M EM O RA N D U M
N o 4/2001
O ut-M igration of Im m igrants: Im plications for A ssim ilation A nalysis
By Pål Longva
ISSN : 0801-1117
D epartm ent of Econom ics
This series is published by the U niversity of O slo
D epartm ent of Econom ics
In co-operation w ith
The Frisch C entre for Econom ic R esearch
P. O .Box 1095 Blindern N -0317 O SLO N orw ay Telephone: + 47 22855127 Fax: + 47 22855035
Internet: http://w w w .oekonom i.uio.no/
e-m ail: econdep@ econ.uio.no
G austadalleén 21 N -0371 O SLO N orw ay
Telephone: +47 22 95 88 20
Fax: +47 22 95 88 25
Internet: http://w w w .frisch.uio.no/
e-m ail: frisch@ frisch.uio.no
List of the last 10 M em oranda:
N o 03 B y Ø ystein K ravdal: The Im portance of Education for Fertility in Sub- Saharan A frica is Substantially U nderestim ated W hen Com m unity Effects are Ignored. 33 p.
N o 02 B y G eir B. A sheim and M artin L. W eitzm an: D O ES N N P G RO W TH IN D ICA TE W ELFA RE IM PRO V EM EN T? 8 p.
N o 01 B y Tore Schw eder and N ils Lid H jort: Confidence and Likelihood. 35 p.
N o 43 B y M ads G reaker: Strategic Environm ental Policy w hen the G overnm ents are Threatened by Relocation. 22 p.
N o 42 B y Ø ystein K ravdal: The Im pact of Individual and A ggregate U nem ploym ent on Fertility in N orw ay. 36 p.
N o 41 B y M ichael H oel and Erik M agnus Sæther: Private health care as a supplem ent to a public health system w ith w aiting tim e for treatm ent:
47 p.
N o 40 B y G eir B. A sheim , A nne W enche Em blem and Tore N ilssen: H ealth Insurance: Treatm ent vs. Com pensation. 15 p.
N o 39 B y D iderik Lund and Tore N ilssen: Cream Skim m ing, D regs Skim m ing, and Pooling: O n the D ynam ics of Com petitive Screening. 21 p.
N o 38 B y G eir B. A sheim and A ndrés Perea: LEX ICO G RA PH IC
PRO BA BILITIES A N D RA TIO N A LIZA B ILITY IN EX TEN SIV E G A M ES. 30 p.
N o 37 B y G eir B. A sheim and W olfgang Buchholz: TH E M A LLEA B ILITY O F U N D ISCO U N TED U TILITA R IA N ISM A S A CR ITER IO N O F
IN TERG EN ERA TIO N A L JU STICE. 22 p.
A com plete list of this m em o-series is available in a PD F® form at at:
http://w w w .oekonom i.uio.no/m em o/
O ut-M igration of Im m igrants: Im plications for A ssim ilation A nalysis
1Pål Longva
D epartm ent of Econom ics, U niversity of O slo P.O . Box 1095 Blindern, N -0317 O slo, N orw ay
Em ail: pal.longva@ econ.uio.no
1 I w ould like to thank Christian Brinch, Svein Longva, N ico K eilm an, O ddbjørn Raaum , Tore Schw eder, and Steinar Strøm for com m ents, and Tom -Erik A abø for help w ith the data organization.
A bstract
The labor m arket behavior of im m igrants is studied in relation to the propensity to out- m igrate. U tilizing a large m icro dataset for individuals in N orw ay, w hich enables
identification of out-m igrants, I find in line w ith other studies that attachm ent to the labor m arket influences positively on the propensity to stay in the host country. A m ong the individuals in the labor force I find that out-m igrants originally from N on-O EC D
countries are negatively selected in term s of labor m arket earnings, from the pool of N on- O ECD im m igrants. A m ong im m igrants from O ECD countries a m ore m ixed picture arises, w ith the out-m igrants draw n from both extrem es of the earnings distribution.
Further, I find som e indications that im m igrants in the upper part of the earnings distribution for O ECD im m igrants are highly m obile and consequently stay only a very short tim e in the host country.
1. Introduction
Im m igrant labor is an essential part of the labor force in m ost w estern countries. They are w illing to undertake jobs in low -status occupations, and provide vital services as self- em ployed in sm all businesses, w orking long hours. For sm all countries especially, im m igrants fill certain highly specialized positions in the labor m arket w here the host country could not provide supply from its ow n labor force. A lso, to som e extent they act as a buffer in the labor m arket by providing a supply of labor in boom s, and by a
w ithdraw al from the labor m arket in dow nturns. O n the other hand, im m igrants are accused of over-utilizing the relatively generous w elfare system found in m ost W estern countries, to generate ethnic conflicts and to underm ine the existing national culture.
W estern governm ents have taken notice of the ongoing trend of declining and ageing native populations (U nited N ations, 2000), w hich could increase the pressure for allow ing m ore im m igrants into these countries. A lso, the labor m arket in certain regions, for instance w ithin EU , are becom ing m ore integrated, and the m obility of the w ork force is, in general, increasing. M oreover, as the EU expands eastw ards m ore countries w ill experience the drastic decline in the overall costs of m oving across borders.2 In sum , the discussion of im m igration policy is on the forefront in m ost w estern countries.
A ny im m igration policy should be based on a base of know ledge as w ide as possible.
Crucial inform ation in this respect is how im m igrants conform to, or assim ilate in the labor m arket. O bviously, one very im portant aspect of the assim ilation process is how long the im m igrant stay in the host country. A t each point in tim e the im m igrant can in principle choose w hether to stay in the host country or not.3 If the im m igrant choose to leave the host country, he/she could either m igrate back to the source country (return m igrate) or to another country (repeat m igration). To this end I denote both these events out-m igration.
2 O n the other hand, the possible inclusion of new East-European m em bers of EU has raised a huge debate on w hether the new m em ber should have free access to the European labor m arket. O ne suggestion is to define a transition period for the new m em bers w here the flow of im m igrants are restricted.
3 I here ignore the existence of contracts w hich lim its the stay in the host country, as w ell as the possibility of involuntary deportations. D ustm ann (2000) provides an overview of the different types of m igrants.
The num ber of im m igrants w ho out-m igrate from W estern countries is substantial.
Figure 1 presents the annual flow s of foreign-born individuals in- and out of N orw ay for the period 1961-1999.4 The outflow as a share of inflow has a m ean equal to 0.79 in the period. That is, for every 10 foreign-born individuals that m ove into N orw ay, roughly 8 individuals m ove out. Studies of questions related to im m igration should therefore ideally include both im m igration and out-m igration, and the selection process im plied by the large difference betw een gross- and net im m igration.
Figure 1. Inflow and outflow of foreign-born individuals into- and out of N orw ay.
