Return Migration, Investment in Children, and Intergenerational Mobility: Comparing Sons of Foreign and Native Born Fathers

Transcription

1 DISCUSSION PAPER SERIES IZA DP No Return Mgraton, Investment n Chldren, and Intergeneratonal Moblty: Comparng Sons of Foregn and Natve Born Fathers Chrstan Dustmann September 2007 Forschungsnsttut zur Zukunft der Arbet Insttute for the Study of Labor

2 Return Mgraton, Investment n Chldren, and Intergeneratonal Moblty: Comparng Sons of Foregn and Natve Born Fathers Chrstan Dustmann Unversty College London, CReAM, CEP and IZA Dscusson Paper No September 2007 IZA P.O. Box Bonn Germany Phone: Fax: E-mal: za@za.org Any opnons expressed here are those of the author(s) and not those of the nsttute. Research dssemnated by IZA may nclude vews on polcy, but the nsttute tself takes no nsttutonal polcy postons. The Insttute for the Study of Labor (IZA) n Bonn s a local and vrtual nternatonal research center and a place of communcaton between scence, poltcs and busness. IZA s an ndependent nonproft company supported by Deutsche Post World Net. The center s assocated wth the Unversty of Bonn and offers a stmulatng research envronment through ts research networks, research support, and vstors and doctoral programs. IZA engages n () orgnal and nternatonally compettve research n all felds of labor economcs, () development of polcy concepts, and () dssemnaton of research results and concepts to the nterested publc. IZA Dscusson Papers often represent prelmnary work and are crculated to encourage dscusson. Ctaton of such a paper should account for ts provsonal character. A revsed verson may be avalable drectly from the author.

3 IZA Dscusson Paper No September 2007 ABSTRACT Return Mgraton, Investment n Chldren, and Intergeneratonal Moblty: Comparng Sons of Foregn and Natve Born Fathers * Ths paper studes parental nvestment n educaton and ntergeneratonal earnngs moblty for father-son pars wth natve and foregn born fathers. We llustrate wthn a smple model that for mmgrants, nvestment n ther chldren s related to ther return mgraton probablty. In our emprcal analyss, we nclude a measure for return probabltes, based on repeated nformaton about mgrants' return ntentons. Our results suggest that educatonal nvestments n the son are postvely assocated wth a hgher probablty of a permanent mgraton of the father. We also fnd that the son's permanent wages are postvely assocated wth the probablty of the father's permanent mgraton. Keywords: Intergeneratonal moblty, return ntentons, educatonal nvestment, earnngs. JEL Classfcaton: J15, J24, J62 Keywords: ntergeneratonal moblty, return ntentons, educatonal nvestment, earnngs Correspondng author: Chrstan Dustmann Economcs Department Unversty College London Gower Street London WC1E6BT Unted Kngdom E-mal: c.dustmann@ucl.ac.uk * I am grateful to Davd Card for many helpful dscussons, and to Jerome Adda, Teresa Casey, Jeff Grogger, Steve Machn, Gary Solon and two anonymous referees for comments. Ths research was funded by the Economc and Socal Research Councl (grant RES ).

4 Ι Introducton Dustmann 2 Immgrants contrbute sgnfcantly to the overall economc performance of ther host economes. It s therefore not surprsng that a large lterature s concerned wth the earnngs moblty of the foregn born populaton, both n solaton, as well as n comparson wth those who are natve born. But mmgrants not only have an mmedate effect on wealth accumulaton and earnngs and skll composton. They transmt ther earnngs status, as well as soco-economc and cultural characterstcs to the next generaton. The economc adjustment process wthn the mmgrant's own generaton has long been recognsed as an mportant step n understandng the economc effects of mmgraton. In understandng the long term consequences of mmgraton, assessment of ntergeneratonal moblty n mmgrant communtes s perhaps equally mportant. Whle the process of ntergeneratonal economc moblty has been ntensvely studed for majorty populatons (see among others Solon 1992; Zmmermann 1992; Bjoerklund and Jantt 1997; Plug 2004; Bjoerklund, Lndahl and Plug 2006; see Solon 1999, 2002 for revews), less s known about ntergeneratonal transmsson n mmgrant communtes. For the U.S., early studes by Chswck (1977) and Carlner (1980) compare earnngs of mmgrants wth ther descendants and ther chldren. More recently, Borjas (1992, 1994) emphasses that ntergeneratonal economc moblty among mmgrants may be more complex. He argues that estmatng the usual models of ntergeneratonal ncome moblty may mss out an mportant aspect of ths process: the sklls of the next generaton may not only depend on parental nputs, but also on the qualty of the ethnc envronment of the parent generaton. Borjas terms ths ethnc captal. In later work Borjas (1995) shows that one reason for the external effects of ethncty s segregaton nto partcular neghbourhoods - a pont that has been re-emphassed n work by Nelsen, Rosholm, Smth and Husted (2001).

5 Dustmann 3 Ths paper focuses on another mportant characterstc of mmgrant communtes that may affect the process of ntergeneratonal moblty: the probablty the mmgrant attaches to a permanent mgraton as opposed to a future return to the home country. There are a number of papers that show that return mgraton may affect dfferent aspects of mmgrants' behavour. Work by Galor and Stark (1990) suggest that postve return probabltes may affect savngs behavour. Dustmann (1997) provdes evdence that marred mmgrant women whose husbands plan to return to ther home countres have a hgher labour force partcpaton rate. Dustmann (1999) develops a model that suggests that mmgrants who have hgher probabltes of returnng are less lkely to acqure human captal specfc to the host country economy f ths human captal has a lower return back home. He fnds evdence for ths by nvestgatng ther nvestment n language sklls. In a recent paper, Cortes (2004), argung along the same lnes, explans the hgher rate of human captal accumulaton by refugee mmgrants n the US wth lower probabltes of ther return mgraton. Smlar consderatons may hold n an ntergeneratonal context. In ths paper, we nvestgate how parental probabltes of a permanent as opposed to a temporary mgraton affect nvestment n ther chldren's educaton, and ntergeneratonal earnngs moblty. We develop a model of ntergeneratonal moblty wth parental nvestments n the chld's earnngs potental, based on prevous work by Becker and Tomes (1979) and Solon (2004), whch we extend by allowng the probablty of a permanent mgraton to affect parental human captal nvestments. The model suggests that, f returns to educatonal nvestment undertaken n the host country, are hgher n the host than n the home country, and f the parent beleves that the chld's return probablty ncreases n hs own, educatonal nvestments should ncrease wth the permanent mgraton probablty, as should the chld's permanent earnngs. Our emprcal analyss s for Germany, and based on detaled data from a long panel that oversamples mmgrants. We concentrate on father-son pars. The data ncludes unque

