LONGITUDINAL ANALYSIS OF EARNINGS ASSIMILATION AMONG IMMIGRANTS IN QUEBEC

Transcription

1 1 LONGITUDINAL ANALYSIS OF EARNINGS ASSIMILATION AMONG IMMIGRANTS IN QUEBEC Michael Beensock Deparmen of Economics Hebrew Universiy of Jerusalem Absrac A cohor of a 1000 immigrans who enered Quebec in 1989 was followed-up during he subsequen 10 years. We find no evidence of immigran wage assimilaion in his cohor. Our resuls join a small bu growing body of research ha shows ha here is no evidence of immigran assimilaion in longiudinal daa. These negaive findings sand in marked conras o corroboraion of assimilaion heory in numerous synheic cohor sudies. Our resuls also show ha for his cohor, he immigran labor marke is segmened from he naive labor marke. March 9, 2006 We wish o hank Pavel Soifer for able and congenial research assisance.

2 2 LONGITUDINAL ANALYSIS OF EARNINGS ASSIMILATION AMONG IMMIGRANTS IN QUEBEC 1. Inroducion 1.1 Synheic Cohors and Cross Secion Daa Empirical research concerning he absorpion of immigrans ino he labor marke of he hos counry has been hampered by lack of longiudinal daa. Chiswick's (1978) pioneering sudy used cross-secion raher han longiudinal daa o show ha immigran earnings vary direcly wih ime since migraion (TSM). He inerpreed his observaion as being consisen wih he hypohesis ha immigran human capial is no compleely porable, and ha i akes ime for immigrans o adjus o heir new milieu. Duraion maers because i akes ime o acquire language skills (Chiswick and Miller 1995) and o learn abou labor marke neworks and insiuions in he hos counry. I has long been recognized (Chiswick 1980, Borjas 1985) ha here are problems wih using saic, cross-secion daa o invesigae dynamic phenomena such as immigran assimilaion. Cross-secional analyses compare differen individuals observed a he same poin of ime, under he assumpion ha apar from heir differen duraions immigrans are oherwise similar. If, however, hey are no similar, dynamic inferences made from saic cross-secions will be misleading. Specifically, if more recen immigran cohors have less abiliy, hen Chiswick's observaion may be consisen wih deerioraing cohor qualiy. More recen immigrans earn less no because hey have had less ime o assimilae, bu because hey belong o less able immigran cohors. Alernaively, if less successful immigrans re-emigrae (Beensock 1996), Chiswick's observaion may simply mean ha more recen immigrans earn less because hey have had less opporuniy o re-emigrae. In he absence of genuine longiudinal daa, Borjas (1985) suggesed he "synheic cohor mehodology", which uses wo or more independen cross-secions, o idenify he effec of duraion on immigran earnings. A comparison of wo immigrans wih idenical duraions, who immigraed a differen imes, idenifies cohor effecs and duraion effecs. The synheic cohor mehodology has been widely

3 3 applied in Canada 1 as well as in many oher immigran receiving counries. Indeed, here has been a deerioraion in he iniial earnings of immigran in Canada, which Aydemir and Skuerud (2005) parially aribue o language abiliy and region of origin. While raes of assimilaion may no have been srong enough for immigran earnings o converge o he earnings of demographically similar Canadian naives, hey provide clear evidence in favor of he immigran assimilaion hypohesis insofar as immigran earnings grow more rapidly in he years following immigraion and grow less rapidly hereafer. See also resuls repored by Green and Worswick (2002) using a differen source of daa on Canadian immigrans and he lieraure review by Hum and Simpson (2004a). Synheic cohors canno solve he problem of selecive re-emigraion, and hey are sensiive o measuremen error on he unobserved characerisics ha define he cohor. Suppose, for example, ha synheic cohor daa show ha immigran earnings display posiive duraion dependence. Suppose also ha he immigran labor marke is Darwinian in he sense ha fier workers survive longer in beer-paid jobs. The observed posiive duraion dependence may be jus a saisical arifac induced by survivorship bias. The apparen assimilaion effec may simply resul from he survival in he labor force of fier workers 2. The same applies if remigrans are negaively seleced. If fier workers do no remigrae, synheic cohor daa will creae he misleading impression ha earnings vary direcly wih duraion. In shor, synheic cohor daa may fail o idenify he assimilaion effec. 1.2 Longiudinal Evidence on Immigran Assimilaion The obvious soluion o his idenificaion problem is o use genuine cohor daa, or longiudinal daa, raher han synheic cohor daa. By observing he same immigrans over ime belonging o he same immigran cohor, here can be no disagreemen abou he source of he phenomena observed in he daa. We are comparing he same immigrans a differen poins in ime raher han differen immigrans a differen poins in ime. Here oo sample ariion induced by selecive re-emigraion is a problem, bu a leas longiudinal daa are informaive abou he ariion process. 1 For example, Baker and Benjamin (1994), Bloom, Grenier and Gunderson (1995), Chiswick and Miller (1996, 2004), Gran (1999), Schaafsma and Sweeman (2001), Warman and Woswick (2004) and Aydemir and Skuerud (2005) who use census daa for Canada for various years. 2 Survivor bias may also apply o naive workers, especially he younger ones. However, survivor bias is likely o be more pronounced among immigrans whose finess has ye o be proved.