O ut-m igration is a topic w hich, due to insufficient and incom plete data, has m ainly been studied from a theoretical point of view .5 H ow ever, inform ation about return propensities
4 Source: Statistics N orw ay, Population Statistics.
5 A s far as I know , only Tysse and K eilm an (1998) have studied the out-m igration of im m igrants in N orw ay.
H ow ever, they did not consider the correlation betw een earnings and out-m igration w hich is the m ain focus in the current study.
0 5 10 15 20 25 30 35 40 45
1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 Year
Tousand
Inflow O utflow
and know ledge of how the out-m igrants are characterized is essential for policy. For instance, if only a sm all percentage out-m igrate w hen the labor m arket w orsen, the scope for using im m igration as a buffer in the labor m arket becom es less attractive. Even w orse, if the m ost able leave the host country such a policy could result in a less able im m igrant population. O n the other hand, if those w ho do not succeed in the host country labor m arket leave, w e are left w ith a gradually positively selected im m igrant w ork force.
The earnings assim ilation studies typically utilize tw o or m ore cross sectional data sets, w hich m akes it possible to identify assim ilation effects under strong assum ptions, a survey is provided in Borjas (1999). O ne of these assum ptions is that the group of out-m igrants is a random sam ple from the im m igrant cohort under study. A ll logic and scattered em pirical evidence contradicts that this is the case. If the out-m igrants are selective, in the sense that the distribution of earnings-determ ining characteristics differ from the distribution to those w ho stay, earnings assim ilation estim ates obtained from cross sectional data w ould be biased (henceforth denoted the out-m igration bias). V ery few attem pts have been m ade in correcting the estim ates for possible out-m igration bias. H ow ever, recently a couple of studies have focused on this problem on U .S. data (H u, 2000; D em om bynes, 1999). Both studies rely on com paring estim ates obtained on data from very different sources, and as such have com e only part of the w ay in identifying the sign and strength of the bias. A lso, there are prom ising ongoing w ork on data from D enm ark and Sw eden (H usted et. al., 2000b; Edin et. al., 2001). These studies utilize data from a very long tim e fram e.
H ow ever, the sam ple sizes are rather sm all, and the availability of covariates are lim ited.
The m ain conclusion in these studies, both on the U .S. data and the N ordic data, is that the out-m igrants are negatively selected in term s of earnings, from the group of im m igrants.
This leads the authors to conclude that standards earnings assim ilation estim ates, obtained from repeated cross-sectionals in w hich out-m igration is ignored, are upw ard biased.
This paper aim s at clarifying these issues using N orw egian data. The m ain advantage is that I observe the entire population of im m igrants at each point in tim e. This allow s m e to rather precisely characterize the different subgroups of the im m igrant population. The out- m igrants are identified by sam ple attrition, w hich allow s for an analysis of the propensity to out-m igrate. The paper is organized a follow s. In the next section I discuss w hat econom ic theory has to say about w hy individuals m ove across international borders. In section 3 I survey the available em pirically related literature on out-m igration, focusing on
those studies w hich analyse labor m arket earnings. In section 4 I present the data structure.
Section 5 contains an analysis of how the out-m igrants differ from those w ho stay. In section 6 I discuss how selective out-m igration could influence on earnings assim ilation estim ates. Finally, in section 7 the results are sum m arized and interpreted.
2. Theory
In general, w hy do individuals m ove across borders? O riginating in Sjaastad (1962) the dom inating explanation has been hum an capital investm ent. B y m aking an investm ent (forgone earnings, travel costs etc.) the m igrants explores w age differentials and/or acquire skills not obtainable in the source country. The basic m odel, along w ith som e com m on interpretations, are neatly sum m arized in Chisw ick (1999). This m odel, in its m ost sim ple version, assum es perfect foresight, no unem ploym ent, fixed w age rates in the source- and host country, and that the m igration decision is irreversible (perm anent m igration). A risk neutral individual w ho m axim izes lifetim e incom e, and w ho lives infinitely, w ill m igrate if the rate of return to m igration, is greater than the real rate of interest cost in financial m arkets.
If all w orkers are identical w e w ill typically end up w ith a corner solution, w ere either all or no one m igrates. In reality both w ages and costs w ill vary according to skill level.
G iven that high skilled w orkers earn m ore than low skilled and that they are m ore efficient in the m igration process in the sense that they face low er direct costs and that they spend less tim e on the m igration process, they w ill have a higher rate of return of return to m igration com pared to low ability w orkers. The m ore the rate of return differs betw een high and low ability w orkers, the m ore selective w ould the m igration be. A lso, a key insight from this very sim ple m odel is that the larger are the direct m igration costs, the low er is the propensity to m igrate, but the greater is the propensity for a favorable (positive) selectivity in m igration.
W ithin this fram ew ork an im portant determ inant of the flow of im m igrants is the ratio of w ages for high skilled versus low skilled, in the source and host country. If for instance the ratio is higher for high skilled w orkers (as it w ould be if the w age distribution is m ore com pressed in the source country com pared to in the host country) this w ould further
increase the positive selectivity in m igration. If the opposite is true, that is if the ratio is low er for high skilled w orkers, this effect w ould w ork in the opposite direction and counteract the effects stem m ing from higher efficiency and low er direct costs.
H ow ever, people m ove across borders also for other reasons than purely econom ical. O ne w ould expect that the m echanism outlined above w ould be less intense am ong those m igrants w ho m ove across borders as refugees or due to fam ily reunification etc. A lso, m ore realistic, the inform ation about the skill level w ould be asym m etric in the sense that the host country em ployer w ould not have full inform ation about the skill level and
productivity for the new ly hired m igrant. O ne sim ple assum ption (K atz and Stark, 1987) is that the m igrant on arrival is paid according to the average skill level am ong im m igrants.
Thus, the high skill w orkers w ould face a less favorable ratio of source country/host country w ages at the outset, w hich w ould w ork in the direction of a less favorably selected m igration.
To sum up, the m odel outlined above identifies w age differentials along w ith cost- and tim e efficiency as the m ain push factors (supply side) behind m igratory behavior. A s such, the m odel does not point to any particular m echanism for returning, besides from the obvious possibility that the w age differential could turn around over tim e in favor of the source country, or in principle of other countries, and induce the m igrant to out-m igrate.
Still, the m odel serves as a useful reference in any discussion and analysis of labor m igration.
W hat then m otivates im m igrants to out-m igrate? D ustm ann (1996b) provides som e suggestions w ithin an optim al life-cycle hum an capital m odel. In particular, it is
rationalized w hy the m igrant w ould out-m igrate despite a higher w age in the host country com pared to the source country, and the optim al duration in the host country is calculated.
Three different m otives are put forw ard: (i) accum ulation of hum an capital; (ii) com plem entarities betw een consum ption and the location for consum ption; and
(iii) differences in relative prices. It is show n that the optim al duration is not necessarily increasing in the w age level in the host country. M ore conventionally, the optim al duration is increasing in the planning horizon, and in the desired stock of savings at the end of the planning horizon.
