of the socal moblty of non-western nhabtants who partcpate n a general pattern lke the suburbanzaton process, and respond n a smlar manner to shortages n urban housng stock However, one can also see that the drecton of suburbansaton for ethnc groups s not a random process It s rather selectve Mgrant movers from large ctes choose muncpaltes where ther co-ethncs are resdng, whle the natve outflow heads for other places Ths pattern may be an ndcaton not only of socoeconomc upward moblty, but also of ethnc preferences 3 Prevous research Prevous research has provded evdence that dfferences n preferences have certanly been nfluental n neghbourhood racal and ethnc transton, even when account s taken of other segregatng forces such as ndvdual and famly-level varables and non-ethnc neghbourhood characterstcs The preferences of the domnant group have two components: movng out; not movng nto an area where the share of ethnc mnorty groups s hgh and growng Earler studes consdered only the outflow of whtes, regarded as whte flght More recently, the role of whtes has been expressed as whte avodance, whch captures both outflow from, and not movng nto, ethncally mxed areas (Qullan, 2002; Crowder, 2000; Clark, 1992) Recent studes argue that the concentraton of blacks and low-ncome famles s persstent owng to the low nter-neghbourhood moblty of these groups Splmbergo and Ubeda,

7 5 demonstrate the mportance n the decson to mgrate of famly attachment, but t s much more mportant for Blacks than for Whtes, despte ther many characterstcs assocated wth hgh geographcal moblty such as a low rate of marrage and home ownershp and hgh unemployment Dawkns, 2006 shows that the nter-neghbourhood moblty of famles wth chldren s negatvely affected by local knshp tes and the socal networks of chldren The effect of local socal tes s partcularly strong for low-ncome famles In ths study we consder an ndvdual s attachment to the resdence locaton, whch s approxmated by the presence of parents n the same muncpalty Studes usng surveys of stated preferences suggest that the wllngness of whtes to lve n a multethnc area decreases as the share of Blacks, Hspancs, and Asans n a neghbourhood ncreases and the tolerance of whtes has ncreased n the course of tme These studes also suggest that the tolerance s hgher for the ethnc makeup of orgn locaton than for a desred ethnc composton of the destnaton neghbourhood (Farley et al, 1978, 1994; Galster, 1990; Clark, 1992; Krysan, 2002) The observed trend n ethnc and racal segregaton confrms the ndcatons of survey research (Crowder, 2000; Cutler et al, 1999) Although some members of the domnant group seem to have hgher degrees of tolerance, ndvdual behavour culmnates n aggregate results that are assocated wth hgher observed level of ethnc resdental segregaton than expected Schellng, 1971 lnked aggregate populaton processes to ndvdual behavour n hs neghbourhood tppng model, whch predcts that unorganzed ndvdual behavour based on slght dfferences n preferences can generate a perpetual process of whte flght, culmnatng n the loss of even more tolerant whtes Drawng on data on ether stated preferences or actual patterns, Amercan researchers have establshed that many urban areas have experenced a drastc transton, gven the extent of housng facltes, whch suggests the exstence of a crtcal pont of racal composton neghbourhood tppng above whch whte avodance perpetuates ethnc and racal segregaton Ths evdence s consstent to lfe course developments captured by a large number of relevant varables that may nfluence spatal moblty (Crowder et al, 2006; Cutler et al,1999; Galster,1990) Data employed n segregaton research usually nclude ether stated preferences or actual moblty behavour, but seldom both, because actual moblty behavour follows the preference statement wth a tme lag that n survey studes s hard to capture Survey studes based on expressed preferences and atttudes provde some ndcatons about the possble effect of an ethnc/racal component of ndvdual moblty, but are unable to dentfy ther nfluences on actual moblty behavour On the other hand, data on actual moblty patterns wthout preference assessment cannot fully measure the role of preferences These are lkely to be correlated wth many ndvdual and famly-level factors as well as non-ethnc neghbourhood characterstcs Perhaps ths s the reason why researchers have not been able to provde drect evdence of the mpact of preferences In ths study, preferences are derved from moblty behavour Dfferences n the ethnc composton of destnaton neghbourhood between natves and mgrants are decomposed nto explaned and unexplaned components, as formally shown n secton 4 In Europe, research on segregaton s scarce, probably because of the relatvely low levels of ethnc concentraton and small number of segregated areas n urban areas (Musterd, 2005) Dutch data suggest that neghbourhoods that already had hgher concentratons of ethnc populaton showed faster transformaton to segregaton than other neghbourhoods (Latten, Ncolaas and Wttebrood, 2005) However, sold evdence of the role of preferences n the segregaton 5