4 4 Unforunaely, longiudinal daases on immigrans are scarce 3. Borjas (1989) invesigaed he earnings dynamics of 1,166 immigran scieniss and engineers in he US over he period He found ha here was no evidence of assimilaion. I is possible ha by 1974 Borjas' immigrans had already compleed heir assimilaion process 4, in which case here would be lile or nohing o be learned abou assimilaion by observing heir earnings during In addiion his resuls may suffer from selecive movemen ino and ou of scienific occupaions since he earnings of Borjas' immigrans are only observed if hey worked as scieniss or engineers. Duleep and Reges (1997) used longiudinal daa for he Unied Saes o show ha beween he earnings growh of 351 immigrans idenified in he roaing panels of he CPS was abou 2% greaer han for comparable naives, which seems o suppor he assimilaion hypohesis. However, hey did no have daa on he duraion of hese immigrans in he Unied Saes, so hey could no es wheher he earnings growh of immigrans during varied inversely wih duraion as prediced by assimilaion heory. Hu (2000) mached longiudinal earnings daa from SSA (Social Securiy Adminisraion) wih he Healh Reiremen Sudy in he US over he period He found ha whereas he synheic cohor mehodology apparenly corroboraes assimilaion heory, his resul is overurned by he longiudinal daa. On he oher hand, Duleep and Dowham (2002) mached he same longiudinal daa o he CPS (bu over a shorer ime period) o show ha he immigran assimilaion hypohesis is corroboraed by he longiudinal daa. They aribue heir difference wih Hu o he fac ha he did no allow he saring wage o vary by cohor. Panel daa covering immigran earnings during he firs hree years in Ausralia (Chiswick, Lee and Miller 2005) and Israel (Beensock and Ben Menahem 1997) corroborae he assimilaion hypohesis over he shor-erm. Beensock, Chiswick and Paliel (2005), however, failed o find evidence of long-erm (12 years) assimilaion in a large sample of immigrans in Israel. Therefore, evidence of shor-erm assimilaion may no corroborae assimilaion heory over a longer ime frame. In his paper, we use a longiudinal daabase ha covering he firs decade in he hos counry, and which sheds ligh on immigran assimilaion over he shor and longer erms. 3 Longiudinal daa sources on immigrans have usefully been reviewed by Black e al (2003). 4 Borjas repors ha in 1974 immigrans' average sojourn in he US was as high as 19.2 years.

5 5 The Immigran Daabase (IMDB) available a Ciizenship and Immigraion Canada (CIC) conains longiudinal daa on ax reurns of immigrans mached wih immigran landing files. However, he panel naure of hese daa has no been aken ino consideraion in he invesigaion of immigran earnings 5. Anoher source of panel daa on Canadian immigrans is he Survey of Labour and Income Dynamics (SLID) which has been used by Hum and Simpson (2000, 2004b) o show ha during he 1990s immigran earnings did no grow faser han naive earnings. I is noeworhy ha hese resuls, which rejec he immigran assimilaion hypohesis, break he empirical consensus in Canada 6, which has been exclusively informed by he large body of evidence accumulaed from applicaions of he synheic cohor mehodology o a number of independen daa sources including census daa, SLID and IMDB. 1.3 Panel Daa for Quebec In his paper we invesigae he assimilaion hypohesis using he ENI (Éablissemen des Nouveaux Immigrans) daabase, which comprises longiudinal daa for a single immigran cohor o Quebec. Since immigrans o Quebec are seleced on a differen basis o Canadian immigrans as a whole 7, he populaion sudy group is specific o Quebec. These daa provide a rare opporuniy o invesigae he assimilaion hypohesis during he firs decade in he hos counry, which happen o be he crucial years for invesigaing immigran assimilaion. Moreover, he daa refer o a single immigran cohor. Maers migh have been more complicaed had our longiudinal daa referred o differen immigran cohors since, as suggesed e.g. by Aydemir and Skuerud (2005), assimilaion may depend upon when immigraion occurred, especially regarding he sae of he business cycle. The main quesion we ask is: given everyhing else, including personal characerisics and he sae of he labor marke in Quebec, do ENI immigran earnings vary direcly wih ime since migraion, and is his relaionship concave as prediced by assimilaion heory? By following he same 5 For example, Green and Worswick (2002) esimae earnings funcions using IMDB ha ignore he panel dimension o he daa. Noe ha IMDB is only available o a small number of auhorized users a CIC, hence we do no use i here. Nor do we use LSIC (Longiudinal Survey of Immigrans o Canada) since i was only sared in April Hum and Simpson (2000) do no conrol for duraion in Canada in 1993, so hey do no properly es he immigran assimilaion hypohesis. The wage growh of immigrans wih shorer duraions should have been greaer han hose wih longer duraions. 7 Quebec is he only province ha has an independen immigraion policy and gives prioriy o francophone immigrans.

6 6 immigrans over ime belonging o he same immigran cohor, a posiive and concave relaionship beween earnings and TSM would serve as direc empirical corroboraion of assimilaion heory. 2. Economeric Mehodology 2.1 The Model Assimilaion heory predics ha immigran earnings converge upon he earnings of demographically comparable naives. Le W i denoe he logarihm of real earnings of immigran i (i = 1,2,..,N) in ime period (= 1,2,,T), le X i denoe a vecor of personal characerisics (such as age, educaion ec) measured a he ime of immigraion ( = 0), and le V i denoe he log earnings of comparable naives, i.e. condiioned on X i. Dropping subscrip i and expressing ime coninuously, we wrie he assimilaion model as: W & = V & λ( W V ) (1) V = v + βx + φ (2) Equaion (1) saes ha when he immigran-naive wage gap is zero, immigran wages grow a he same rae as he wages of comparable naive. If however, he wage gap is negaive, immigran earnings grow faser han naive earnings. Equaion (2) saes ha naive earnings grow a a common rae φ, and he wage level depends upon X. For example, more educaed workers earn more. The general soluion o equaions (1) and (2) is: λ W = Ae + v + βx + φ (3) where he iniial condiion A = W 0 V 0 < 0 is equal o he log wage gap on immigraion. Equaion (3) saes ha immigran wages converge on naive wages from below. Empirical evidence suggess (Aydemir and Skuerud, 2005) ha he iniial wage gap may depend upon observed heerogeneiy measured by X. We herefore le A = A 0 + ax. Subsiuing his ino equaion (3) and resoring subscrip i, equaion (3) may be reparameerized in erms of average naive earnings: λ W i = ( A0 + ax i ) e + βx i + V (4) Equaion (4) saes ha immigran earnings vary direcly wih ime since migraion, hey depend upon X in erms of levels and raes of change, and hey are proporionae o average naive earnings.