W hat do the above m entioned theories im ply for the labor m arket behavior of those w ho out-m igrate versus those w ho stay? A ssum ing that there is a w age prem ium associated w ith the initial m ove, the tem porary m igrant w ill typically, due to the lim ited horizon of the stay in the host country, supply m ore labor, save m ore of their incom e, and invest less in host country specific hum an capital. A lso, to the extent that the hum an capital
requirem ents in the labor m arket varies across borders, short-term m igrants w ould have less incentive to invest in host-country hum an capital, w hich could lead to a less favorable selective m igration.
D ustm ann (2000) contains an excellent discussion of these questions. The m otives behind the m ove across borders are investigated from a theoretical point of view , and the
im plications for em pirical analysis is discussed. The m ain point is that the length of the stay in the host country could affect earnings-influencing investm ent to a high degree, and hence em pirical specifications w hich do not take this into account risk to obtain estim ates that are biased. A s investm ents in hum an capital in general m ay have strong im pacts on the earnings profile, differences in assim ilation rates betw een im m igrant groups of
different origins can be explained by different return probabilities. H ow ever, it is not clear how to approach this problem , even w ith full inform ation on com pleted duration, as it is the unobserved intention w hich in principal determ ines the investm ents in general- and country specific hum an capital.
H ow ever, the extent of w hich the im m igrants behave rationally according to the factors outlined above could vary betw een groups of im m igrants. I w ould expect that both the m otivation behind the initial m igratory m ove as w ell as the geographical and cultural distance are im portant in this respect. Im m igrants w ho arrive as refugees and asylum seekers w ould be expected to behave less rational than pure labor m igrants. A lso the less the geographical and cultural distance betw een the source and the host country, the m ore w ould w e expect the im m igrant to respond to for instance changes in the w age structure and job opportunities.
3. A brief review of the literature
M y review of the related em pirical literature is threefold. Firstly, I discuss the existing literature on earnings assim ilation. Secondly, I give a review of those studies w ho analyse out-m igration in general, and w hich focus on to w hich extent the out-m igrants are selected from the group of im m igrants in the host country. Lastly, I discuss those contributions in the literature w hich explicitly have gone into the issue of how (selective) out-m igration affects earnings assim ilation estim ates.
The em pirical literature on how im m igrants conform to the host country labor m arket is substantial, and although m ost of the focus has been on the U .S. labor m arket a num ber of studies have been undertaken for European countries, Canada and A ustralia as w ell. The typical m odern study of earnings assim ilation utilize tw o or m ore cross sections, as in Borjas (1995), in order to separate cohort- and years-since-im m igration effects. Evidence for N orw ay is provided by H ayfron (1998) and Longva and Raaum (2001a). A guilar and G ustafson (1991) provides results for Sw eden, w hile H usted et. al. (2000) studies D anish data. There are huge differences across nations w hen it com es to the history of
im m igration, im m igration policy, and also in how an im m igrant, as w ell as earnings, are defined and m easured in the data sets. This, together w ith the variations of m ethods applied to cope w ith the assorted data deficiencies, m akes a cross country com parison difficult. H ow ever, in m ost countries there seem s to be a underlying declining earnings capacity of the m ost recent im m igrant cohort over tim e, especially in the 80’s. This justifies the use of tw o or m ore observations in tim e w hen m easuring the effect of years- since-m igration on earnings. N ext, im m igrants starts out w ith an earnings disadvantage com pared to natives, but im prove their earnings over tim e. The speed at w hich this happens (the degree of assim ilation) varies. For instance, Longva and Raaum (2001a) finds that im m igrants from N on-O ECD countries im prove their earnings by about 6 percent relative to natives during their initial 10 years of residence in N orw ay.
Turning to the out-m igration studies, Borjas and Bratsberg (1996) analyzes the return m igration of im m igrants in the U .S. Based on m easures of the return m igration flow by source country, data from the 1980 Census, and various aggregated data from the source countries, they find that im m igrants tend to return to rich countries and to countries not far aw ay from U .S. A lso, the data gives som e support to the hypothesis that return m igration
tend to accentuate the selection originally characterizing the im m igrant flow . A sim ilar study is conducted by Ram os (1992) w ho finds that the m igrants from Puerto Rico to the U .S. are negatively selected, both on observables and unobservables. Those w ho return from U .S. are, on the other hand, positively selected from the pool of Puerto Rican m igrants in the U .S. These findings are interpreted as a support of a m odel w here
individuals from the low er part of the earnings distribution m ove from econom ies w ith the larger incom e inequality to econom ies w ith the less incom e inequality.6
W ithin the European context, D ustm ann (1996a) provides a sim ple em pirical analysis of the determ inants of the intended stay in the host-country am ong im m igrants to G erm any.
A lso, on a sm all sam ple of returned im m igrants (after a stay in G erm any) to Turkey, the out-m igration decision is evaluated ex ante. H e finds that the propensity to out-m igrate increases w ith the age at entry, but declines w ith the num ber of years in the host country, holding age at entry constant. This latter effect is interpreted as a strong assim ilation effect. H ow ever this could also be caused by selection if those w ho have the highest propensity to out-m igrate do so after a short duration of stay.
H usted et. al. (2000b) studies out-m igration on a com prehensive D anish data set. The m ain focus is on estim ations of the probability am ong the stock of im m igrants in 1986 of leaving D enm ark in the subsequent nine years. Like D ustm ann (1996a) they find a negative duration dependence in the sense that the probability of leaving is a decreasing function of the num ber of years the im m igrant has resided in the host country. Further, the closer the im m igrant has been attached to the labor m arket, the low er the probability of leaving, w hich is interpreted as an indication of that the least successful out-m igrate. O n the other hand, registered unem ploym ent w orks in the other direction, as those w ith a high num ber of m onths receiving unem ploym ent benefits yield a low out-m igration probability.
Interestingly, the ratio betw een incom e from w ork and gross incom e, interpreted by the authors as the ability to provide for oneself, enters negatively in the probit equation. This leads the authors to infer that the m ore able the im m igrant are to provide for her/him self, the low er is the probability of out-m igration.
6 H ow ever, as the inhabitants of Puerto Rico are U .S. citizens, the relevance of this analysis in an international m igration context could be questioned.
Borjas (1989) is the first study I am aw are off that explicitly focuses on the problem s, due to selective out-m igration, inherited in the standard approach to earnings assim ilation estim ation. A t hand he has a sm all panel sam ple of scientist and engineers in the U .S., for the years 1972 and 1978. H e finds that im m igrants are m ore likely to leave the sam ple relative to natives (as expected), and that the probability is decreasing w ith the num ber of years since im m igration, controlling for a num ber of other characteristics. Regarding earnings, it is show n that it is the least skilled (m easured by w eekly earnings) that
disappear from the sam ple. Three separate estim ations of the earnings assim ilation m odel are undertaken, one for the full population, one for the subsam ple w ho stays in the sam ple, and one for the subsam ple w ho disappears from the sam ple. The results show that the stayers start out w ith a sm aller w age differential com pared to natives than m overs.