8 6 tendency s lackng n European research Ths study contrbutes to our understandng of the dynamcs of segregaton 4 Data We have used a rch ndvdual admnstratve databank from 2002 and 2003, the Socal Statstcal Database (SSD) housed by Statstcs Netherlands The data cover the entre populaton of the Netherlands The occurrence of moblty between neghbourhoods s dentfed by comparng the resdence neghbourhoods on the thrd Frday of September 2002 and 2003 The avalablty of data on two ponts n tme enables us to observe the orgn and destnaton neghbourhoods and the changes n ndvdual and household statutes as well as neghbourhood condtons, although the last wll probably be very small Snce the data cover the entre populaton and natonal geographcal area, we have been able to construct aggregate varables on the neghbourhood level, whch s the smallest avalable spatal area n the data, wth an average of about 1700 resdents After estmatng the neghbourhood transton matrx dsplayed n table 2, we took a random sample of 1 percent from the entre populaton of around 164 mllon to be able to perform more sophstcated regresson analyses that would otherwse be mpossble to conduct wth the techncal capacty avalable Furthermore, we restrcted the sample to people aged 18 to 64 years who are potentally decson makers n the household, snce the head of household s not dentfed n the data For the analyss, three major groups are dstngushed: natves, non-western mgrants, and western mgrants The last category contans mgrants from developed countres The nonwestern category conssts of the mgrants from developng countres It s domnated by four large groups: Turks, Surnamese, Moroccans, and Antlleans Table 1 gves the descrptve statstcs for a large number of ndvdual and neghbourhood-level varables for the three man groups In addton to the man demographc and household characterstcs, the transton n household structure s also gven by two dummy varables ndcatng the transton between sngle and marred status The socoeconomc poston s measured by dummy varables defned on the bass of ncome source as employed, unemployed, student, recever of another beneft, and nactve, whch s the reference category Addtonally, separate dummes have been constructed to control for the levels of earned ncome defned as the quartles of the dstrbuton The neghbourhood characterstcs are approxmated by a set of varables ndcatng the mean ncome, the unemployment rate, the share of owner-occuped houses, the mean value of houses, and the share of non-western mgrants To approxmate the degree of neghbourhood attachment, three control varables are used to ndcate the presence of parents n the muncpalty (both parents; one parent; no parents n the muncpalty) wth parents not alve as the reference We have consdered the muncpalty level rather than the neghbourhood, because muncpaltes and dstances are relatvely small and transportaton n the Netherlands s easy Unfortunately, ths rch data fle does not nclude nformaton on educaton, whch s a very mportant varable n ths knd of research It mmedately becomes apparent that there are substantal dfferences between the characterstcs of natves and non-western mgrants Western mgrants are smlar to natves; more than half of them were born n the Netherlands, whle only 15 percent of non-western mgrants are second generaton Non-western mgrants are relatvely young and are less lkely to be employed or 6

9 7 homeowners Ther household sze s large and they are less lkely to resde apart from ther parents Ther concentraton s substantally hgh n the three largest ctes of Amsterdam, Rotterdam, and The Hague They lve more often n neghbourhoods where the value of houses s relatvely low and the unemployment rate s hgh Table 1 Mean values of varables used, populaton between 18 and 64 years old n 2002 Varable Natve Non-western Western Age Woman Second Generaton na Homeowner The value of home n 1000s Marred Number of chldren n household Number of persons n household From sngle to marry From couple to sngle Employed Unemployed Recever of a beneft Student Labour ncome Q1-Q Labour ncome Q2-Q Labour ncome >Q Mean value of homes n neghbourhood (n 1000s) Mean annual labour ncome n neghbourhood % owner occuped homes n neghbourhood Unemployment rate n neghbourhood % non-western n the orgn neghbourhood Both parents n the same muncpalty One parent n the same muncpalty No parent n the same muncpalty Amsterdam Rotterdam The Hague Probablty of movng Share of natves n destnaton negh N Moblty pattern The dfferences between the resdental moblty patterns of natves and mmgrant groups can be shown usng some categores for the degrees of concentraton of non-western mgrants We defne seven neghbourhood types usng 10 percent ntervals up to 60 percent and one nterval for the hgher degrees Snce we are nterested n examnng a sngle moblty (2002 to 2003), the nter-(and ntra-)neghbourhood moblty among these seven categores can be captured by a K x K transton matrx, A 7