6 Dustmann 4 nformaton on parental return ntentons whch we use to construct our measure for a permanent mgraton probablty. To understand the precse nature of how parental background may affect earnngs n the next generaton, we commence by estmatng nvestment equatons, relatng educatonal nvestments to parental earnngs, as well as return probabltes. We fnd strong and consstent evdence that, for mmgrants, parental nvestment n educaton ncreases wth the permanent mgraton probablty. Estmaton of nvestment equatons that relate educatonal achevements to permanent parental earnngs show estmates of smlar magntude for mmgrants and natves. For earnngs moblty, we agan fnd that the son's permanent earnngs ncrease n the father's permanent mgraton probablty, condtonal on father's permanent earnngs and educaton. We also fnd that educatonal achevements of mmgrant parents are not correlated wth educatonal achevements of ther chldren. Ths s n contrast wth prevous work by Card, DNardo and Estes (1998) for the U.S., who show that much of the ntergeneratonal lnk between the economc status of mmgrant fathers and ther sons and daughters works through educaton. Our fndngs are n lne wth smlar results reported by Gang and Zmmermann (2000). One explanaton for ths weak lnk n educatonal outcomes for mmgrants s that t s the permanent earnngs poston of the father that matters for the son's educatonal attanment; the father's educaton s then only correlated wth the son's educaton f educaton s a good predctor for earnngs. We show that n our sample, ths s not the case for mmgrants. We also estmate ntergeneratonal moblty coeffcents. Our analyss extends some prevous work on Germany by Couch and Dunn (1997) and Wegand (1997) for the natve born, but provdes more robust estmates based on a longer panel, allowng us to address the problem of measurement error n permanent parental earnngs whch bedevls studes of ntergeneratonal transmsson (see Solon 1989). In addton, we dstngush between

7 Dustmann 5 mmgrant and natve father-son pars. Our results reveal that ntergeneratonal moblty between natve born fathers and ther sons s larger than between foregn born fathers and ther sons, condtonal on return probabltes of foregn born fathers, although the dfference s not statstcally sgnfcant. The paper s structured as follows. In the next secton we develop a theoretcal model, and dscuss ts emprcal mplcatons. In secton 3 we descrbe the data and the sample. Secton 4 presents the results, and secton 5 concludes. ΙΙ Theory Our model s a permanent ncome model of ntergeneratonal moblty wth parental nvestments n the chld's earnngs potental, followng early work by Becker and Tomes (1979) and Solon (2004). It extends Solon (2004) by takng account of the way the probablty of a permanent mgraton as opposed to a future return, may affect the decson of the parent to nvest n ther offsprng's human captal. We consder a one-person household wth one chld. There are two perods. In the frst perod (perod 0) both parent and chld lve n the host country. In the second perod (perod 1), the parent returns to the home country wth probablty 1 - p, and remans n the host country wth probablty p. The parent retres n perod 1, and has earnngs n perod 0 equal to y 0. The chld s n full tme educaton n perod 0, and partcpates n the labour market n perod 1, ether n the parent's home country, or n the host country. The parent s altrustc and maxmses an ntertemporal utlty functon, by choosng frst perod savngs s 0, and nvestment n the chld's human captal n the frst perod, I0. When makng these choces, the parent attaches the same probablty p to hs chld's locaton choce n the second perod as to hs own. Ths assumpton smplfes the algebra, but can be relaxed

8 Dustmann 6 wthout affectng the key mplcatons of our model, as long as the parent perceves the chld's probablty to reman n the host country to ncrease n hs own probablty of remanng. The parent's nter-temporal utlty functon s gven by (1) I 0 1 γ I E 1 1 V= uc ( ) + puc [ ( ) + vy ( )] + (1 p)[ uc (, b) + γvy ( ) E 1 where u(.) and v(.) are the parent's and the chld's utlty functons, defned over parental consumpton n perod 0 c 0, and parental consumpton and the chld's earnngs n perod 1, J c 1 and J y 1. Here J = E; I stands for emgraton or mmgraton country. The parameter γ s an altrustc weght. If γ = 0, the parent does not consder the chld's welfare n perod 1. The parameter b s a preference parameter, reflectng a preference for consumpton at home over consumpton abroad. If b > 1, more utlty s obtaned from consumng n the home country as compared to the host country. We assume that parental nvestments translate nto human captal of the chld (h 1 ) accordng to the followng producton technology: (2) h 1= θ log I0 + e0. The parameter θ s a technology parameter measurng the productvty of nvestments. The term e 0 s the human captal the chld receves wthout any drect parental nvestments (see Becker and Tomes, 1979 and Solon, 2004 for a smlar formulaton). Ths term represents the attrbutes endowed upon the chld, dependng on characterstcs of the parent, the chld's upbrngng, genetc factors, envronment, and luck. It may also depend on exstng networks, as well as the lack of opportunty to move out of socal and economc structures from one generaton to the next. Ths latter factor may be partcularly mportant for mmgrants, and we wll dscuss ts mplcatons later. It ncludes what Borjas (1992) calls "ethnc captal" - the qualty of the envronment n whch parental nvestments are made.