D ustm ann (1993) investigates how the earnings assim ilation pattern depends on the
intended duration of the stay (at arrival) in the host country. O n theoretical grounds he first argues that the am ount of hum an capital investm ent undertaken by the im m igrant depends negatively on the intended duration of the stay, im plying a flatter earnings profile as the intended duration decreases. Secondly, the selectivity of the tem porary im m igrants depends crucially on the state of the labor m arket in the source country as w ell as in the host country, at the tim e of entry. Specifically, high unem ploym ent along w ith low levels of unem ploym ent benefits, in the source country relative to the host country, could give a negatively selected pool of im m igrants in the host country. D ata from G erm any give som e support to these hypotheses. H ow ever the sam ple is sm all, w ith no tim e dim ension, hence different cohort effects in the pool of im m igrants are im plicitly assum ed aw ay. D ustm ann (1999) tests the sam e hypothesis on data for language acquisition, a test that supports the theory.
There have been, and still are, huge controversies regarding the assim ilation of im m igrants in the U .S. labor m arket. Thus, one should not be surprised of the fact that the potential role of out-m igration has com e into focus recently, in spite of the considerable difficulties in obtaining reliable data for this purpose. The studies that I am aw are of are D em om bynes (1999), H u (2000) and Lubotsky (2000). The com m on approach is to com pare earnings assim ilation patterns obtained from standard repeated cross-sections and longitudinal data.
The source and approach in constructing the longitudinal data varies betw een the studies7 but they all face, to a varying degree though, data problem s like censoring, eligibility etc.
This is a com m on problem for the recent U .S. studies although it seem s that
Lubotsky (2000) goes a long w ay in clarifying the com parability problem , as w ell as taking this into account in the approach. H u (2000) and Lubotsky (2000) find that assim ilation estim ates based on the Censuses are biased upw ards caused by negatively selected return m igrants, in the sense that the out-m igration flow is substantial and consisting of individuals w ho’s potential earnings are below those w ho rem ain.8
D em om bynes (1999) on the other hand finds indications of a m ore rapid earnings grow th in the 90’s using the longitudinal data.
Lastly, a very interesting study is conducted by Edin et. al. (2001) on Sw edish data. They observe a three percent sam ple of the population in each year 1970-97, and are able to distinguish betw een stayers and m overs am ong the im m igrant population by a sim ple sam ple attrition procedure. W ithin a cross-sectional fram ew ork they estim ate earnings profiles for im m igrants by region of origin. They find no assim ilation effect for im m igrants from the N ordic countries and from the O ECD region, these groups stay roughly 15-20 percent below the average native in incom e. Im m igrants from N on-O EC D countries on the other hand enters the econom y w ith a huge disadvantage in incom e, and experience only m odest earnings assim ilation, concentrated on the first five years of their stay. N ext, the consequences of out-m igration for the earnings assim ilation estim ates are investigated. They find that the least successful outm igrate. That is, w ithin each group by region of origin, the group of out-m igrants could be characterized as those: (i) least attached to the labor m arket, m easured by w hether or not they are observed w ith zero earnings; and (ii) in the low er part of the earnings distribution, am ong those w ith positive earnings. Thus, by restricting the sam ple to those w ho rem ain in Sw eden, they find that the assim ilation estim ates w eakens com pared to the benchm ark analysis w ere bias from out-m igration are not taken into account. The authors state quite strongly that this is the
7 D em om bynes (1999) uses m erged CPS (Central Population Study) data for households, w hile H u (2000) and Lubotsky (2000) use Social Security data.
8 Interestingly, Lubotsky (2000) also finds that m isclassification in Census based studies of transient im m igrants as m ore recent arrivals than they actually w ere, leads to an overstatem ent of the declining earnings potential am ong successive im m igrant cohort in the U .S.
case. H ow ever, in the regressions they only adjust for age and gender, hence they do not take into account that other variables could vary betw een the groups. M oreover, the sam ple is quite sm all.
To sum up, a large em pirical literature exists on how im m igrants fare in the host-country labor m arket w ith regards to earnings. A s a rule, these studies use the so called synthetic panel approach w here tw o or m ore cross sectional data sets are utilized in order to follow im m igrant cohorts over tim e. In contrast, there exist only scattered em pirical evidence on how out-m igration influence the assim ilation estim ates. The com m on finding is that the out-m igrants are negatively selected from the pool of im m igrants, in the sense that their labor m arket earnings are less than for those w ho rem ain. Taken at face value this m eans that standard earnings assim ilation estim ates are upw ard biased. M y study aim s at adding to this recent literature.
4. D ata
In tw o previous studies of the earnings assim ilation of N orw egian im m igrants w e have utilized m icrodata w ith tw o observations in tim e, 1980 and 1990 (Longva and Raaum , 2001a; 2001b). The possible influence of selective out-m igration w as not specifically accounted for. Thus, it could be interesting in itself to apply these data, in order to get a validity check of our previous results. H ow ever, the 1980-90 tim e dim ension is ham pered by a num ber of com plicating factors. Firstly, the m acroeconom ic situation w as very different at the tw o points in tim e, w ith the aggregate unem ploym ent alm ost three tim es higher in 1990 as in 1980. Longva and Raaum (2001a) show that im m igrant earnings are m uch m ore affected by high unem ploym ent com pared to natives, thus the com m on assum ption of equal period effects is hard to em ploy. Secondly, the characteristics of the im m igrant population changed dram atically, as the im m igration in the 1980’s w as dom inated by refugees, asylum seekers, and fam ily reunification, w hile the im m igration until the m id 1970’s w as dom inated by labor m igrants. Each of these factors are by them selves w orth a study. The question of how out-m igration influences the results, adds to the com plexity of the analysis. In addition, for those w ho leave the sam ple during the 80’ies, and thus under certain assum ptions could be characterized as out-m igrants, w e do not observe w hen they arrived in N orw ay for the first tim e. This puts a lim it on the use of
this data set to investigate the precise effect of out-m igration on earnings assim ilation estim ates. Therefore, in addition to a characterization of m overs versus stayers, I resort to study the out-m igration decision only, by estim ating the probability to out-m igrate. The set builds on inform ation of all im m igrants and a random sam ple of natives in 1980 and 1992.9 D etails on the underlying sam pling procedure are docum ented in A ppendix 1. The set is denoted the 1980 Sam ple.