10 8 a A = a a a where the j th element a j represents the probablty of movng from neghbourhood n 2002 to neghbourhood j n 2003 The th row contans the probabltes of movng from neghbourhood n 2002 to each of K neghbourhood types n 2003 Thus, the elements of each row sum to 1 We estmate the elements of A from observed transtons usng the estmator M a j = N j 2003,2002 where M j2003 s the number of nhabtants n neghbourhood j n 2003 who were n neghbourhood n 2002, and N,2002 s the number of nhabtants n neghbourhood n 2002 We have estmated transton probabltes separately for natves, non-western, and western mgrants Table 2 presents the estmates of a matrx of transton probabltes The last two columns have been added to reveal the dstrbuton of the populaton and the propensty to move over the neghbourhood type The second column from the rght shows that the propensty to move among natves ncreases monotoncally wth the share of non-western mgrants n the neghbourhood The propensty to move s slghtly lower when the share of non-western mgrants s less than 21 percent A smlar pattern holds for western mgrants On the other hand, for non-western mgrants the propensty to move s almost constant at around 14 percent across the neghbour types Ethnc sortng becomes more vsble when we look at the destnaton-neghbourhood type of movers The entres n the rows of the table show the redstrbuton of ndvduals lvng n neghbourhood n 2002 over the neghbourhood types one year later The dagonal of the matrxes shows the preservaton n the same neghbourhood types for natves and mmgrant groups The dstrbuton above the dagonal shows the dsperson of populaton towards locatons wth more mgrants, whle the dstrbuton below the dagonal shows dsperson n the opposte drecton Note that also not movers are ncluded n the calculatons, n addton to movers, whereby the change n the neghbourhood ethnc composton looks lke small It s mmedately apparent that natves are manly dffused over the neghbourhoods wth a lower percentage of non-western mgrants (that s, there s more dsperson below the dagonal), whle the dstrbuton of mgrants especally non-western mgrants s more promnent above the dagonal Natves resdng n the neghbourhoods where less than 10 percent of the populaton s non-natve are less nclned to move If they do, ther destnaton s more lkely to be a low (mmgrant) concentraton area When they resde n neghbourhoods wth more non-western neghbours, ther destnaton s often a less concentrated neghbourhood 8

11 9 Table 2 Transton probabltes and the dstrbuton of movers over neghbourhood types referrng to the share of non-western mgrants n the neghbourhood % non-western n the % non-western n the destnaton neghbourhood (2003) orgn neghbourhood 0-1% 11-20% 21-30% 31-40% 41-50% 51-60% % % Movers Total Natves 0-10% % % % % % % Total Non-western 0-10% % % % % % % Total Western 0-10% % % % % % % Total Note ths table s based on the entre populaton of the Netherlands 5 Analytcal strategy We frst consder the subset of movers alone and regress the natve concentraton onto the mover s destnaton, y on a vector of some ndvdual demographc and economc varables, W and a vector of the mmgrant groups dstngushed, y = W δ + I η + µ The parameters of ths model are estmated by an ordnary least square (OLS) estmator neglectng the possble selectvty of movers To assess the mpact of varable groups, four models were estmated, startng wth a smple model wth the most exogenous varables Subsequently, the model was extended wth varables reflectng the demographc, household, and socoeconomc poston of ndvduals, the neghbourhood ethnc composton, and the three largest ctes Standard errors have been adjusted at the muncpalty level, snce neghbourhoods are nested n muncpaltes The estmates are presented n table 3 All varables refer to the frst year pror to movng (2002), except those ndcatng the change n martal status (from sngle to marred and from couple to sngle) The analyss shows that the natve concentraton n the destnaton neghbourhood s 12 to 21 ponts lower for non-western mmgrant groups (that s, the Turks, Moroccans, Surnamese, Antlleans and other non-western groups) after controllng for all background varables, except the percentage of non-western mgrants n the orgn neghbourhood and the three largest ctes Includng ths last varable has a substantal nfluence on the mmgrant groups coeffcents; they are almost 50 percent lower Ths suggests that almost half the segregatory behavour of non-western mgrants can be explaned by the fact that I 9