7 7 I may be shown ha varians of equaion (4) are obained if he assimilaion process in equaion (1) is no firs order. For example, if i is second order here will be wo roos insead of one, and he convergence process may be quadraic, as originally suggesed by Chiswick (1978). Also, equaion (2) is unlikely o be saic, and wages are likely o be influenced by he business cycle. Indeed, as suggesed by Aydemir and Skuerud (2005), he business cycle migh affec naive and immigran wages differenially, in which case i would be appropriae o specify he sae of he business cycle in equaion (4). In his conex here are wo quie differen hypoheses. In he firs, wha maers is he sae of he business cycle on enering he labor marke. Aydemir and Skuerud sugges ha he long erm prospecs of naives and immigrans are adversely affeced if hey happened o ener he labor marke during a recession. The second hypohesis predics ha he business cycle affecs he relaive pay of immigrans as hey assimilae, i.e. afer hey have enered he labor marke. Since ENI immigrans belong o a single cohor and enered he labor marke a abou he same ime, i is impossible o invesigae he former effec. Therefore, we only invesigae he laer effec. 2.2 Panel Daa Analysis The assimilaion hypohesis predics ha given X, V and he sae of he business cycle (Z) here is a concave relaionship beween earnings and ime since migraion (TSM). In equaion (5) we use Chiswick's original quadraic specificaion: W i = α + X β + θz + φv + γtsm ηtsm + ε i i i Equaion (5) is a one-way specific effecs model where α i denoes he individual specific effec. The α's capure immigran heerogeneiy; immigrans who assimilae more successfully have larger α. We do no specify ime specific effecs because he number of rounds in he panel is small, because TSM naurally varies direcly wih ime, and because equaion (5) includes ime dependen conrols (Z and V). Given everyhing else, we expec a posiive relaionship beween immigran earnings and naive earnings. Indeed, he model in Secion 2.1 predics ha φ = 1. If he immigran wage gap is anicyclical we expec θ < 0. Following equaion (4) TSM may be ineraced wih X. We include sandard demographic conrols in X, including educaion, age and is square, and region of origin. The role of age here serves a double funcion because i no only reflecs wages over he life-cycle, i also capures he affec of age a 2 i i (5)

8 8 immigraion. Schaafsma and Sweeman (2001) poin ou ha older immigrans find i more difficul o assimilae, so given everyhing else, older immigrans in ENI are likely o earn less. On he oher hand, older immigrans are likely o earn more because hey are more experienced. Therefore, he ne effec of ageing is unclear. We prefer specifying random effecs raher han fixed effecs on a priori and pracical grounds. Fixed effecs are more suiable when he daa refer o populaions, and random effecs are more suiable when he daa refer o random samples 8. ENI is a sample of immigrans who were randomly observed on firs enering Quebec in 1989, herefore heir specific effecs are a random sample of he specific effecs in he populaion. In any case, Hsiao (2003) recommends using random effecs esimaors in shor panels, since esimaes of fixed effecs are biased when T is small. Finally, ε is a residual error assumed o be normally disribued wih consan variance. I is hard o hink of reasons why eiher ε or α should be relaed o X, Z, V and TSM. Usually hese random componens capure unobserved cohor effecs, bu here is only one cohor in ENI. Since he explanaory variables in equaion (5) are independen of ε and α, random effecs esimaion of equaion (5) yields consisen parameer esimaes. A Hausman es is used o invesigae he maer. An alernaive formulaion of equaion (5) specifies he model in erms of wage differenials beween immigrans and demographically similar naives. In his case he dependen variable is defined as d i = W i V i, where V i is he prediced value of naive wages condiional upon X i, and d is hypohesized o depend upon TSM and TSM 2. This would be equivalen o equaion (5) if he Mincer model used o predic naive wages was also valid for immigrans. If, however, i happens o be invalid he esimaed wage gap daa will be incorrec and he es of assimilaion heory will be invalid. By conras, equaion (5) does no impose he resricion ha immigran wages mus equal naive wages in he long run. As explained in Secion 3, no all immigrans repored posiive earnings, and an even smaller number repored posiive earnings in all 4 rounds. We decided o use a balanced sample 9, and herefore iniially exclude immigrans reporing posiive earnings in 3 rounds or less. Alhough we concenrae upon he balanced sample wih T = 4, we also experimen wih unbalanced samples, i.e. in which T < 4. Equaion (5) wih random effecs is esimaed by feasible GLS, and he esimaes are consisen as 8 See Greene (2003) p293-4 and Hsiao (1986) p43. 9 In a balanced sample immigrans ener and exi he sample a he same ime.