The second data set available consists of all residents in N orw ay in 1993 and in 1997, w ith full inform ation about arrival date and country of origin,10 w hich enables a calculation of the length of the residence for all im m igrants, including those w ho out-m igrate. H ence, I can study the relationship betw een the propensity to out-m igrate, earnings and years-since- m igration. M oreover, I can undertake an explicit com parison of earnings assim ilation estim ates obtained from sam ples w here w e can include/exclude those w ho leave the sam ple due to out-m igration. D etails on the underlying sam pling procedure are docum ented in A ppendix 2. The set is denoted the 1993 Sam ple.
Im m igrant
I classify an individual as an im m igrant if he/she is born in a foreign country, w ith tw o foreign born parents. H ow ever, as explained in A ppendix 1, I do not have available the country of origin for the out-m igrants in the 1980 sam ple. For this group I therefore assum e that the country of origin is equal to the observed citizenship. That is, naturalized im m igrants that out-m igrate are not captured by m y definition of out-m igrants. This w ould represent a problem for the interpretations of the results (or rather the applicability of the analysis) if this group: (i) system atically differ, in observed or unobserved characteristics, from the group of out-m igrants w ithout a N orw egian citizenship; and (ii) is large in
num bers. Regarding the first point I have no definite opinion, how ever regarding the latter I w ould expect that the group in question is relatively sm all in size. It should be noted that in order to becom e a N orw egian citizen one m ust have been in the country for at least
9 In light of the discussion of the 1980-90 period, the reference to 1992 m ay seem strange. H ow ever, the 1990 sam ple utilized in Longva and Rauum (2001a, 2001b), is selected conditional of their presence in N orw ay tw o years later, in 1992. Thus, as residence in N orw ay is crucial for the classification of out- m igrants, I find it m ost appropriate to refer to 1992 in the presentation of this sam ple.
10 This is part of the data set utilized in Barth et. al. (2000).
seven years. It is w orth noting that a num ber of studies use only citizenship as the criterion for the classification of im m igrants, w hich yields an inferior sam ple as a num ber of foreign born individuals naturalize.11 A s I have available both country of origin and citizenship for the m ajority of the sam ple in 1980, I am able to com pare these tw o
classification rules, see A ppendix 1. Lastly, I divide each sam ple of im m igrants into those w ho w ere born in a O ECD country, and those born in a country outside the O ECD .12 N atives constitutes the residual, i.e. those born in N orw ay or in a foreign country w ith at least one N orw egian born parent.
O ut-m igrant
O ut-m igrants are defined by a sam ple attrition procedure. That is, an individual in the 1980 sam ple is defined as an out-m igrant if he/she is not present in 1992. Correspondingly in the 1993 sam ple, w here the classification is done based on the residence in 1997. O ne possible source of bias is that I w rongly classify those w ho die betw een the tw o points of observation, as out-m igrants. A sim ple correction could be done using the publicly
available statistics of the m ortality rate in N orw ay by age and gender.13 H ow ever, the m ain focus of this paper is on the correlation betw een the propensity to out-m igrate and
individual earnings, hence w ithout any inform ation about the correlation betw een m ortality and earnings, I doubt w hether such a correction w ould influence the m ain results. A lso, it w ill be show n that the num ber of out-m igrants are very large com pared to the num ber of deaths at the present m ortality rate.14 H ence, I have not prioritized such an adjustm ent.
N ote that the different sam pling fram e in the tw o data sets allow s us to capture the out- m igrants defined by tw o different lengths of stay. That is, if w e capture out-m igrants by sam ple attrition, w e then select the group by a 12 and 4 years-of-stay-lim it for the m ost
11 For instance O ECD (2000) use citizenship to calculate the com parable share of im m igrants in all European O ECD countries.
12O ECD (N ordic, w ith the exception of N orw ay, O ECD -Europe as of 1990, N orth-A m erica, A ustralia/N ew Zealand), Non-O ECD (Eastern-Europe, A sia, A frica, Latin-A m erica).
13 Thanks to Tore Schw eder for pointing this out.
14 The annual average m ortality rate is about 0.37 percent per 1999, SSB (2000). Interestingly,
Schoeni (1997) reports som e evidence of low er m ortality am ong foreign-born in the U .S. com pared to the U .S. born.
recent cohort in the 1980 and 1993 set respectively. H ow ever, the term length-of-stay m ust be used w ith caution, as the im m igrants have spent a different num bers of years in N orw ay w hen I first observe them in 1980 and 1993. For instance, from the 1980 sam ple the subsam ple of out-m igrants w ill consists of those in the k cohort of im m igrants w ho spend betw een (1980-k) and (1993-k) years in N orw ay.15 Thus, for the different im m igrant cohorts w e capture the out-m igrants at different stages of their stay. This is im portant to have in m ind w hen interpreting the results.
Labor force participation
A s I eventually w ould study labor m arket earnings, I find it appropriate to categorize each individual according to its labor m arket status. M ost studies use a cutoff-point in earnings to select the sam ple (including Longva and Raaum , 2001a; 2001b). H ow ever, such a procedure is not ideal as the sam ple selection criteria is highly correlated w ith the
dependent variable under study. The 1980 Census data includes inform ation about hours of w ork w hich allow s m e to avoid the cutoff-point procedure. Thus, in the 1980 sam ple I define an individual as a m em ber of the labor force if the individual: (i) reported that they w orked 100 hours or m ore during the Census year; and (ii) is registered w ith nonzero earnings.16 A s w e m iss inform ation about the actual num ber of hours w orked in the 1993 data set w e are forced to use an incom e criterion to assign the crucial labor force status.
The annual threshold level is, set to approxim ately the average m onthly earnings for full tim e w orkers.
Age, Student, Self-em ployed
In order to m inim ize the im pact of the possible bias caused by the inability to distinguish betw een out-m igration and death, I have chosen to restrict the sam ple to those w ho are (or w ould have been) 64 years of age at the right endpoint in each sam ple. H ence, in the 1980 sam ple I select those aged 18-52, and in the 1993 sam ple I select those aged 18-60. Lastly, I throw out all students, as w ell as all registered as self-em ployed from the 1980 sam ple,
15 I here ignore the possibility that the im m igrant m ay out-m igrate, and then im m igrate again betw een the tw o points of observation.
16 Interestingly, the average annual earnings am ong the individuals categorized as not in labor force by m y selection rule are approxim ately equal to the cut-off point used in Longva and Raaum (2001a, 2001b), w hich is the average m onthly earnings for full tim e w orkers.
w hich excludes about 7 per cent of the natives, and 5 per cent of the im m igrants.
U nfortunately, I am not able to apply this selection criteria on the 1993 sam ple.
G iven the com plex data structure, and the large num ber of variables available, I face a num ber of difficult decisions regarding the em pirical approach. The literature offers few guidelines in this respect. A s discussed in section 3 the few studies w hich have been undertaken have designed the analysis according to specific structure of their data. In light of this I have chosen a rather broad em pirical approach, focusing m ore on providing basic figures, rather than estim ating heavily param eterized m odels. For instance, I have chosen not to undertake separate analyses for each gender as I am not convinced that a separate analysis is justified, given the other alternatives. For instance, a separate analysis by country of origin could potentially, from m y point of view , provide as m uch insight.