12 10 they were lvng n segregated areas An alternatve explanaton s that mgrants, especally nonwestern mgrants lvng n segregated areas, are less lkely to move to natve neghbourhoods Table 3 Regresson estmates of the destnaton-neghbourhood type of movers (Dependent varable s percentage of natves n the destnaton neghbourhood) Varable Model I Model II Model III Model IV Woman 0787 * 0590 * 0920 ** 0967 ** Age *** *** * ** Age Age Age Moroccan *** *** *** *** Turksh *** *** *** *** Surnamese *** *** *** *** Antllean *** *** *** *** Other Non-western *** *** *** *** Western *** *** *** *** Second generaton 5540 *** 5244 *** 5062 *** 3427 *** Homeowner 3954 *** 3853 *** 1111 *** Marred 2909 *** 2576 *** 2335 *** Number of chldren n household * * * Number of persons n household 1183 *** 1084 *** 0609 ** From sngle to marred 4303 *** 3493 *** 3242 *** From couple to sngle *** ** *** Employed 2098 ** 2418 *** Unemployed * 0030 Recever of a beneft Student * * Labour ncome Q1-Q Labour ncome Q2-Q Labour ncome >Q %non-western n the orgn neghbourh *** Amsterdam * Rotterdam * The Hague *** Constant *** *** *** *** R-sq N * p<05; ** p<01; *** p<001 Standard errors are clustered on the muncpalty level However, OLS estmates wll not provde consstent and asymptotcally effcent parameter estmates f the decson to move s correlated wth the choce of destnaton-neghbourhood type Intutvely and theoretcally, there are two reasons why these two decsons could be nterrelated Frst, ndvduals may decde to move when they ntend to sort themselves by ethnc background Second, more fundamentally, an ndvdual facng a choce set wll potentally make a nonrandom selecton Agan, ths suggests the selectvty of movers The decson to move and the choce of destnaton-neghbourhood type are both choces Preferences bounded by household ncome and housng market condtons are lkely to play an mportant part n generatng selectve 10

13 11 moblty Ths lne of argumentaton has led us to estmate a selecton model whch allows these decsons to be correlated (Heckman, 1979; Wooldrdge, 2002) 51 Selecton Model The patterns of resdental moblty of both natves and ethnc mnortes reflect the process of ethnc resdental segregaton In ths process, the movers choce of destnaton-neghbourhood type s central to ths study Snce the decson to move s also a choce gven all relevant restrctons on the moblty, movers are lkely to be nclned to choose a certan neghbourhood type The selectvty of movers wll dffer strongly across ethnc groups f an ethnc sortng process s n operaton We consder the selectvty of movers n examnng the choce of the destnaton-neghbourhood type n terms of the percentage of natves The locaton choce of an ndvdual household follows a two-equaton selecton model The destnaton of the move s gven by a lnear model (the outcome equaton), whle the nclnaton to move s determned by a probt model (selecton equaton) Let the choce of destnaton locaton and the decson to move for ndvdual be expressed by the followng functons: The outcome equaton: P = β + ε (1) X * The selecton equaton: M = Z γ + u (2) where P s the percentage of natves n the destnaton neghbourhood, X s a vector of ndvdual * characterstcs determnng the locaton choce and the mmgrant poston, M s a latent varable ndcatng the propensty to move that s observed as a bnary varable takng the value 1 f moved and 0 f not moved, Z s a vector of determnants of movng and the mmgrant poston, β and γ are the assocated parameter vectors to be estmated, andε and u are error terms that follow a bvarate normal dstrbuton 0,0, σ ε, σ u, ρ ( ) The probablty of movng s represented by a probt model where Pr Φ() * ( M f 0) = Pr( u f Z ) = Φ( Z ) γ γ s the standard normal cumulatve dstrbuton functon The destnaton locaton type s only observed for those for whom s determned as: E M * ( P M f 0) = X β + E( ε u f Z γ ) * f 0, so that the expected neghbourhood type of movers = X β + θλ (4) where ρσ λ φ Z γ Φ Z γ φ s the standard normal densty functon The locaton choce functon n the presence of sample selectvty to be estmated s expressed as θ =, = ( ) ( ) and ( ) (3) 11