9 9 eiher T or N ends o infiniy. In a shor panel such as ENI, i is he laer ha is imporan. Taylor (1980) shows ha in shor panels FGLS rapidly approaches he Cramer-Rao bound wih N, so ha N > 50 may be considered as a "large" sample when T = 4. While he sample size in ENI maybe small when compared wih samples obained from censuses and oher large daases, i is sufficienly large o carry ou saisical inference since he balanced sample size is 80 and he unbalanced sample size is abou 250. The immigran assimilaion hypohesis predics γ > 0 and η > 0, implying a concave relaionship beween earnings and duraion in Quebec. The efficiency of he esimaes of hese parameers varies direcly wih he variance of TSM in he daa. TSM naurally varies beween rounds of ENI, bu as explained in Secion 3.2, i also varies wihin rounds because immigrans enered Quebec over a 6 monh period and because hey were inerviewed over a 6 monh period in each round. Had hey all arrived on he same day and been inerviewed on he same day, maers would have been differen, and he variance in TSM would have been smaller. The firs 3 rounds in ENI are equally spaced (1990, 1991 and 1992), bu he 4 h round ook place in This means ha observaions on TSM exis more or less coninuously during he firs 4 years in Quebec and during years 10 and 11, bu hey do no exis over he range of 4 9 years. This dead pach makes i impossible o idenify he assimilaion curve over his range. Neverheless, he firs 3 observaions are informaive abou immigran assimilaion in he shor erm, while he las observaion is informaive abou he longer erm Ariion Bias Our balanced sample may be self-seleced since immigrans who re-emigraed canno repor in all 4 rounds, bu especially in Round 4. There may also be self - selecion in employmen. Members of he balanced sample did no re-emigrae and were employed in all 4 rounds. Forunaely we know which immigrans belong o our balanced sample and which do no. Le D i = 1 if immigran i belongs o he balanced sample and D i = 0 oherwise. D i = 1 when laen variable D i * 0 wih: D = ϕ + Yψ + u * i i i (6) 10 Rerospecive earnings daa for were colleced in he 4 h round, bu we have decided no o use hem since hey are less reliable han daa colleced in real ime.

10 10 where Y denoes a vecor of variables hypohesized o deermine balanced sample membership. If u and ε have a bivariae normal disribuion wih correlaion ρ we may correc equaion (5) for sample seleciviy by specifying in equaion (5) he inverse Mills raio as a supplemenary regressor: φ( ϕ + Yiψ ) λi = σ ε 1 Φ( ϕ + Yψ ) i (7) where φ( ) is he sandard normal densiy, Φ( ) is is cumulaive counerpar, and λ has slope coefficien ρ. If ρ > 0, immigrans in he balanced sample are posiively selfseleced. If, however, ρ = 0 seleciviy does no induce bias in he esimaes of he parameers in equaion (5). Idenificaion ideally requires ha X be a subse of Y, i.e. selecion is influenced by insrumenal variables ha do no direcly affec earnings. If such insrumenal variables do no happen o be available, i.e. Y = X, hen he effec of ariion bias is enirely idenified paramerically, by assuming ha u and ε have a bivariae normal disribuion The Daa 3.1 The Sampling Frame for ENI The populaion under sudy consiss of immigrans in Quebec who obained heir immigran visas for Quebec prior o immigraion, and who enered Quebec direcly via Monreal's wo airpors (Dorval and Mirabel) or via he Blackpool land crossing wih he Unied Saes. This populaion excludes immigrans in Quebec who obained visas for he res of Canada and subsequenly chose o reside in Quebec, and holders of immigran visas for Quebec who enered Quebec indirecly 12. I also excludes holders of immigran visas for Quebec who chose o live in he res of Canada. ENI herefore recognizes ha immigrans in Quebec wih prior experience in he res of Canada migh be differen o immigrans who enered Quebec direcly. In he absence of reliable daabases from which o sample immigrans 13, he daa were colleced as follows bearers of Quebec immigran visas, aged 18+ years, enered Quebec via Dorval and Mirabel airpors or via Blackpool during mid 11 In he absence of convincing insrumens, we followed Horowiz and Manski (1998) by calculaing non-parameric bounds on ariion bias. However, hese bounds urn ou o be very wide, and we herefore do no repor hem. 12 For example, via airpors in he res of Canada or via oher land crossings. This means ha Asian immigrans are under-represened since many of hem ener Canada via Vancouver. 13 For example he daabase of Regie de l Assurance Maladie du Quebec (RAMQ) excludes many immigrans since hey do no have regular addresses.

11 11 June o November These 9,645 immigrans consiue he cohor populaion. All of hese immigrans were invied on heir arrival o paricipae in ENI. During his period here were 12,591 immigrans who enered Canada wih Quebec visas, bu his figure includes immigrans who enered Quebec indirecly. 1,880 of he 9,645 immigrans acceped he inviaion o paricipae in ENI, bu only 1000 of he laer were inerviewed in Round 1 some 6-15 monhs laer. The main causes of his ariion were problems wih elephone numbers (488 cases) and people who had lef Quebec (169 cases). 64 refused o be inerviewed. These daa refer herefore o a single cohor of 1000 immigrans who arrived in Quebec in 1989, and who were followed up in 4 face-o-face inerviews. The ENI sample as of Round 1 herefore consiues 10.4% of he populaion in he cohor. In conras o oher daabases, ENI conains rue informaion on he populaion. Therefore, whereas wih oher daabases i may be difficul o check heir fideliy because accurae daa on he populaion are unknown, we are able o check he fideliy of ENI. To hese ends a logi model was esimaed for sample paricipaion in which he covariaes include variables on he visa applicaion form. According o he logi model 14 women, married people, francophones and bilinguals are more likely o be sampled. However, he sampling error for hese parameers never exceeds 10%. Since represenaive samples of immigrans are paricularly hard o consruc and validae, ENI consiues a saisfacory if no perfec daabase for our presen purposes. Table 1 presens demographic characerisics based on daa colleced in Round 1. No surprisingly he immigrans are relaively young and educaed. The mos imporan source region is Norh Africa and he Middle Eas. Indeed, a quarer of he immigrans came from Libya. For 39% of immigrans he moher ongue was Arabic. The nex mos frequen moher ongue was French (10%). However, 50.6% of visa applicans and 47.5% of he sample knew French, while 42.8% of visa applicans and 40.2% of he sample knew English. These daa naurally underscore he imporance of French in he visa selecion process. Despie he represenaion of Arab counries in he daa, only 12.6% of immigrans in he sample declared hemselves o be Muslims. 3.2 The 4 Inerview Rounds of ENI 14 Deails of his model are available on reques.