Table 1 provides som e basic figures according to the chosen classifications. To sim plify the exposition I denote out-m igrants as M overs and the residual as Stayers. For instance, 12 825 individuals from O ECD countries are observed outside the labor force in 1980.
A m ong these, 46 per cent are not observed in 1992 and thus classified as m overs, and 69 per cent are fem ales. W hile N on-O ECD im m igrants out-num bered the O ECD
im m igrants in the 1980 sam ple, it is opposite in the 1993 sam ple. This is due to the large inflow of N on-O ECD im m igrants during the 1980’s and underscores the im portance of taking account of the region of origin distribution in the em pirical analysis. O therw ise w e risk to confound the findings w ith pure com position effects due to the dram atic shift in the population structure. Lastly, by any standard the sam ple sizes, are large w hich enables a rather detailed specification.17 Com paring the sam ple sizes by labor force status w e observe that the overall labor force participation rate is clearly higher am ong natives com pared to the tw o groups of im m igrants. This finding is in line w ith H usted et. al.
(2000a) w hich finds that that the unem ploym ent rate of im m igrants in D enm ark is m uch higher than for D anish born. The high share of im m igrants outside the labor force points to the im portance of not neglecting this group.
17 For instance, D ustm ann (1993) observe a total of 1 064 im m igrants w hile D ustm ann (1996a) apply a sam ple of 6 901. Edin (2001) observe a 3 per cent sam ple each year from 1970 to 1990, w hich adds up to 15 574 im m igrants. The sam ple size in H usted et. al. (2000b) is not reported.
Table 1. Sam ple sizes, share m overs and share fem ales.
1980 Sam ple 1993 Sam ple
Country of origina Size M overs Fem ales Size M overs Fem ales
(share) (share) (share) (share)
Not in the labor force
N orw ay 23 145 .00 .82 90 120 .02 .54
O ECD 12 825 .46 .69 14 196 .30 .54
N on-O ECD 5 155 .26 .63 33 951 .14 .51
In the labor force
N orw ay 95 470 .00 .44 340 893 .01 .48
O ECD 28 244 .25 .46 33 963 .10 .50
N on-O ECD 11 412 .12 .29 37 460 .05 .36
a For non-residents 1992, in the 1980 sam ple, the country of origin is based on the citizenship.
Turning to the extent of out-m igration w e observe, as expected, that only a very sm all fraction, 1-2 per cent, of the native sam ple leave the sam ple betw een 1993 and 1997. The finding of zero attrition in the native sam ple betw een 1980 and 1992 is due to the
construction of the data set, as explained in A ppendix 1. The out-m igration behavior for the im m igrants, as represented by the share of m overs, could be sum m arized as follow s:
(i) in the 1980 sam ple the share of out-m igrants am ong those outside the labor force is about the double com pared to the share am ong those w ithin. In the 1993 sam ple the share is roughly three tim es as high for those outside com pared to those w ithin; (ii) the share of out-m igrants is roughly the double am ong O ECD im m igrants com pared to am ong N on- O ECD im m igrants; and (iii) the share in the 1980 sam ple is higher than in the 1993 sam ple. That the overall propensity to leave the sam ple is higher in the 1980-92 sam ple com pared to in the 1993 sam ple could easily be explained by the difference in the tim e span betw een the tw o observations over tim e used to identify the out-m igrants. W hile I
define those w ho disappear from the sam ple during the follow ing 12 years in the 1980 sam ple, the corresponding num bers of years is 4 in the 1993 sam ple.
A lthough the large share of out-m igrants am ong the O ECD im m igrants could in principle be caused by som e kind of com position effect, it should not be controversial to state that O ECD im m igrant cross the borders m ore frequently than the N on-O ECD im m igrants. O ne com m on explanation could be found in the im m igration law s w hich allow s in principle unrestricted m ovem ent w ithin the N ordic countries and the EU , another in the sm all cultural distance betw een N orw ay and m any w estern countries. M ore puzzling, a larger percentage of the individuals outside the labor force out-m igrate com pared to the individuals in the labor force. Taken at face value, this could be interpreted as an early indication that those w ho do not succeed in the labor m arket leave. This finding is in line w ith Edin et. al. (2001). H ow ever, there are several objections to such an interpretation.
First, people m ay com e for other reasons than labor, for instance to take education, due to fam ily unification, or as refugees. Second, m y m easure of earnings is not necessarily a good m easure for success in the labor m arket.
Regarding the gender distribution I find that fem ales are over-represented outside the labor force in the early sam ple (1980), w hile the distribution is m ore equal in 1993. This is partly due to an overall increase in the labor force participation am ong fem ales, and partly due to the severe w orsening in the labor m arket betw een these tw o points in tim e, w hich affected a num ber of the em ployed m ales. It should also be noted that the share of fem ales is low am ong the N on-O ECD im m igrants com pared to O ECD im m igrants and natives.
A lso (not show n), the tendency to outm igrate is m uch stronger am ong m en com pared to w om en. W e know that m en typically are the m ain breadw inners in the fam ily, and hence w ould be over-represented am ong short term labor m igrants, w hile w om en on the other hand are over-represented am ong the fam ily-reunification m igrants.
A s rem arked upon earlier in this section I observe the im m igrant cohorts at different stages of their stay. This m otivates a look at the share of m overs for each arrival cohort. Table 2 provides an overview for the 1993 sam ple, using 5-year cohorts, by labor force status. For instance, am ong the im m igrants born in an O ECD country, not in the labor force in 1993, w ho arrived before 1965, 5 percent (first row , first colum n) are not observed in 1997 and thus classified as m overs. It seem s like the longer the im m igrant has been in N orw ay as of
1993 the less chance for outm igration during the next four years. This can be interpreted as a cohort effect, a years-since-im m igration effect, or a com bination. In the first case the increasing share of out-m igrants as the arrival tim e get closer to 1993 is due to inherent differences betw een the arrival cohorts, w ith the early cohort characterized by a low out- m igration propensity, and the m ore recent cohort characterized by a higher out-m igration propensity. In the latter case the pattern is interpreted as a declining out-m igration
propensity as the length of the stay increases, w hich is w hat w e w ill observe if the arrival cohorts consists of a m ix of short-term and long-term im m igrants.18 The pattern for O ECD im m igrants is very strong. W hile only five percent of the rem aining im m igrants from the pre 1964 arrival cohort outside the labor force out-m igrated during 1994-97, the
corresponding num ber is alm ost sixty percent for the m ost recent cohort. The pattern for those in the labor force is also clear, especially from the 1980-84 cohort on.
Table 2. Share m overs, by arrival cohort, 1993 sam ple.