14 12 E ( P M 0) = X β + θλ + * f υ (5) where υ s the error term The parameters of (1) and (2) can be estmated by the maxmum lkelhood (MLE) method allowng the correlaton of the error terms 1 (Wooldrdge, 2002) The selecton model can be successfully estmated, because we have a number of explanatory varables potentally affectng the nclnaton to move, but not necessarly the destnatonneghbourhood type (Sartor, 2003) (see table 4) These varables are the neghbour characterstcs measurng the economc and housng market condtons n the neghbourhood such as the mean value of houses, ncome level, the share of owner-occuped homes, together wth the presence of parent(s) to approxmate the local attachment Ths large number of varables s certanly a good bass to estmate the proposed selecton models, compared to prevous applcatons of selecton models, although not all of these varables are supposed to be exogenous to the share of mgrants n the destnaton neghborhoods The sgn of the selecton coeffcent s expected to be postve f those who are more lkely to choose natve neghbourhoods have a hgher probablty of movng It wll be negatve when movers to mmgrant neghbourhoods are more lkely to move For the pooled sample, the sgn of the selecton coeffcent s hard to predct, because natves and mgrants may have dfferent motves for movng and dfferent preferences for the destnaton-neghbourhood type 52 Emprcal Results Frst, we estmated the parameters of the outcome and selecton models ncludng a varous set of varables usng the pooled sample Table 4 dsplays the estmates of the outcome equaton and correspondng selecton equaton, snce both models are mportant The selecton equaton s, n fact, a probt model estmatng the probablty of movng The frst model ncludes the ndvdual characterstcs n the outcome equaton that are strctly exogenous and, addtonally, some neghbourhood characterstcs n the selecton equaton to ensure model dentfcaton The followng models utlse varables for household stuaton, socoeconomc poston, the concentraton of non-western people n orgn neghbourhoods, and controls for the three largest ctes In ths way, the nfluence of varable groups can be demonstrated The selecton correcton term, lambda, s sgnfcantly postve n the models I-IV, whch suggests a postve selecton by movers Those who tend to move to neghbourhoods wth a hgher share of natves are more lkely to move Note that ths postve selecton does not necessarly hold for all groups When the same model s estmated separately for natves and mmgrant groups, the sgn of the selecton coeffcent s postve for natves and western mgrants, but statstcally not sgnfcant for nonwestern mgrants, as dscussed n the next secton (see table A1 n appendx) Concentratng frst on the estmates of the outcome equaton, t s mmedately apparent that nonwestern mgrants, especally Surnamese and Moroccans are less lkely to move to neghbourhoods wth a hgher share of natve populaton than natves, the reference group Model 1 Nawata and Nagase (1996) suggests that the MLE s asymptotcally much more effcent than Heckman s two step estmator, especally when there s a hgh degree of multcollnearty between the hazard rato and explanatory varables We have not encountered the convergence problem often mentoned n the lterature as a dsadvantage of the MLE method 12

15 13 I ndcates that the percentage of natves n the destnaton neghbourhood of Surnamese and Moroccans s 2329 and 2277 ponts lower than that n the destnaton neghbourhood of natve movers when only age and gender controls are used After controllng for the concentraton of non-western mgrants n the orgn locatons, ther estmates declne substantally, but reman sgnfcantly hgh at about 104 and 118, as shown by model IV Those who lved n ethncallysegregated neghbourhoods are more lkely to move to neghbourhoods wth a lower share of natves Ths last model ndcates a comparable moblty pattern for Turksh and other nonwestern mgrants The natve-mmgrant dfference n the natve concentraton n destnaton neghbourhoods s the smallest for western mgrants by almost 6 ponts On the other hand, the second generaton and those who marred are more lkely to move neghbourhoods wth a hgher share of natves Ths also holds for employed people The effect of gender and age on the locaton choce seems to be neglgble The estmates of the selecton equaton gve the probablty of movng After controllng for observed ndvdual and neghbourhood characterstcs, only other non-western mgrants have a hgher probablty of movng No sgnfcant dfference was found between natves and other mmgrant groups The hgher moblty among non-western mgrants s possbly the result of ther relatvely short stay n the Netherlands The category of non-western mgrants contans a substantal share of asylum mgrants and other recent mgrants (Zorlu and Mulder, (forthcomng)) Hgher