12 12 Firs Round inerviews were carried ou beween June and Ocober 1990 by which ime he immigrans had been in Quebec beween 6 and 15 monhs. Round 2 inerviews ook place beween July and November 1991, in which 729 immigrans were inerviewed. Round 3 inerviews ook place beween May and Augus 1993, in which 508 were inerviewed. Noe ha immigrans inerviewed in Round 3 are a subse of immigrans inerviewed in Round 2. This means ha immigrans who were no inerviewed in Round 2 were also no inerviewed in Round 3. Finally, Round 4 ook place during January o April 2000, when 429 were inerviewed. In his las round an effor was made o race respondens los since Round 1 using records a RAMQ (Régie de l'assurance-maladie du Quebec). Table 2 repors ha 25.8% of immigrans were inerviewed in all 4 rounds, 33.3% were inerviewed in 3 rounds (25% in Rounds 1,2 and 3, and 8.3% in Rounds 1,2 and 4), 22.6% in 2 rounds and 18.3% in Round 1 only. Noe ha he cell for Rounds 1 and 3 is empy because, as menioned, ENI did no race immigrans in Round 3 who were los in Round 2. Table 1 shows ha he sample proporions for he balanced panel (inerviewed in all 4 rounds) are quie similar o heir counerpars in Round 1. This means ha sample selecion on observables is no a major problem in he balanced panel. Similar resuls apply o he non-balanced panels. Sample ariion a he provincial level is bound o be larger han a he naional level. Immigrans who lef Quebec o live in anoher province will consiue ariion in ENI, bu no in a naional sample (such as SLID or LSIC). Daa from IMDB show ha 11.3% of immigrans who arrived in Quebec in 1989 were living in oher provinces in Table 3, which repors he number of employed immigrans who repored posiive earnings, shows ha many immigrans were no coninuously employed. In he balanced panel of 258 immigrans only 81 immigrans repored posiive earnings in all 4 rounds. 168 repored posiive earnings in 3 rounds, of which 148 repored earnings in Round 1, 158 in Round 2, 127 in Round 3 and only 71 in Round repored posiive earnings in 2 rounds, of which 160 were in Round 1, 124 in Round 2, 43 in Round 3 and 65 in Round 4. Finally, 214 repored earnings in only one round, of which 137 were in Round 1, 26 in Round 2, 16 in Round 3 and 35 in Round 4. Table 3 also repors he raes of employmen in each round. In Round 1 he column sum is 15 IMDB daa show ha in 1995 here were 760 immigrans in Quebec who had immigraed o he res of Canada in 1989.

13 so he rae of employmen in 1990 was 52.6%. In Round 2 he rae of employmen is 53.4%, in Round 4 i is 52.6% and in Round 4 i is 58.7%. Therefore, he rae of employmen was sable during he firs 3 years afer immigraion, bur rose subsequenly. 3.3 Balanced Sample Seleciviy We define D1 i = 1 if immigran i was inerviewed in all 4 rounds, and herefore is a member of he balanced sample, oherwise D1 i = 0. We define D2 i = 1 if immigran i who was inerviewed in all 4 rounds was also employed in all 4 rounds, oherwise D2 i = 0. In Secion 2.3 we defined D i = 1 if immigran i was employed in all 4 rounds, and zero oherwise. Therefore, D i = D1 i D2 i. Since D2 is naurally nesed in D1 (firs decide wheher o remain in Quebec, hen decide wheher o be employed) a mulinomial selecion model is inappropriae. In Table 4 we repor probi models for D1 and D2 esimaed separaely. The covariaes ha feaure in Table 4 resul from applying he general-o-specific mehodology 16 applied o he variables in Table 2. The probi model for D1 shows ha survivorship in he sample depends mainly on region of origin. Immigrans from he US and Wesern Europe were more likely o be inerviewed in all four rounds, whereas immigrans from Sub-Saharan Africa were leas likely o be inerviewed in all four rounds. The probi model for D2 shows ha employmen raes among immigrans inerviewed in all four rounds depends mainly on educaion, wih universiy graduaes being he mos likely o be employed in all four rounds. Also, immigrans from he US and Wesern Europe were more likely o be employed in all four rounds. 3.4 Time Series Monhly ime series daa from Saisics Canada were obained for he following variables. The consumer price index for Monreal was used o deflae immigran earnings. Since he vas majoriy of ENI immigrans chose o reside in greaer Monreal, we chose o deflae heir wages by he CPI for Monreal raher han he CPI for Quebec as a whole. Naive wages are represened by average hourly wages of employees in Quebec. Unforunaely, ime series daa on naive wages for Monreal are no available. Neverheless, we deflae naive wages by he CPI for Monreal. Nor 16 See e.g. Thomas (1997) for a descripion of he general-o-specific mehodology.

14 14 are ime series daa available on he rae of unemploymen in Monreal, herefore we use he rae of unemploymen for Quebec. 3.5 The Economic Conex During he early 1990s he economy of Quebec experienced a deep recession ogeher wih he res of Canada. Real wages in Quebec barely grew during and unemploymen increased from abou 9% in 1989 o almos 15% in The naional rae of unemploymen increased from 6.2% o 10.5% over he same period. In 2000 real wages in Quebec were only 16% higher han in 1990, recording an average rae of growh of only 1.5% per year, of which he majoriy was achieved in he second half of he decade. Also, by 2000 he rae of unemploymen in Quebec had fallen o abou 7%, which was less han half of is peak in he early 1990s. Therefore ENI immigrans arrived a a paricularly difficul ime. However, maers eased during he second half of he 1990s. In Figure 1 we plo real wages in Quebec and he real wages of he immigrans in he sample. Noe ha hese sample means are based on all immigrans reporing in each round. We have cenered hese sample means in July 1990, July 1991, June 1992, and February 2000 when mos of he immigrans were sampled. Figure 1 shows ha immigrans earned less han naives in 1990 and 2000 and more han naives in This comparison is obviously unfair because immigrans were more educaed han naives. More imporan han absolue earnings of immigrans and naives is relaive earnings beween hese groups. The assimilaion hypohesis predics ha relaive immigran earnings should rise. In erms of Figure 1 his means ha immigran earnings should grow more rapidly han naive earnings. This appears o happen during , bu maers are quie differen when we compare 1992 and This issue is examined economerically in he nex secion.