9 Dustmann 7 Human captal translates nto earnngs accordng to the followng relatonshp: j j j (3) log y1 = µ + r h1, where j = I ; E. Our formulaton allows for dfferent base wages µ, as well as dfferent returns to the chld's human captal r j n the two countres. It follows from (2) and (3) that the chld's earnngs n the second perod are related to parental nvestments as: j j j j (4) log y1 = µ + r θ log I 0 + r e0. The parent's consumpton n perod 0 equals c 0 = Y 0 - I 0 - s 0, where Y 0 s frst perod earnngs. As the parent retres n perod 1, perod 1 consumpton s equal to perod 0 savngs. Choosng a smple logarthmc utlty functon, and substtutng (4) for the chld's earnngs nto (1), the optmsaton problem of the parent can be expressed as I I (5) V = Y0 I0 s0 + p s0 + γ µ I + r θ I0 + r e0 si, max log( ) [log ( log )] E E + ( 1 p)[ b logs0 + γ ( µ E + r θ log I 0 + r e0 )]. Maxmsng (5) w.r.t. savngs and nvestment, and solvng the frst order condtons for the optmal nvestment I 0 yelds: I E γθ( pr + (1 p) r ) (6) I0 = Y0 γθ I E ( pr + (1 p) r )) + (1 + p + b(1 p)) E I = Γ( p; r, r, b, γ, θ ) Y0. The term n the numerator, whch s equal to the frst term n the denomnator, s the

10 Dustmann 8 expected utlty gan to one unt of parental nvestment n the chld's human captal. The second term n the denomnator s the expected lfetme utlty from one log unt of addtonal consumpton. Smple comparatve statcs show that nvestment n the chld's human captal ncreases n p, the probablty of a permanent mgraton, as long as the return to human captal nvestments s hgher n the host country (r I > r E ). Furthermore, a lower probablty of a permanent mgraton ncreases the expected gan n utlty by consumng n the home country, as long as b > 1: the parent prefers to save more resources for ther own future consumpton, thus reducng nvestment n the chld. The combned effect leads to an ncrease n nvestment wth the probablty of a permanent mgraton. Fnally, nvestment ncreases wth altrusm γ and productvty of nvestment θ. An estmable verson of (6) s obtaned by takng logs, and addng an error term: ' (7) log Ed = a1 + a2p 0 + X a + b log y0 + e, where Ed s a measure of chld's educatonal attanment, P s a varable measurng the probablty of a permanent mgraton, X are addtonal control varables, and y 0 are permanent earnngs of the father. Below we test whether nvestment nto chldren's educaton ncreases wth the probablty of a permanent mgraton. If r I > r E, we should expect a 2 to be postve. Now consder the relatonshp between the chld's earnngs and the parent's earnngs for chldren who are observed n the host country n perod 2. Substtutng (6) nto (4) and rearrangng terms yelds: I I I I I (8) log y1 = µ + r θ logγ + r θ logy0 + r e1. Equaton (8) s smlar to the relatonshp between son's earnngs and father's earnngs, as derved n Solon (1999, 2004), except for the term Γ whch, as we show above, ncreases n the

11 probablty of a permanent mgraton. Dustmann 9 An estmable verson of equaton (8) s gven by: (9) log y 1 = α1 + α 2P 0 + α 3k Dk0 + β log y0 + e0, K k = 1 where as before P s a varable measurng the probablty of a permanent mgraton of the parent, and y 1 and y 0 are permanent earnngs of the chld and parent. The varables D k0 are dummy varables for the orgn country k of ndvdual 's father. They capture, among other factors, dfferences n ethnc captal (see Borjas 1992). Agan, f r I > r E, α 2 should be postve. We wll test for ths below. ΙΙΙ Background, Data and Descrptves A.1 Background Between the md 1950's and 1973, the strong economc development n Northern Europe and the resultng demand for labour, led to a large nflow of mmgrants manly from the perphery countres of Europe, but also from Turkey, North Afrca, South Amerca and Asa. The man recevng countres were Belgum, France, Germany, the Netherlands, Swtzerland, and the Scandnavan countres. The West-German economy experenced a strong upward swng after 1955, accompaned by a sharp fall n the unemployment rate (between 1955 and 1960, the unemployment rate fell from 5.6 percent to 1.3 percent) and an ncrease n labour demand. Ths generated a large mmgraton of workers from Southern European countres and Turkey nto Germany. The percentage of foregn born workers employed n West Germany ncreased from 0.6 percent n 1957 to 5.5 percent n 1965, to 11.2 percent n 1973, and declned thereafter.

12 Dustmann 10 Blateral recrutment agreements were set up between Germany and Italy, Span, Greece, Turkey, Portugal, and Yugoslava n the 1950's and 1960's. Labour mgraton over ths perod was ntally consdered as temporary by both the mmgraton countres and the emgraton countres. Indvduals were not expected to settle permanently. The German recrutment polcy was based on the assumpton that foregn workers would after some years return to ther home countres. Stll, although return mgraton has been qute consderable (see Bohnng 1987), a large fracton of foregn born workers settled permanently. A.2 The Data and Sample The data we use for ths analyss stems from 19 waves of the German Soco-Economc panel (GSOEP ), whch s a household-based panel survey, smlar to the US Panel Study of Income Dynamcs (PSID) or the Brtsh Household Panel Study (BHPS). Intated n 1984, the GSOEP oversamples the then resdent mmgrant populaton n Germany, whch stems from the mgraton movement we have descrbed above. In the frst wave, about 4500 households wth a German born household head were ntervewed, and about 1500 households wth a foregn born household head. The data s unque n provdng repeated nformaton on a boost sample of mmgrants over a long perod of tme. For our analyss, we use observatons of the foregn born from the over-sample, as well as from the standard sample. Each ndvdual n a household and over the age of 16 s ntervewed. The household head provdes nformaton about all other ndvduals n the household and below the ntervewng age. Indvduals who leave households and form ther own households are ncluded n the panel. When ndvduals are 16 years old, they receve ther own personal dentfer, and ponters connect them to ther mother and ther father. We construct a sample of father-son