observed moblty among other ethnc mnorty groups s ascrbable to ther young age structure and household status n comparson wth natves In lne wth the lterature, larger households wth a hgher degree of local attachment and homeowners are less lkely to move, whle a change n martal status s strongly assocated wth moblty So, the presence of (a) parent(s) n the same muncpalty has a sgnfcant negatve effect on the probablty of movng Indvduals who lve n a better home n the neghbourhood, gven a hgher value of ther own home, are less lkely to move Notably, the moblty behavour seems to be negatvely nfluenced by the concentraton of all non-western mmgrant groups n the orgn neghbourhood Table 4 Estmates of the choce of destnaton neghbourhood wth correcton for selecton Model I Model II Model III Model IV Outcome equaton (dependent varable s percentage of natves n the destnaton neghbourhood) Age *** 8319 *** 8597 *** 2620 * Age *** 7173 *** 6185 *** 2461 * Age ** 2806 * Age Woman * 1017 ** Moroccan *** *** *** *** Turksh *** *** *** *** Surnamese *** *** *** *** Antllean *** *** *** *** Other Non-western *** *** *** *** Western *** *** *** *** Second generaton 5243 *** 4817 *** 4580 *** 3247 *** Homeowner 2717 * 2536 * 0427 Marred Number of chldren n household *** ** ** Number of persons n household 1186 *** 1066 *** 0650 ** From sngle to marry 7783 *** 6956 *** 4825 *** 13

16 14 From couple to sngle 9496 *** 9480 *** 2473 Employed 2329 ** 2547 *** Unemployed * Recever of a beneft Student * Labour ncome Q1-Q Labour ncome Q2-Q Labour ncome >Q * %non-western n the orgn neghbourh *** Amsterdam * Rotterdam The Hague ** Constant *** *** *** *** Selecton equaton (Dependent varable s the propensty to move) Age *** 0836 *** 0849 *** 0844 *** Age *** 0593 *** 0581 *** 0582 *** Age *** 0274 *** 0262 *** 0263 *** Age ** 0087 ** 0078 ** 0077 ** Woman Moroccan ** * * 0036 Turksh * * * 0012 Surnamese *** ** ** 0008 Antllean Other Non-western * 0084 ** 0175 *** Western Second generaton Homeowner *** ** ** *** Log value of home n 1000s *** *** *** *** Marred *** *** *** *** Number of chldren n household * Number of persons n household From sngle to marry 0210 *** 0348 *** 0337 *** 0336 *** From couple to sngle 0875 *** 1062 *** 1066 *** 1081 *** Employed Unemployed ** * Recever of a beneft Student Labour ncome Q * * * * Labour ncome Q Labour ncome Q Log mean value of houses n negh % owner occuped houses n negh Unemployment rate n negbourhood 0014 ** 0014 ** 0014 ** 0015 *** Log mean labour ncome n negh 0398 *** 0407 *** 0406 *** 0353 *** %non-western n the orgn negh * * * * Both parents n the same muncpalty ** ** ** ** One parent n the same muncpalty No parent n the same muncpalty 0006 *** 0006 *** 0006 *** 0001 Amsterdam Rotterdam The Hague Constant *** *** *** *** rho 0744 *** 0753 *** 0747 *** 0470 *** lambda *** *** *** 6755 *** N

17 15 N-Censored * p<05; ** p<01; *** p<001 Standard errors are clustered on the muncpalty level 53 Decomposng dfferental The estmates of lnear models show that the types of destnaton neghbourhoods dffer substantally between natve and mmgrant groups Sgnfcant dfferences n the age dstrbuton, household, and socoeconomc characterstcs of natves and mmgrant groups explan some part of ethnc moblty dfferentals, but much of the dfferentals remans unexplaned by the factors observed In order to uncover the explaned and unexplaned parts of moblty dfferences between natves and separate ethnc groups, we adopted the Oaxaca-Blnder lnear decomposton technque (Oaxaca, 1973; Blnder, 1973) usng the separate regresson models for each group We clustered the mgrants nto two groups: western and non-western Ths technque s wdely used n the economcs lterature to decompose wage and employment dfferentals between varous ethnc groups or between men and women nto two components: dfferentals attrbuted to dfferences n endowments or observed characterstcs, and an unexplaned component, whch may be supposed to capture labour market dscrmnaton (Oaxaca,1973; Blnder, 1973; Catton, 1988; Neumark, 1988, Oaxaca & Ransom, 1994; Neuman & Oaxaca, 2004) In ths study, we appled ths decomposton technque to assess the natve-mmgrant dfferentals n destnaton-neghbourhood types that cannot be explaned by the characterstcs of the ethnc groups n the models Ths unexplaned part may be attrbuted to two man groups of factors Frst, the majorty populaton can hnder the spatal assmlaton of ethnc mnorty groups For nstance, dscrmnaton n the housng market, both for renters and buyers, and other nsttutonal factors can hamper the moblty of ethnc mnorty groups towards places wth a hgher share of natves Second, ethnc mnorty groups own preferences may be a reason for remanng n an mmgrant neghbourhood These two components of unexplaned dfferentals are hard to dstngush n the admnstratve data wthout addtonal nformaton about atttudes However, ths unexplaned part of the choce behavour mght reflect the preferences of natves and ethnc mnortes to lve n neghbourhoods wth a hgher share of ther own ethnc group, and dscrmnaton The absence of sound emprcal evdence on dscrmnaton hnts at the dffculty of assessng dscrmnaton n the housng market, snce n the Netherlands dscrmnaton s lkely to be subtle Ths contenton also supports our mplct strategy to assess the unexplaned component of moblty After controllng for relevant demographc and socoeconomc characterstcs, resdng closer to co-ethncs may ndcate mgrants preferences shaped by comparatve advantages of ethnc good and networks 15

18 16 Accordng to the Oaxaca-Blnder technque, the observed mean dfferental of the percentage of natves n the destnaton neghbourhood, Pn Pm, s decomposed nto two components by the followng equaton 2 : n m n ( X X ) + X ( ˆ β ˆ β ) P P = ˆ β (6) n m m n m where the subscrpts n and m denote natves and mmgrant groups, clustered nto two groups as Surnamese/Antlleans and Turks/Moroccans n vew of the smlartes between the groups X n and X m are the mean values for the regressors; ˆn β and βˆm are the assocated coeffcents The frst term on the rght hand sde, βˆ n ( X n X m ), s a dfferental owng to the characterstcs X β * ˆ β gves unexplaned dfferentals and the second term ( ) m m In the orgnal form of the decomposton, Oaxaca (1973) proposes ether a male or female wage structure as the non-dscrmnatory wage structure Later studes suggest a non-dscrmnatory * wage structure β be estmated, so that (6) becomes n m ( X n X m ) [ X ( ˆ * + n β n β ) + X m ( β ˆ β m )] ˆ * n P P = β (7) where the estmated non-dscrmnatory structure s gven as ( Ω) ˆ m * β = Ω ˆ β n + I β Catton [5] suggests a weghtng matrx ( Ω ) reflectng the share of the majorty group n the sample (I n ), Ω = I n I Neumark (1988) proposes a least-squares crteron to estmate a weghtng * 1 matrx from the pooled sample of all the groups dstngushed, β = ( X X ) ( X P) = ˆ β, where X s the observaton matrx, P s the observaton vector of the response varable and βˆ s the OLS estmate obtaned from the pooled sample Neumark (1988) and Oaxaca & Ransom (1994) show that the extent of the unexplaned part s senstve to the choce of a non-dscrmnatory structure In our case, the unexplaned part of locaton-choce dfferentals would cover preferences as ndcated Hence, an ndfferent preference structure may be equvalent to a non-dscrmnatory wage structure There s, however no unambguous crteron to defne an ndfferent-preference structure We therefore calculated alternatve decompostons usng three weghtng matrxes: n the frst and second cases, an ndfferent-preference structure s assumed to be equal to the preference structure of natves and mmgrant groups respectvely In other words, the locaton choce of the majorty group (natves) or mmgrant groups prevals Alternatvely, as Neumark suggests (1988), a common-preference structure derved from the pooled sample of natves and mgrants s assumed to represent an ndfferent-preference structure Ths approach mples that natves and ethnc mnortes 2 Ths decomposton formula does not nclude a separate selecton component, snce we have estmated MLE selecton models rather than Heckman two-step models In ths case, any possble selectvty effect has already been captured by the estmated coeffcents 16

19 17 contrbute to an ndfferent-preference structure accordng to ther weghted share n the entre populaton 54 Results To decompose ethnc dfferentals n the destnaton-neghbourhood types accordng to the share of natves, we frst estmated selecton models wth dentcal varables for natves, western, and non-western mgrants usng the MLE estmator wth selectvty correcton The Parameter estmates for both models are dsplayed n Table A1 n the appendx A lkelhood-rato test ndcates that the null hypothess of a zero correlaton between the resduals of locaton choce equaton and the selecton equaton (ρ = 0) can be rejected for the selecton models for natves and western mgrants, wth χ 2 =3961 for natves and χ 2 =775 for western mgrants The ndependence of the two equatons cannot, however, be rejected for non-western mgrants, wth χ 2 =066 Ths result suggests that the MLE estmaton of the selecton model provdes consstent, asymptotcally effcent estmates for all the parameters n the models for natves and western mgrants, but for non-western mgrants OLS provdes unbased parameter estmates We therefore estmated the locaton-choce equaton usng the OLS estmator not only for nonwestern mgrants, but