15 15 4. Resuls 4.1 Esimaing Equaion (5) We begin by esimaing an unresriced version of equaion (5) in which he X vecor includes dummy variables for age group, gender, schooling, source region, and language skill. We also include in he specificaion he log of real wages and he rae of unemploymen (V and Z in equaion 5) in Quebec 17. Finally, we specify a polynomial in TSM, which is measured in weeks, alhough he esimaes for γ and η have been muliplied by 52 and 52 2 respecively o annualize hem. We use a prioriized general-o-specific specificaion search mehodology in which we give prioriy o TSM, which is he main variable of ineres, and also o naive wages (W). In Table 5 we repor esimaes of he resriced models in which saisically insignifican variables have been eliminaed using he general-o-specific specificaion search procedure. Model A in Table 5 is esimaed using daa on a balanced sample of 81 immigrans who repored posiive earnings in each of he 4 rounds. Model B is esimaed using an unbalanced sample in which 249 individuals have paricipaed in a leas 3 rounds. The main resul in boh models is ha here is no evidence of assimilaion. The coefficien on TSM should be posiive and he coefficien on TSM 2 should be negaive. The opposie ends o be he case; he coefficien on TSM is negaive, bu for he mos par is no saisically significan, while he coefficien on TSM 2 is posiive. Boh models imply ha effec of TSM on earnings during is negaive. We esed for higher order effecs in TSM, bu hey were no saisically significan. Oher resuls in Table 5 are as follows. Men earned abou 30% more han women according o boh models, and earnings vary direcly wih schooling. According o boh models older immigrans (aged 41+ years) earn more han younger immigrans (26 40 years). As menioned in Secion 3.3, his capures wo effecs, age a immigraion and life-cycle effecs. According o models A and B francophones earned subsanially less han anglophones. Indeed, here is no evidence ha knowledge of French or bilingualism is advanageous 18. Immigrans from US and Wesern Europe earned subsanially more han immigrans from elsewhere. Anoher surprising resul 17 We also conrolled for he rae of unemploymen (see Fig 1), bu i had no significan effec on immigran earnings. 18 Nor do Chiswick and Miller (2003) find ha immigrans who speak neiher English or French are disadvanaged.

16 16 is ha immigran earnings are no, on he whole, affeced by naive earnings, suggesing ha he immigran labor marke is segmened from is naive counerpar. Nor are hey affeced by he business cycle as measured by he rae of unemploymen. While he parameer esimaes of models A and B are qualiaively similar, and in several respecs quaniaively similar, here are some differences oo. For example, francophones earn 45% less han Anglophones according o model A bu 25% less according o model B. While boh models are saisically significan, R 2 ends o be small indicaing a large degree of unobserved heerogeneiy in immigran earnings, which, however, is broadly in line wih resuls obained by ohers. In boh models he Hausman es saisic is far from significan, so ha, as expeced, he esimaed random effecs are independen of he regressors. Therefore he esimaes repored in Table are consisen. 4.2 Tesing for Seleciviy and Ariion Bias We use he wo models in Table 4 o calculae he inverse Mills raios using equaion (7), λ 1i and λ 2i, which we add as regressors o model A in Table 5. Their - values are respecively 0.6 and -1.05, and he Wald es saisisic for excluding hese selecion regressors from model A is F = 0.23, which is far below is criical value. These resuls sugges ha here is no selecion bias in model A in Table 5. Alhough Table 4 esablishes ha selecion is no random a each selecion hurdle, he Wald es indicaes ha he random componen of earnings in model A is no significanly correlaed wih he random componen of selecion in boh hurdles. 4.3 Has he Assimilaion Curve Shifed? In equaion (5) he coefficiens γ and η, which deermine he shape of he assimilaion curve, are assumed o be ime invarian. However, empirical sudies of immigran assimilaion indicae ha he shape of he assimilaion curve may change. I may become seeper or flaer even if i reains is same overall asympoic form. The assimilaion curve may differ beween economies a he same poin in ime, and i may differ wihin economies a differen poins in ime. If immigrans face a greaer incenive o inves in hos counry skills hey will accumulae hese skills more rapidly and he assimilaion curve will become seeper. The same will happen if he reurn o hos counry skills happens o increase. Indeed, hese wo effecs will end o reinforce each oher.