13 Dustmann 11 pars for foregn born and natve born fathers. We defne a second generaton mmgrant as an ndvdual who s born n Germany, and whose father s born abroad. The defnton of our sample s n the tradton of prevous studes on ntergeneratonal moblty based on the PSID (see, for example, Solon 1992). v Our reference group are ndvduals who are lkewse born n Germany, and whose father s German born. In our analyss we use only father-son pars. v We analyse below the relatonshp between permanent mgraton probabltes, permanent parental earnngs, and nvestment nto sons' educaton. We also analyse ntergeneratonal earnngs moblty, and how the son's permanent wages relate to the father's permanent mgraton probablty. Our analyss of educatonal achevements consders secondary track choce. Track choces n Germany are made at the age of 10, and determne whether ndvduals wll be enttled to study at unversty, or wll receve secondary educaton that enttles them only for vocatonal or apprentceshp based post-secondary educaton. Secondary track choce s strongly correlated wth later earnngs (see Dustmann 2004). For ths analyss, we nclude all sons above the age of 16, as ther track choce has already been made, even f they are stll n secondary educaton. We also analyse completed educaton, and dscuss results brefly n the text. Here we use only ndvduals above the age of 20, and who have completed educaton. For the analyss on earnngs moblty, we exclude sons who are younger than 20 years; our age range s between 20 and 34. A.3 Sample Characterstcs and Varables Our resultng samples contan a total of 795 sons born to natve born fathers, and 334 sons of foregn born fathers. Because some fathers have more than one son, we have 640 correspondng natve born fathers, and 251 foregn born fathers. In table 1 we provde a breakdown of fathers and sons n the two samples.

14 Dustmann 12 In table 2 we dsplay nformaton on the fathers' observable characterstcs. Foregn born fathers have about 2 years less educaton than natve born fathers. The age of fathers when the son was born s smlar, at about 30 years. Foregn born fathers have on average been n Germany for 5.5 years when ther son was born. As a measure for earnngs, we use real hourly log wages for both fathers and sons. The GSOEP provdes nformaton on average monthly gross earnngs n the month precedng the ntervew, and on actual weekly hours worked for pay durng that month. From that nformaton, and usng a consumer prce ndex, we compute an hourly real log wage rate n 2002 prces. A common problem n the lterature on ntergeneratonal moblty s the lack of suffcent nformaton for the estmaton of a permanent wage for fathers as well as ther sons, whch s partcularly severe n short panels. Zmmerman (1992), Solon (1989, 1992), and Dearden et al. (1997) among others, emphasse the problems of measurement error n parental earnngs for the estmaton of ntergeneratonal moblty parameters. The length of our panel allows us to address ths problem. Table A1 n the appendx dsplays vald wage spells of fathers n our sample, where the frst set of columns report numbers for foregn born fathers and the second set of columns for natve born fathers. There s a total of 19 possble years an ndvdual can report earnngs, correspondng to a total of 2,577 earnngs spells for foregn born fathers, and a total of 7,262 earnngs spells for natve born fathers. About 75 percent of fathers n the two samples have more than sx wage observatons, consttutng, for each group, more than 90 percent of all wage spells n the data. For sons, we dsregard earnngs spells whle beng on apprentceshp schemes. Apprentceshp schemes are vocatonal tranng schemes whch pay low and regulated wages, and these wages are not approprate measures for permanent earnngs status. As stated above, we also exclude wage spells when ndvduals were younger than 20, for smlar reasons. Table A2 n the appendx dsplays the wage nformaton for sons. Among those wth foregn born fathers,

15 Dustmann 13 there are 161 ndvduals wth 606 vald wage spells; among those wth natve born fathers, there are 373 ndvduals wth 1,824 vald wage spells. In table 3 we dsplay percentles as well as the mean and varance of average log real wages n our data. The entres n the table show that mean wages of natve born fathers are about 13 percent hgher than mean wages of foregn born fathers. For sons, ths dfference has reduced to 4 percent. Wages of both sons and fathers n the natve sample are consderably more dspersed than wages of sons and fathers n the sample of foregn born fathers. For natves, earnngs of sons are slghtly more dspersed than earnngs of fathers - whch s smlar to what s found n US studes usng smlar data (see Solon 1992). For the foregn born, however, fathers' earnngs seem to be more dspersed than sons' earnngs. Dfferences between sons of natve and foregn born fathers seem to be smlar throughout the dstrbuton, whle dfferences for natve and foregn born fathers are slghtly more substantal n the upper percentles of the dstrbuton. Permanent Earnngs To elmnate measurement error, we essentally follow the lterature whch averages wages over a number of years, thus ncreasng the sgnal-nose rato n earnngs nformaton (see for example Solon 1992 and Zmmerman 1992). Our method s slghtly more general, and allows the ncluson of ndvduals wth a mnmum number of wage spells, but observed n dfferent years, and wthout requrng subsequent vald spells. We do ths by estmatng fxed effects wage regressons, condtonng on a quadratc n age. Our regressons have the followng form: (10) 2 log w k k k t = a1 + a2age + a3age + v + ut,

16 Dustmann 14 where log w t are log real wages of ndvdual n perod t, v are ndvdual fxed effects, and u t are d error terms whch nclude measurement error. The ndex k s an ndex for the two groups of foregn and natve born ndvduals. We estmate equaton (10) separately for foregn born and natve born fathers. Uncondtonal on age, the sum of estmates k aˆ1 + vˆ s the mean wage for ndvdual n group k. Condtonng on age fxes ndvduals at the same pont n ther lfe cycle. aˆ k k k ˆ ˆ ˆ 1 + v + a2 age + a3 age 2 As for our measure of permanent earnngs, we predct v at age 40 for both natve and foregn born fathers. Whle estmates for k aˆ 2 and k aˆ 3 are unbased and consstent, the estmates for v are unbased but nconsstent for small t, and estmates of permanent earnngs wll suffer from measurement error f the sample contans ndvduals wth small t (that s ndvduals that have reported earnngs for a small number of years only). For our estmaton, we wll successvely ncrease the mnmum number of perods we requre ndvduals to have vald earnngs nformaton to nclude n the sample. The last rows of table 3 report predcted earnngs, and the dstrbuton of predcted earnngs for natve and foregn born fathers. Probablty of Permanent Mgratons To compute the probablty of a permanent mgraton, we use survey nformaton on the father's assessment of whether or not he wshes to return to the home country n the future. Our data s unque n provdng nformaton about these evaluatons. In each year between 1984 and 1995, mmgrants have been ntervewed regardng ther ntenton to ether stay permanently n Germany or to return home at some pont n the future. v These ntentons are lkely to be subject to measurement error; furthermore, permanent mgraton ntentons may also change over the mmgrant's lfe cycle. v To obtan a measure for the probablty that mmgrants may have assgned to a possble permanent mgraton when makng nvestment decsons about ther chld, we have frst coded ths nformaton nto a bnary

17 Dustmann 15 varable (assumng the value 1 when the response was 3: "I want to reman permanently"). Smlar to obtanng permanent earnngs measures, we then estmate fxed effects regressons, where we condton on years snce mgraton and years snce mgraton squared: 2 (11) P = b + b ysm + b ysm + ξ + e, t t where P t s equal to 1 f ndvdual reports n perod t the ntenton of a permanent mgraton, and ysm s a measure of the years snce mgraton of. The ξ are ndvdual fxed effects, and the e t are d error terms whch nclude measurement error. As our measure of a permanent mgraton probablty we compute ˆ 2 1+ ˆ ˆ ˆ, b b ysm% b ysm% ξ where ysm% and 2 ysm% are the father s years of resdence and years of resdence squared when the chld was ten years x old. In Germany, ths s the age when secondary track schools are decded, see Dustmann (2004) for detals. We then normalse ths varable between 0 and 1. We present the dstrbuton of the resultng varable n table 4. On average, the probablty of remanng permanently n Germany for fathers whose son s ten years old, s about 40 percent. Selecton ssues The sample of the foregn born father-son pars we use for analyss s one whch s selected - we observe more father-son pars where fathers have a hgher propensty to stay permanently. Ths nduces a selecton whch s correlated wth our measure for a permanent mgraton probablty: those wth a hgher probablty of a permanent mgraton (measured by past ntentons) wll be more lkely to be n the sample. If sons of those mmgrants who reman n the sample perform dfferently than sons of those who return (condtonal on father's permanent earnngs and permanent mgraton probablty), then ths wll bas the coeffcent of the return probablty. The bas wll be downward f resduals n the selecton equaton and the

18 Dustmann 16 ntergeneratonal moblty equaton are postvely correlated (ndcatng that sons of father-son pars who reman n the sample do better than sons of father-son pars who drop out of the sample, condtonal on father's earnngs and permanent mgraton probablty). The ntuton s that those who have a low probablty to reman permanently, but are nevertheless n the sample, must have unobserved characterstcs that are postvely related to the son's performance, whch leads to a reducton of the coeffcent estmate on the permanent mgraton probablty n absolute sze. x We can therefore nterpret the coeffcents on the permanent mgraton measure as a lower bound. ΙV Results B.1 Investment n Educaton We commence by estmatng nvestment equatons. Our theoretcal model relates nvestment n chldren to fathers' permanent earnngs, as well as to the probablty of a permanent mgraton. As a measure for nvestment, we use the son's educatonal outcomes. We concentrate here on secondary school track choce. Results on completed educaton are very smlar, and we brefly dscuss them. In Germany, age ten marks an mportant decson n the school career of chldren. At ths age, the chld transfers from prmary to secondary school, and, at the same tme, has to decde between three secondary school tracks: lower secondary (wth graduaton typcally at age 16), ntermedate secondary (wth graduaton typcally at age 16-17), and upper secondary (wth graduaton typcally at age 18-19). Although swtchng tracks s possble, t rarely occurs (see Pschke 1999 for evdence). Only hgh school allows for contnuaton to Unversty; lower and ntermedate secondary schools qualfy for blue collar and whte collar apprentceshp degrees. x

19 Dustmann 17 Intal track choce s therefore very mportant for future career prospects. Dustmann (2004) llustrates the strong correlaton between secondary school track, post-secondary educatonal achevements, and earnngs. x The man objectve of our analyss s to understand the relatonshp between the probablty the father assgns to a permanent mgraton, and the educatonal qualfcatons of the son. We also nvestgate the relatonshp between the father's permanent earnngs and school track enrollment of the son. And fnally, we nvestgate the possble correlaton between the chld's educatonal achevements and educaton of the father, and compare ths for mmgrants and natves. In table 5 we dsplay secondary school degrees for sons of natve and foregn born fathers. The numbers show that sons of natve born fathers have a hgher probablty of attendng hgher track schools than the sons of foregn born fathers. Whle about 64 percent of sons of natve born fathers attend at least an ntermedate secondary school, ths number s only 46 percent for sons of foregn born fathers. On the other hand, about 8 percent of sons of foregn born fathers do not complete secondary school tranng, whle ths s the case for only 3 percent of sons of natve born fathers. The numbers n the lower panel gve the total number of years of educaton for the two groups. There s a dfference n the number of years of full tme educaton between sons of natve and foregn born fathers of about 1.3 years, whch s a sgnfcant reducton compared to the fathers' generaton, where ths dfference was 2.4 years (see table 2). Our sample ncludes all sons above the age of 15, even f they are stll n secondary educaton, because the secondary school choce has been made before that age. We also report results for completed educaton, measured as the total number of years n full tme educaton for those who have fnshed full tme educaton. In table 6 we present results from an ordered probt model of secondary school choce.

20 Dustmann 18 We report results for sons of foregn born fathers (upper panel), and of natve born fathers (lower panel). As n table 5, we dstngush between four levels: no completed secondary school, lower secondary school, ntermedate secondary school, and hgh school. x In all regressons we condton on the son's brth cohort, the age of the father when the son was 10 years old, and (for foregn born fathers) on the country of father's brth (coeffcents are not reported). In the left panel, we present estmaton results of coeffcents on the father's permanent mgraton probablty only (columns 1), and when we condton n addton on the father's permanent log wage (where we use estmates based on a mnmum of fve wage spells), and hs years of educaton (columns 2). In table A3 n the appendx, we report the margnal effects of these varables on the probablty that the son does not obtan a secondary school degree (frst column), and acheves a hgh school degree (second column). These probabltes are calculated at sample means for each sample. The coeffcent on the measure of a permanent mgraton probablty s sgnfcantly dfferent from zero (table 6, columns 1). Uncondtonal on the father's permanent earnngs and educaton, an ncrease n the probablty of remanng permanently of one standard devaton s assocated wth an ncrease n the probablty of hgh school attendance of about 4 percentage ponts. Gven that only about 17 percent of sons of foregn born parents attend hgh school, ths mpact s qute large. In columns 2 we condton n addton on the father's permanent wages and on the father's years of educaton. Ths specfcaton corresponds to the nvestment equaton n (4). Condtonng on the father's schoolng should further elmnate nfluences that affect both the son's educatonal attanment, and re-mgraton probabltes. The sze of the coeffcent of the permanent mgraton probablty reduces only slghtly f we condton on these varables. Thus, condtonal on the father's level of educaton, age and permanent wages, there remans a szeable dfference n enrollment n hgher track schools between those whose fathers tend to reman permanently n the host country, and those whose fathers do not. The

21 Dustmann 19 coeffcent estmate on the father's permanent wage s postve and sgnfcant. An ncrease n father's permanent log wage by one standard devaton (0.21) ncreases the probablty that the son wll enroll n a hgher secondary school by about 4 percentage ponts (see table A3). In columns 3 and 4, we report results for estmaton equatons condtonng only on permanent log wages, and on fathers' educaton, for both foregn (upper panel) and natve born (lower panel) father-son pars. The latter equaton s frequently estmated n the lterature on the ntergeneratonal moblty of educaton. Wthn our nvestment model, t can be nterpreted as a reduced form equaton, where the effect of fathers' educaton on sons' educatonal attanments works through fathers' permanent wages. For foregn born fathers, the father's years of educaton s nsgnfcant, but t s hghly sgnfcant (and much larger n sze) for natve born fathers. For the latter group, each year of addtonal educaton ncreases the probablty that the son attends hgh school by 7.3 percentage ponts. These results support estmates by Gang and Zmmermann (2000), who do not fnd any assocaton n educaton between foregn born parents and ther chldren. In ther case, ths may be due to the fact that ther sample ncludes mmgrants who arrved before the age of 16, and have therefore obtaned or started educaton n ther home countres. In our case, all sons of foregn born parents are born n Germany; despte that, we fnd only small and nsgnfcant assocatons between fathers' and sons' educatonal attanments. We have also estmated the same specfcatons for the completed number of years of educaton for those sons who completed full tme educaton on father's permanent mgraton probabltes (results not reported). Estmaton results are very smlar. As for the track choce, sons of fathers who have a hgher probablty of remanng permanently have sgnfcantly hgher completed levels of educaton. Condtonng on the father's permanent log wage and level of educaton n addton, reduced the coeffcent estmate only slghtly, and s largely n lne wth track choce results reported above. When we estmate nvestment equatons condtonng only on the father's permanent wage or the father's years of educaton n addton, the coeffcent on

22 Dustmann 20 the father's permanent earnngs s remarkably smlar for natves and mmgrants, suggestng that an ncrease n permanent wages of 10 percent ncreases years of educaton by about As before, whle for natve fathers, educaton s strongly and sgnfcantly assocated wth sons' educaton, t s smaller n sze and nsgnfcant for mmgrants. What explans the small coeffcent we estmate for the ntergeneratonal correlaton n educaton for mmgrants? One reason mght be measurement error. As educaton of the foregn born s obtaned abroad, t s more lkely to be ms-coded than educaton obtaned n the host country. Ths may lead to a downward bas n estmates; however, t s unlkely to fully explan the large dfference n pont estmates for foregn and natve born father-son pars. xv A further explanaton s that t s permanent earnngs rather than educatonal achevements of the father that drves educatonal outcomes of the son. Ths nterpretaton s compatble wth an ntergeneratonal permanent ncome model, as the one we develop above. The assocaton between fathers' permanent earnngs and sons' educaton that we estmate, s smlar for school track choce of foregn and natve born (table 6), and near dentcal for completed full tme educaton. If educaton of the father affects the son's educaton manly through the father's earnngs, a low correlaton between permanent earnngs and educaton - as often found n mmgrant communtes - explans why some studes (lke Gang and Zmmermann 2000) fnd only a modest assocaton of educatonal achevements between parent and offsprng n mmgrant samples. xv B.2 Intergeneratonal earnngs correlaton Comparng Foregn and Natve Born Father-Son Pars We now turn to analyss of ntergeneratonal moblty n permanent earnngs, and on the