also for natves and western mgrants, to show the consequences of gnorng selectvty The sgn of the selecton coeffcent lambda for natves and mgrants s nterestng As expected, t s postve and statstcally sgnfcant for natves and western mgrants, whch ndcates the postve selectvty of natve movers: those who tend to move to neghbourhoods wth a large proporton of natves have a hgher probablty of movng Ths result confrms the partcular role of the moblty of the domnant group n the sortng process (Clark, 1991) The lambda coeffcent s not statstcally sgnfcant for non-western mgrants Usng MLE and OLS estmates, we calculated alternatve decompostons for the dfferentals between natve and non-western mgrants, and between natve and western mgrants, based on the same type estmator Subsequently, relyng on the above nformaton about the consstent and effcent estmates for natves and non-western mgrants, we calculated the decompostons for these groups usng the MLE estmates for natves and the OLS estmates for non-western mgrants Hence, the dscrepancy between the results from ths last estmaton and the results obtaned from the OLS or MLE estmates wth selectvty correcton wll mplctly reflect the dfferental resultng from the selecton of natve movers The results of the decomposton analyses are presented n table 5 As mentoned above, three weghtng matrxes are used to calculate explaned and unexplaned parts of the dfferentals n the natve share of the destnaton neghbourhood The frst column n table 5 refers to the weghtng matrx appled The frst and second decompostons are obtaned from the weghtng matrx assumng that the majorty and mgrants respectvely are the reference The thrd decomposton uses the pooled sample as the reference 17

20 18 Table 5 Decomposton of natve-mmgrant dfferentals n the concentraton of natves n the destnaton neghbourhood of movers Natve non-western Natve western OLS MLE MIXED OLS MLE Reference Decomposton Dff SE Dff SE Dff SE Dff SE Dff SE Explaned By %n-west Natve Unexplaned Explaned By %n-west Immgrant Unexplaned Explaned By %n-west Pooled Unexplaned Predcted Dfference The fgures n bold are sgnfcant at the 05 level Consder frst the estmated dfferentals between natves and non-western mgrants The predcted mean dfferentals obtaned from OLS and the selecton models are farly small: 1865 and 1677 ponts respectvely These predctons are very close to the observed dfferental as shown n table 1 The dstrbuton of the dfferental across explaned and unexplaned parts vares slghtly across the type of weghtng matrx appled The contrbuton of observed characterstcs rses slghtly when the mmgrant group s taken as the reference, whereas the unexplaned part s relatvely hgher when natves are the reference Snce an ndfferentpreference structure s derved from the pooled sample, the proporton of the explaned part s even greater than these two extreme cases In ths case, 62 to 65 percent of natve mmgrant moblty dfferences (about 11 percentage ponts) s explaned by observed characterstcs The remanng 35 percent s lkely to be upper bound for the dfference owng to preferences Ths last approach that reles on the composton of current socety as whole rather than takng natves or mgrants as reference, probably gves a more realstc pcture Ths result s smlar to the outcome for Asans (53 percent) n the study of Bayer and colleagues (2004), who used a dfferent strategy from that appled n ths study They could explan 925 percent of the segregaton for Hspancs and approxmately 30 percent for Whtes and Blacks The results obtaned from OLS and selectvty-corrected MLE estmators are small, as mentoned above However, the dstrbuton of the dfferental across explaned and unexplaned s much more pronounced when the decompostons are based on the selectvty-corrected MLE model for natves and OLS estmator for non-western mgrants, accordng to the crtera of unbased parameter estmates as dscussed above Ths last method provdes a lower dfferental than the other estmates do: 796 versus 1865 and 1677 percentage ponts The entre dfferental s derved from observed characterstcs when the ndfferent-preference structure s supposed to be derved from the pooled sample Comparng ths result wth the decomposton based on the OLS estmator may suggest that a decomposton-based OLS estmator overestmates the natve mmgrant dfference by about 10 percentage ponts and the relatve share of unexplaned dfferental when the selectvty bas of natve movers s not consdered These results also mply that ethnc resdental segregaton s manly derved from the selectve moblty behavour of natves Movers among non-western mgrants are not self-selected consderng the ethnc composton of neghbourhoods, nether postvely nor negatvely 18