Not in the labor force 1993 In the labor force 1993
Cohort O ECD N on-O ECD O ECD N on-O ECD
A rrived pre. 1965 .05 .07 .02 .03
A rrived 1965-69 .08 .11 .03 .03
A rrived 1970-74 .14 .10 .04 .03
A rrived 1975-79 .21 .09 .05 .03
A rrived 1980-84 .26 .11 .08 .04
A rrived 1985-89 .39 .14 .13 .06
A rrived 1990-92 .59 .18 .34 .10
18 This process could be illustrated w ithin a sim ple form al fram ew ork m odeling the stock and outflow of im m igrant cohort over tim e.
5. M overs versus Stayers
The em pirical analysis w ill proceed as follow s. I start out w ith som e basic descriptive statistics follow ed by a sim ple m ultivariate analysis of the earnings differential betw een m overs and stayers. A s the m ain focus is on earnings, this part of the analysis is
undertaken on the subsam ple of labor force participants. N ext I provide an analysis of the discrete choice of w hether to out-m igrate or not.
D escriptive statistics
Table 3 provides descriptive statistics for those in the labor force by out-m igration status, separately for the tw o sam ples. First, looking at the earnings statistics, w e observe that natives on average earned less than the im m igrants w ith the exception of the group of N on-O ECD im m igrants in 1993. This som ew hat peculiar observation w as also found in Longva and Raaum (2001b), but w as to som e degree explained by a different age- and gender structure in the sam ples. N evertheless, w ith the m entioned exception, these sim ple figures paint a fairly positive picture of im m igrant labor m arket behavior. Taken at face value, and by using earnings as a m easure of the im m igrants contribution to the econom y (Borjas, 1999), the high m ean earnings provide som e indications of a positive contribution to the econom y from im m igrants participating in the labor force.19
Turning to the earnings differential betw een m overs and stayers w e observe that the m ean annual earnings am ong O ECD m overs are w ell above the corresponding statistics for stayers. H ow ever, the reported percentiles show that the earnings distribution am ong the m overs is skew ed to the extrem es, w ith the 90 percentile roughly tw ice as large as the corresponding figure for the stayers, and w ith the 10 percentile below that of the stayers.
Thus it seem s like O ECD m overs consist of a num ber of high-earnings individuals, m ixed w ith m overs w hich have annual earnings clearly below that of the stayers. The earnings distribution am ong the N on-O ECD m overs is m ore in line w ith the stayers, at least in the upper part of the earnings distribution. H ow ever, both the m edian and the 10 percentile are clearly below that of m overs, in both sam ples. Thus it seem s like
19 O f course, such a statem ent is build on a num ber of strong assum ptions and prem ises, w hich I w ill not go into in this paper (Borjas, 1999). H ow ever, from m y point of view it is striking that w e observe a positive earnings gap for large groups of im m igrants in N orw ay.
Table 3. M eans, individuals in the labor force.
O ECD N on-O ECD
V ariable N atives Stayers M overs Stayers M overs
1980 Sam ple
Earningsa, m ean 153 169 261 163 171
Earningsa, 90 percentile 248 273 523 247 255
Earningsa, 50 percentile 154 162 152 161 151
Earningsa, 10 percentile 49 52 44 70 57
Log Earnings 11.74 11.81 11.89 11.86 11.79
Fem ales, share 0.44 0.49 0.37 0.30 0.20
A ge, in years 33.85 37.67 31.34 33.71 32.32
Education, in yearsb 10.33 11.43 11.85 11.15 10.98
Education m issing, share 0.01 0.14 0.22 0.35 0.46
Y ears since im m igration - 11.35 - 7.67 -
1993 Sam ple
Earningsa, m ean 164 190 258 130 141
Earningsa, 90 percentile 271 317 527 226 245
Earningsa, 50 percentile 160 175 174 124 97
Earningsa, 10 percentile 46 64 49 33 25
Log Earnings 11.81 11.96 12.03 11.55 11.42
Fem ales, share 0.48 0.51 0.44 0.37 0.30
A ge, in years 37.23 41.74 36.81 34.98 34.29
Education, in yearsb 11.58 12.57 12.95 11.43 12.35
Education m issing, share 0.01 0.43 0.79 0.52 0.55
Y ears since im m igration - 15.87 7.20 10.01 7.58
aThousand 1990 N O K
bA m ong those w ith registered educational attainm ent
the m overs on average are draw n from the low er part of the earnings distribution am ong the N on-O ECD im m igrants.
Further, the m overs are younger than the stayers, especially am ong the O ECD im m igrants.
A t face value the m overs are m ore educated than the stayers am ong O ECD im m igrants, and less educated am ong N on-O ECD im m igrants. H ow ever, I m iss inform ation on educational attainm ent for a large share of the im m igrant groups, especially for the N on- O ECD m overs w here I m iss inform ation for nearly half of the group. This fact hints at that w e should be careful in using educational attainm ent as a regressor later on. Lastly, w e note that the average O ECD im m igrants has spent less tim e in N orw ay com pared to the average N on-O ECD im m igrant. From the 1993 sam ple w e observe that the m overs, as expected, have spent less tim e in N orw ay com pared to the stayers.
M ultivariate analysis
This exercise is m otivated by the different distributions of socioeconom ic variables by out-m igration status, and by country of origin, as show n in Table 3. Specifically I w ould like to investigate w hether the presented raw earnings differentials by country of origin, as w ell as by out-m igration status, could be explained by differences in the distribution of m arital status, gender, age and education. I estim ate the follow ing m odel by ordinary least squares, separately for each sam ple (1980 and 1993):
0 1 2 *
i i i i i i
y =α α+ I+α I O U T+X β ε+ (1)
w here yi denotes the natural logarithm of annual earnings for individual i,I is a dum m y variable for im m igrant status, O U T is a dum m y variable w hich takes the value 1 if the individual is classified as a m over, 0 otherw ise. X is a vector of sosioeconom ic variables, andα0,α1,α2 and β are unknow n param eters. B y the specification in (1) w e restrict the variables in X to have the sam e im pact on log earnings, independent of region of origin, w hich is a highly questionable assum ption in light of the presum able very different underlying return to observable variables, for instance educational attainm ent. H ow ever, for the case of sim plicity I stick to this restriction. A lso, note that in light of the earnings
distributions presented in Table 3, the use of a logarithm ic transform ation of the dependent variable is not unproblem atic.20
Table 4 provides estim ates of the im m igrant dum m ies and the interaction term s, obtained by estim ations of (1) on individuals in the labor force, in 1980 and 1993 respectively.
These estim ates are approxim ations of the group specific m ean earnings, relative to natives. The controls included are indicated in the bottom of the table. For instance, in 1980, from colum n 5, w e observe from the first row that the average stayer from an O ECD country earned 2.6 percent m ore than natives, controlling for the differences in the
distribution of gender, age and m arital status. The second row is accordingly interpreted as the earnings differential betw een O ECD stayers and m overs, evaluated at the m eans (in the 1980 sam ple) of the above m entioned socioeconom ic variables.