17 17 Dropping subscrip i from equaion (5), bu permiing he parameers of he assimilaion curve o vary over ime gives: W = α + γ ( ) TSM η( ) TSM 2 The assimilaion curve seepens eiher when γ increases or when η decreases. If hese parameers are ime invarian he rae of change in immigran earnings is equal o γ - 2ηTSM > 0. If, however, hey are ime varying he rae of change of immigran earnings is equal o: dw d = ( γ 2η TSM ) + ( γ TSM dγ ηtsm d 2 (8) dη ) (9) d Even if he firs erm in brackes happens o be posiive, he sign of equaion (9) is indeerminae. If equaion (9) happens o be zero, i will appear as if assimilaion is no aking place, when in fac i is being concealed by he changing shape of he assimilaion curve. Or, equaion (9) may be negaive insead of posiive if γ is decreasing and η increasing. In his case i will look as if dissimilaion is aking place. Finally, if γ is increasing and η is decreasing immigran earnings may grow more rapidly over ime, in which even i will look as if super assimilaion is aking place. Longiudinal daa are incapable of handling srucural change in he assimilaion curve, unless auxiliary hypoheses are made abou he specific naure of srucural change. Ideally, we need o hold calendar ime consan, while duraion in he hos counry varies. Bu his is obviously no possible. Therefore, even longiudinal daa canno unambiguously idenify he assimilaion effec in he presence of srucural change. I may be shown ha he idenificaion of srucural change is made easier if here are daa on successive immigran cohors. In our case, however, we have daa on only one immigran cohor, which makes i very difficul o disenangle srucural change in assimilaion from our longiudinal daa. 5. Conclusions Economeric esing of he immigran assimilaion hypohesis has been hampered by he absence of appropriae longiudinal daa on immigrans. This absence has prevened follow-up of individual immigrans during heir firs years in he hos counry. In Canada, as elsewhere, researchers have resored insead o he synheic cohor mehodology as he work-horse o es he immigran assimilaion hypohesis.

18 18 Resuls obained using he synheic cohor mehodology ypically suppor he immigran assimilaion hypohesis in he sense ha during heir firs years in he hos counry immigrans narrow he earnings gap wih respec o naives, even if hey do no manage o close i enirely. The synheic cohor mehodology also ypically indicaes, in Canada as elsewhere, ha more recen immigran cohors are less posiively seleced on unobserved characerisics. In his paper we have used longiudinal daa on a single immigran cohor in Quebec o es he immigran assimilaion hypohesis. This is mos probably he firs ime ha longiudinal daa are used o es he immigran assimilaion hypohesis during he firs decade in he hos counry. The immigrans in he sample arrived in Quebec in 1989, and were followed-up four imes during heir firs decade in he hos counry. Two sriking conclusions arise ou of he resuls. Firs, here is no evidence a all in favor of he immigran assimilaion hypohesis. Immigran earnings do no iniially grow more rapidly han naive earnings, and hen grow less rapidly as duraion in he hos counry increases. Secondly, immigran earnings seem o be independen of labor marke condiions in he hos counry as measured by naive earnings and he naive rae of unemploymen. The former resul suggess ha puaive evidence in favor of he immigran assimilaion hypohesis obained using he synheic cohor mehodology may be spurious. The laer resul suggess ha immigran labor markes are segmened from he naive labor marke. Signally, our resuls are consisen wih Hum and Simpson (2000, 2004b), who used longiudinal for Canada as a whole daa over a shorer horizon, and who rejec he immigran assimilaion hypohesis. While wo similar findings hardly consiue a rend, i may be more han jus a coincidence ha whereas almos every synheic cohor sudy corroboraes he immigran assimilaion hypohesis, he handful of longiudinal sudies rejecs i. More longiudinal sudies are obviously required before any sronger conclusions can be reached. If, indeed, furher longiudinal sudies are consisen wih our findings and he findings of Hum and Simpson, his would imply ha he synheic cohor mehodology is flawed. The flaw lies in he fac ha he synheic cohor mehodology compares differen immigrans over ime raher han he same immigrans a differen poins in ime. If survivors in he labor marke are posiively seleced, he synheic cohor mehodology may simply be picking up survivorship bias raher han genuine immigran assimilaion. Immigran earnings

19 19 appear o grow faser iniially, no because immigrans are assimilaing ino he labor marke, bu simply because he immigrans whose earnings are observed say 10 years afer immigraion happen o be more posiively seleced han he immigrans whose earnings are observed afer say only 5 years afer immigraion. In shor, he assimilaion effec in synheic cohor daa may be a ficion of he daa. We have shown ha srucural change in he assimilaion curve may conceal underlying assimilaion behavior. I is conceivable ha in our case srucural change during he 1990s made i appear as if assimilaion has failed o ake place when he converse is rue. We canno rule ou his possibiliy. However, i seems o us ha i would be oo much of a coincidence if all he effecs of assimilaion jus happened o be oally eclipsed by srucural change. Resuls for Quebec do no necessarily apply o Canada as a whole because Quebec has is own selecion crieria for immigraion. Also, he ENI cohor arrived a a paricularly difficul ime. Perhaps maers migh have been differen had hey arrived when he economy was less recessed. On he oher hand, we found no evidence of assimilaion afer he economy began o recover. Even if immigran assimilaion prospecs are permanenly affeced by he sae of he business cycle on heir arrival, i would be unreasonable o mainain ha here is nohing o be learn from ENI. Their prospecs migh have been harmed bu no compleely desroyed. Finally, despie our effors o check for sample seleciviy, we canno rule ou he possibiliy ha remigraion beween Quebec and oher provinces is concealing evidence of assimilaion. If he successful immigrans qui Quebec for oher provinces, his would creae he misleading impression ha assimilaion was no aking place in Quebec.

20 20 Table 1 Sample Frequencies (%) Round 1 Balanced Panel Females Males Aged Aged Aged Married 56.0 Chrisians 58.4 Muslims 12.6 Buddhiss 7.6 No schooling Primary school Secondary school Pos secondary Universiy 25.4 English English and French French No English or French Lain America USA & W Europe E. Europe N. Africa & M Eas Oher Africa Asia & Pacific Round Number Table 2 Panel Compleion Raes Percen and ,2 and ,2,3 and and ,2 and 4 8.3

21 21 Table 3 Employmen Number of Rounds Round 1 Round 2 Round 3 Round 4 Toal Employed Employmen % Table 4 Probi Models for Balanced Sample Seleciviy D1 D2 Inercep (9.921) (8.008) Age (1.733) Male (1.941) Secondary (2.278) (5.491) College (3.227) Universiy (6.186) US and W. Europe (2.483) (3.422) N.Africa & M.Eas (3.613) Sub-Saharan Africa (1.826) No French & English (1.98) Log L N D = Absolue -values in parenheses.