23 Dustmann 21 mpact of a permanent mgraton probablty on permanent wages. In table 7 we frst report ntergeneratonal correlaton coeffcents for both mmgrants and natves, usng the measure for fathers' permanent log wage as descrbed above. We gnore here the permanent mgraton varable for mmgrants, but we use measures for fathers' permanent ncome that gradually remove measurement error. We use all wage nformaton for sons, and report standard errors that allow for an equ-correlated covarance matrx. Estmates are based on equaton (9), but nclude n addton a quadratc n son's age and a lnear tme trend. The upper panel of table 7 reports results for foregn born fathers, and the lower panel reports results for natve born fathers. The results n the frst column are based on all avalable observatons for constructng fathers' log wages. Intergeneratonal correlaton coeffcents for natves as well as mmgrants are small, and smlar n magntude to those reported by Crouch and Dunn (1997). In the second column, we use as a measure for parental permanent earnngs predctons based on at least fve wages spells. Ths should reduce any downward bas through measurement error. The effect on estmates s qute dramatc, wth coeffcents ncreasng to 0.25 for natves, and 0.36 for mmgrants. The last column reports estmates where only father-son pars wth fathers reportng at least 8 wage spells are ncluded. Coeffcents rse further, to 0.29 for natves, and 0.40 for mmgrants. Restrctng the samples even further (we expermented wth at least 11 wage observatons for fathers) does not lead to any sgnfcant further change n the coeffcents. The ncrease n estmated coeffcents s n lne wth other studes, suggestng substantal underestmaton n the degree of ntergeneratonal mmoblty through measurement error n parental earnngs. The numbers further suggest a larger ntergeneratonal mmoblty for mmgrant father-son pars than for natve father-son pars. The dfference n pont estmates, based on estmates n column 3, s about 0.12; however, the dfference s not sgnfcantly dfferent from zero, whch s perhaps not surprsng gven the small sample sze. The

24 Dustmann 22 magntude of estmates for mmgrants s smlar to those reported for the US by Solon (1992) and Zmmermann (1992), who obtan coeffcents of about 0.4, also usng mult-year averages to reduce measurement error n parental earnngs. Our pont estmates for natves are smaller than those n the U.S. studes. Bjoerklund and Jantt (1997), usng a method for estmatng ntergeneratonal ncome correlaton on ndependent samples for Sweden, also conclude that Swedsh estmates are smaller than those obtaned for the U.S. Permanent Mgraton Probabltes In table 8 we report estmates for mmgrants where we condton addtonally on the father's probablty of a permanent mgraton, as suggested by equaton (9). Other regressors nclude, as before, a quadratc n the son's age, the father's orgn dummes, and a tme trend. The estmated coeffcent of the measure for a permanent mgraton suggests a hgher log wage for sons whose fathers tend to consder the mgraton as permanent rather than temporary. The effect s szeable: an ncrease n the probablty of a permanent mgraton by one standard devaton (0.26) ncreases the son's permanent real wages by about 5 percent. In the next columns we condton n addton on the father's permanent wages. Columns 2 consder all father-son pars wth at least one observaton on fathers' earnngs, and columns 3 and 4 restrct the sample to father-son pars wth more than 4 or 7 earnngs observatons for the father. In columns 2, the coeffcent on the father's probablty of a permanent mgraton remans roughly the same as n column 1, wth about the same standard error. Condtonal on fathers' permanent earnngs, an ncrease n the probablty of the father stayng permanently contnues to ncrease the son's permanent wage. When usng measures for fathers' earnngs that are based on more than four or seven wage observatons (columns 3 and 4), the coeffcent on the measure for a permanent mgraton propensty decreases n sze, and s less precsely estmated. One reason for ths s the reducton

25 Dustmann 23 n sample sze when we move to fathers wth more than four or seven vald wage spells. The decrease n the coeffcent on the permanent mgraton probablty could be due to the father's permanent earnngs beng postvely correlated wth the father's probablty to mgrate permanently. Therefore, a downward bas n the effect of fathers' permanent earnngs on the son's wage translates nto an upward bas n the coeffcent estmate of the permanent mgraton probablty. Overall, return probabltes seem to be related to sons' permanent earnngs n a way that s compatble wth the model, although the last estmates are not sgnfcant. Fnally, note that n ths specfcaton the sze of the coeffcent on the father's permanent wage measure s slghtly decreased, as compared to results reported n table 7, but stll larger n magntude to results for natve born fathers. Robustness checks Smultanety: One concern wth our estmates s that fathers condton ther own mgraton plans on the son's labour market performance, whch may lead to smultanety bas f we use nformaton on the father's re-mgraton plans after the son has entered the labour market. To check ths, we have re-estmated regressons n table 8 usng permanent mgraton probabltes that are constructed from the father's responses before the son was 16 years old (whch s the mnmum age for labour market entry). We present results n the lower panel of table 8. The estmates for the permanent mgraton probablty ncrease slghtly n sze, and reman stable and sgnfcant throughout the four specfcatons. Estmates for the three measures of the father's permanent earnngs are more smlar n sze across specfcatons than before. The reason s that the sample whch s based on permanent mgraton ntentons before sons were 16 years old, excludes many of those fathers who have only a small number of wage observatons, thus reducng the measurement error bas from the start. Results confrm that there s a postve relatonshp between the father's permanent

26 Dustmann 24 mgraton propensty and the chld's earnngs along the lnes we have hypothessed above. Parents may have knowledge about ther chld's future earnngs performance before ther chld's entry nto the labour market, and condton ther return ntentons on ths, whch may then stll lead to based estmates. When usng father-son pars where mgraton probabltes are computed when the chld was even younger leads to small samples. We have expermented wth that, and used father-son pars for whch we can compute return probabltes for fathers based on survey nformaton when the son was 13 years of age or younger. The sample sze drops to 164 observatons when estmatng specfcatons as n column 3 of table 8. The coeffcent on the permanent mgraton probablty s smlar to that reported n table 8: 0.266, wth a standard error of Sblngs: We nclude n our sample sblngs who have the same father (see table 1 for frequences). To check f our results are robust to usng only the oldest son for father-son pars wth more than one son, we re-estmate the models restrctng our sample to father-oldest son pars f a father has more than one son. The results for the ntergeneratonal correlaton coeffcents are smlar, wth pont estmates and assocated standard errors of (0.058) and (0.061) for natves, usng wage nformaton based on more than fve or seven years of the father's wage data, and correspondng coeffcents of (0.165) and (0.197) for mmgrants. Pont estmates reman hgher for mmgrant father-son pars. When condtonng on the permanent mgraton probablty n addton, coeffcents on the father's permanent earnngs decrease slghtly, as before. The estmated coeffcent on the father's permanent mgraton probablty s smlar: pont estmates for regressons n columns 1-4 of table 8 are 0.20, 0.16, 0.15, and 0.17.

27 V Dscusson and Conclusons Dustmann 25 Earler research by Borjas (1992) suggests that the way wealth and earnngs potental s transmtted from one generaton to the next may dffer between mmgrant and natve born communtes. Borjas argues that for mmgrants, the qualty of the ethnc envronment may provde an externalty n the producton of human captal of the next generaton, whch affects parental nvestments. In ths paper we nvestgate a further reason why parental nvestment may be dfferent across mmgrant groups: dfferences n parents' assessment about a permanent or a temporary mgraton. We also estmate and compare ntergeneratonal correlaton coeffcents for educaton and permanent earnngs, dstngushng between mmgrant and natve born fatherson pars. Our emprcal analyss s based on a long panel that oversamples mmgrants. The data provdes unque and repeated nformaton on permanent mgraton ntentons of foregn born ndvduals. To analyse parental nvestment n the son's educaton, we nvestgate secondary school track choce. We fnd a strong assocaton between the probablty of fathers' permanent mgraton, and sons' educatonal attanments, condtonal on the father's age and orgn country, and the son's cohort. These estmates reman smlar f we condton n addton on fathers' educaton, and on fathers' permanent earnngs. For both natve and foregn born fathers, the father's permanent earnngs postvely affect educatonal achevements of the son. For father-son pars wth natve born fathers, we also fnd a strong correlaton between educatonal attanment of father and son, whle ths correlaton s small and nsgnfcant for mmgrants. We explan ths wth the low assocaton between educaton and earnngs for foregn born fathers: f parental earnngs affect nvestment n the chld's educaton, as suggested by the permanent ncome model of ntergeneratonal moblty, then the father's educaton s a weaker proxy for permanent earnngs for mmgrants

28 than for natves. Dustmann 26 Estmatng ntergeneratonal correlaton coeffcents n permanent earnngs for father-son pars of foregn and natve born fathers, and takng account of measurement error usng a flexble averagng method, we fnd ntergeneratonal correlaton coeffcents of about 0.3 for father-son pars wth natve born fathers, and about 0.4 for father-son pars wth foregn born fathers. Consstent wth our hypothess, we fnd for foregn born fathers that the son's permanent wages are postvely assocated wth the probablty of the father's permanent mgraton. Our results therefore suggest that the effects of permanent as opposed to temporary mgraton expectatons on mmgrants' performance, as establshed for nstance by Dustmann (1997) and Cortes (2004), are not restrcted to the mmgrant generaton, but are lkely to carry over to the next generaton.

29 References Dustmann 27 Bauer, Phlpp and Regna T. Rphan Heterogenety n the Intergeneratonal Transmsson of Educatonal Attanment: Evdence from Swtzerland on Natves and Second Generaton Immgrants. Mmeographed. Basel: Unversty of Basel. Becker, Gary. and Ngel Tomes An Equlbrum Theory of the Dstrbuton of Income and Intergeneratonal Moblty. Journal of Poltcal Economy 87 (6): Becker, Gary and Ngel Tomes Human Captal and the Rse and Fall of Famles. Journal of Labor Economcs 4(3): S1-S39. Bjoerklund, Anders and Markus Jantt Intergeneratonal Income Moblty n Sweden Compared to the Unted States. Amercan Economc Revew 87(5): Bjoerklund, Anders, Mkael Lndahl, and Erk Plug The Orgns of Intergeneratonal Assocatons: Lessons from Swedsh Adopton Data. The Quarterly Journal of Economcs 121(3): Bohnng, W. Roger Studes n Internatonal Mgraton. New York: St. Martn's Press. Borjas, George J Assmlaton, Changes n Cohort Qualty, and the Earnngs of Immgrants. Journal of Labour Economcs 3(4): Borjas, George J Ethnc Captal and Intergeneratonal Moblty. The Quarterly Journal of Economcs 107(1): Borjas, George J The Intergeneratonal Moblty of Immgrants. Journal of Labor Economcs 11(1): Borjas, George J Long-Run Convergence of Ethnc Skll Dfferentals: The Chldren

30 Dustmann 28 and Grandchldren of the Great Mgraton. Industral and Labor Relatons Revew 47(4): Borjas, George J. 1994b. The Economcs of Mgraton. Journal of Economc Lterature 32(4): Borjas, George J Ethncty, Neghbourhood, and Human Captal Externaltes. Amercan Economc Revew 85(3): Card, Davd, John DNardo and Eugena Estes The More Thngs Change: Immgrants and the Chldren of Immgrants n the 1940s, the 1970s, and the 1990s. NBER Workng Paper Carlner, Geoffrey Wages, Earnngs and Hours of Frst, Second and Thrd Generaton Amercan Males. Economc Inqury 18(1): Chswck, Barry Sons of Immgrants: Are they at an Earnngs Dsadvantage? Amercan Economc Revew 67(1): Chswck, Barry The Effect of Amercanzaton on the Earnngs of Foregn Born Men. Journal of Poltcal Economy 86(5): Couch, Kenneth A. and Thomas A. Dunn Intergeneratonal Correlaton n Labour Market Status: a Comparson of the Unted States and Germany. Journal of Human Resources 32(1): Cortes, Kalena E Are Refugees Dfferent from Economc Immgrants? Some Emprcal Evdence on the Heterogenety of Immgrant Groups n the Unted States. The Revew of Economc and Statstcs 86(2): Dearden, Lorrane, Steve Machn and Howard Reed Intergeneratonal Moblty n