From the 1980 sam ple w e observe from colum n (1) and (2) that the earnings advantage for the group of im m igrants from N on-O ECD countries disappear once w e control for som e basic individual characteristics. The earnings differentials in colum n (3) are identical to those that result from a com parison of the m ean log earnings presented in Table 3.
Controlling for gender w e find that im m igrants from O ECD countries earn m ore than im m igrants from N on-O ECD countries, both am ong stayers and m overs. Colum n (6) presents the differentials after controlling for the full set of variables. W e observe that w e end up w ith a structure sim ilar to w hat w e found by com paring the raw m eans, nam ely that the m overs earn m ore (less) in 1980 than the stayers am ong the O ECD (N on-O ECD ) im m igrants. The results for the 1993 sam ple show that m uch of the earnings advantage of O ECD im m igrants over natives could be explained by socioeconom ic characteristics.
Further, by controlling for the full set of variables, the other differentials, as observed in colum n (3), becom e stronger. For instance, w hile the average N on-O ECD m over earned 13 per cent less than the average N on-O ECD stayer, this differential increases to 18.8 per cent after controlling for gender, age, m arital status and education.
20 The w idely used log transform ation of earnings in the literature is norm ally justified by tw o factors: (i) the distribution of earnings is positively skew ed such that the distribution of log earnings is closer to the norm al distribution; (ii) the transform ation sim plifies the interpretations of the estim ated coefficients from the regressions. Regarding (i) no such regularity is observed in the subsets of stayers and m overs, judging from Table 3. In spite of this I have undertaken the transform ation as this greatly sim plifies the interpretation of the estim ates.
Table 4. Estim ates of the im m igrant dum m y and interaction w ith out-m igration status, individuals in the labor force.
V ariable (1) (2) (3) (4) (5) (6)
1980 Sam ple
O ECD .090 .057 .070 .109 .026 -.032
(.005) (.004) (.006) (.005) (.005) (.005)
O ECD *O U T .081 -.008 .119 .114
(.010) (.009) (.009) (.009)
N on-O ECD .108 -.017 .116 .020 -.005 -.034
(.007) (.006) (.008) (.007) (.007) (.007)
N on-O ECD *O U T -.073 -.145 -.102 -.095
(.022) (.015) (.019) (.018)
1993 Sam ple
O ECD .161 .048 .154 .165 .035 -.002
(.004) (.003) (.004) (.004) (.004) (.004)
O ECD *O U T .074 .049 .125 .148
(.013) (.012) (.011) (.011)
N on-O ECD -.266 -.337 -.259 -.301 -.328 -.317
(.004) (.003) (.004) (.004) (.003) (.004)
N on-O ECD *O U T -.130 -.155 -.169 -.188
(.016) (.016) (.014) (.014)
Controls
gender X X X X
age X X X
m arital status X X X
education X
Notes. Standard errors in parentheses. G ender is entered as a dum m y for fem ale. A ge is entered as a fourth- order polynom ial. M arital status is entered as a dum m y for m arried and a dum m y for previously m arried.
Education is entered as the num ber of years for those w ith registered education, and a dum m y for education m issing. The com plete table of the results is available from the author.
To sum m arize, m uch of the observed difference in average earnings betw een natives and O ECD im m igrants as w ell betw een natives and N on-O ECD im m igrants in 1980 can be explained by socioeconom ic and dem ographic characteristics. M y earlier observations w ith regards to the earnings differentials by region of origin, and by out-m igration status, are upheld.
The decision to out-m igrate
In order to m ore precisely characterize how m overs differ from stayers I undertake a probit analysis, in line w ith H usted et. al. (2000b) and D ustm ann (1996a). Form ally, the decision to out-m igrate is m odeled as a binary choice of w hether to stay or not, w ith the probability of not staying given by:
Pr(out-m igrate)= Φ(Zγ) (2)
w here Z is a vector of variables, influencing the out-m igration decision, γ is an unknow n vector of param eters, and Φ is the standard norm al distribution function. (2) is estim ated separately for O ECD and N on-O ECD im m igrants, Table 5 gives the results for the 1980 sam ple. For sim plicity I present only the result w ith the full set of variables included, as a stepw ise introduction of the variables did not provide any m ajor new insights. If a variable enters positively in the table it has a positive effect on the probability to out-m igrate. In order to sim plify the interpretations I have calculated the m arginal effect, M Ej, for variable j given by:
( ˆ ˆ)
j j
M E =φ γ γZ (3)
w hereφ is the norm al density function, interpreted as the change in out-m igration propensity of a sm all change in variable j, calculated at sam ple m eans.
A ge enters negatively for O ECD im m igrants, w ith a m arginal effect of m inus 1.2 percent.
That is, the older the im m igrant is, the low er is the probability of out-m igration during 1981-92, and being 10 years older reduces the probability of out-m igration by 12 percent.
O ne interpretation is that older im m igrants have spent a larger num ber of years in the host country and thus are m ore established com pared to younger im m igrants. A lso, the group of O ECD out-m igrants could be dom inated by young, highly m obile, labor m igrants. This
Table 5. Probit analysis of the propensity to leave N orw ay 1981-92, by region of origin.
O ECD N on-O ECD
V ariable, 1980 Est. St.dev. M E Est. St.dev. M E
Intercept 1.417 .065 .420 -1.283 .118 -.239
A ge -.041 .001 -.012 -.001 .002 .000
Fem ale -.329 .018 -.098 -.212 .038 -.040
M arried -.510 .023 -.151 -.237 .041 -.044
Previously m arried -.407 .036 -.121 -.273 .080 -.051
N ever m arried ref. - - - - -
Education, years .001 .003 .000 -.002 .007 .000
Education, m issing .227 .045 .067 .152 .079 .028
N ordic ref. - - - - -
W estern-Europe -.232 .020 -.069 - - -
N orth-A m erica .470 .027 .140 - - -
Eastern-Europe - - - ref. - -
A sia - - - .330 .046 .061
A frica - - - .321 .060 .060
South/M iddle A m . - - - .324 .070 .060
Earnings, 1.quartilea -.081 .025 -.024 .211 .045 .039
Earnings, 2.quartilea -.270 .026 -.080 .032 .047 .006
Earnings, 3.quartilea -.308 .026 -.091 -.062 .049 -.012
Earnings, 4.quartilea ref. - - ref. - -
Log-likelihood -13548.2 -3975.6
Num ber of observations 28 244 11 412
Num ber of out-m igrants 7 105 1 338
Notes. M E=M arginal effects, calculated at sam ple m eans. The dependent variable takes the value 1 if the individual is not residence in N orw ay in 1992 (out-m igrant), 0 otherw ise.
a Q uartiles defined by the gender specific earnings distribution for the group under study