22 22 Table 5 Panel Regression Esimaes of Equaion (5) A: 4 Rounds B: 3 4 Rounds β -sa β -sa Inercep lnw Age Age Male Secondary school College Universiy English & French French No English French US and W Europe TSM TSM R RMSE N Hausman Tes Dependen variable: logarihm of real earnings. Esimaed wih random effecs. Base: Women, primary school, aged 18-25, English speaker.

23 23 Figure 1: Earnings and Unemploymen in Quebec % % real weekly wages % 10% 8% 6% rae of unemploymen Dae Earnings in Quebec a Consan Prices. Immigran Earnings a onsan Prices Rae of Unemploymen in Quebec % 2% 0%

24 24 References Adeymir A. and M. Skuerud (2005) Explaining he deerioraing enry earnings of Canada's immigran cohors, Canadian Journal of Economics, 38: Balagi B.H. (2005) Economeric Analysis of Panel Daa, 3 rd ediion, John Wiley, New York. Baker M. and D. Benjamin (1994) The performance of immigrans in he Candian labor marke. Journal of Labor Economics, 12: Beensock M. (1996) Failure o absorb: reurn migraion by immigrans o Israel. Inernaional Migraion Review, 34: Beensock M. and Y. Ben Menahem (1997) The labour marke absorpion of CIS immigrans o Israel: Inernaional Migraion, 35: Beensock M., B.R. Chiswick and A. Paliel (2005) Endogenous assimilaion and immigran adjusmen in longiudinal daa, mimeo. Black R., T. Fielding, R. King, R. Skeldon and R. Tiemoko (2003) Longiudinal Sudies: An Insigh ino Curren Sudies and he Social and Economic Oucomes for Migrans. Working Paper 14, Sussex Cenre for Migraion Research. Bloom D.E., G. Grenier and M. Gunderson (1995) The changing labour marke posiion of Canadian Immigrans. Canadian Journal of Economics, 28: Borjas G.S. (1985) Assimilaion, changes in cohor qualiy, and he earnings of immigrans. Journal of Labor Economics, 3: Borjas G.S. (1989) Immigran and emigran earnings: a longiudinal sudy. Economic Inquiry, 27: Chiswick B.R. (1978) The effec of Americanizaion on he earnings of foreignborn men. Journal of Poliical Economy, 85: Chiswick B.R.,Y.L.Lee and P.W. Miller (2005) Immigran earnings: a longiudinal analysis. Review of Income and Wealh, 51: Chiswick B.R. and P.W. Miller (1995) The endogeneiy beween language and earnings: inernaional analyses. Journal of Labor Economics, 13: Chiswick B.R. and P.W. Miller (1996) Language and earnings among immigrans in Canada: a survey. H. Duleep and P. Wunnava (eds) Immigrans and

25 25 Immigraion Policy: Individual Skills, Family Ties and Group Ideniies. JAI Press, Greenwich, CT. Chiswick B.R. and P.W. Miller (2003) The complemenariy of language and oher human capial: immigran earnings in Canada. Economics of Educaion Review, 22: Chiswick B.R. and P.W. Miller (2005) Review of Income and Wealh Duleep H.O. and M.C. Reges (1997) Measuring immigran wage growh using mached CPS files. Demography, 34: Duleep H.O. and D.J. Dowhan (2002) Insighs from longiudinal daa on he earnings growh of US foreign born men. Demography, 39: Frenee M. and R. Morissee (2003) Will hey ever converge? Earnings of immigrans and Canadian-born workers over he las wo decades. Saisics Canada, Oawa. Gran M.L. (1999) Evidence of new immigran assimilaion in Canada. Canadian Journal of Economics, 32: Green D.A. and C. Worswick (2002) Earnings of immigran men in Canada: he roles of labor marke enry effecs and reurns o foreign experience. Ciizenship and Immigraion Canada, Oawa. Greene W.H. (2003) Economeric Analysis, 5 h ediion, Prenice Hall, New Jersey. Horowiz J.L and C.F. Manski (1998) Censoring of oucomes and regressors due o survey nonresponse: idenificaion and esimaion using weighs and impuaions, Journal of Economerics, 84: Hsiao C. (1986) Analysis of Panel Daa. Cambridge Universiy Press. Hsiao C. (2003) Analysis of Panel Daa, 2 nd ediion. Cambridge Universiy Press. Hu W-Y. (2000) Immigran earnings assimilaion: esimaes from longiudinal daa. American Economic Review, 90: Hum D. and W. Simpson (2000) Closing he wage gap: economic assimilaion of Canadian immigrans reconsidered. Journal of Inernaional Migraion and Inegraion, 1: Hum D. and W. Simpson (2004a) Economic inegraion of immigrans o Canada: a shor survey. Canadian Journal of Urban Research, 13: Hum D. and W. Simpson (2004b) Reinerpreing he performance of immigran wages from panel daa. Empirical Economics, 29:

26 26 Labeaga J. (1999) A double-hurdle raional addicion model wih heerogeneiy: esimaing he demand for obacco. Journal of Economerics, 93: Renaud J. (2003) Wha a Difference 10 Years Can Make!: The Selemen Experience of Immigrans Admied o Quebec in 1989, Les Publicaions du Québec. Schaafsma J. and A. Sweeman (2001) Immigran earnings: age a immigraion maers. Canadian Journal of Economics, 34: Taylor W.E. (1980) Small sample consideraions in esimaion from panel daa, Journal of Economerics, 13: Thomas R.L. (1997) Modern Economerics, Addison-Wesley, New York. Warman C.R.,and C. Worswick (2004) Immigran earnings performance in Canadian ciies: 1981 hrough Canadian Journal of Urban Research, 13: