NEW APPROACHES AND ESTIMATES OF FOREIGN BORN EMIGRATION. Jennifer Van Hook. Bowling Green State University. Jeffrey Passel. The Urban Institute

Transcription

1 EW APPROACHES AD ESTMATES O OREG BOR EMGRATO Jennfer Van Hook Bowlng Green State Unversty Jeffrey Passel The Urban nsttute Wewe Zhang Bowlng Green State Unversty rank D. Bean Unversty of Calforna rvne

2 EW APPROACHES AD ESTMATES O OREG BOR EMGRATO Abstract Post-censal populaton estmates depend on the accuracy of estmates of demographc components of change. Of these components, emgraton of the foregn born s probably the least well known. ndrect methods.e., the resdual method are partcularly poor at estmatng emgraton among recent arrvals. We ntroduce a new method for estmatng emgraton that takes advantage of the unque sample desgn of the CPS. The CPS follows housng unts but not necessarly ndvduals over a perod of 6 months. ndvduals n the March CPS not successfully followed up nclude those who ded, nternal mgrants, and emgrants. We use statstcal methods to estmate the proporton of emgrants among those not followed up. Our method produces emgraton estmates that are comparable to resdual-based methods n the case of earler arrvals mmgrants who arrved more than ten years ago, but yelds much hgher estmates for earler arrvals.

3 EW APPROACHES AD ESTMATES O OREG BOR EMGRATO Post-censal populaton estmates depend on the accuracy of estmates of demographc components of change: brths, deaths, mmgraton, and emgraton. Of these components, emgraton of the foregn born s probably the least well known. Yet populaton estmates bult up through the cohort component method wll be too low f emgraton s over-estmated and too hgh f emgraton s under-estmated. n the case of resdual estmates of unauthorzed mgrants, the accuracy of emgraton rates among the legal foregn born s crtcal. f the emgraton estmates of the legal foregn-born are too hgh, the legal populaton wll be too low and the unauthorzed mgrant populaton wll be overestmated. Offcal statstcs on emgraton from the Unted States are vrtually non-exstent. The S collected data from departng foregn born emgrants from 908 to 957 Woodrow-Lafeld 998, but dscontnued ths practce due to concerns about the qualty of emgraton data Kraly 998. n contrast, the number of brths, deaths, and arrvals of legal mmgrants are known wth relatve accuracy because the Unted States vtal regstraton system and the former mmgraton and aturalzaton Servce now U.S. Ctzenshp and mmgraton Servces collects data on these events. Out of necessty, emgraton has been estmated wth a varety of ndrect demographc methods, most promnent of whch s the resdual method. We develop a new method for estmatng emgraton that takes advantage of the longtudnal nature of the Current Populaton Survey. By producng new foregn-born emgraton estmates wth whch resdual estmates can be compared, we offer a new way to evaluate and update the emgraton estmates currently used by the Census Bureau and others for the producton of populaton estmates.

4 Prevous Research Warren and Peck 980 frst developed ndrect methods for estmatng emgraton. Ther technque referred to here as the resdual method has snce served as the major method used by the U.S. Census Bureau for developng and updatng foregn-born emgraton estmates. Warren and Peck s 980 orgnally estmated an annual number of emgrants of 4,000 for the decade. Ths fgure was later ncreased to 33,000 on the bass of Warren and Passel s 987 analyss of S Alen Address Regstraton data for The 33,000 fgure was used as an offcal pont estmate by the U.S. Census Bureau untl the md-990s, when the number was ncreased to 95,000 based on the resdual estmates of Ahmed and Robnson 994. The number may be updated agan on the bass of Mulder s 2003 more recent work, whch estmated an annual number of 225,000 foregn-born emgrants durng the 990s. The resdual method for estmatng emgraton n the decade between two censuses nvolves the comparson of two populaton estmates: the expected populaton f no emgraton had occurred durng the decade, and the enumerated populaton at the end of the decade. Resdual estmates of emgraton durng the 990s Mulder 2003, for example, were constructed by survvng mmgrants who arrved pror to 990 forward to 2000.e., by agng all cohorts by ten years and subtractng the estmated numbers of deaths and then comparng the survved populaton wth the enumerated populaton n The dfference s attrbuted to emgraton. The number of emgrants among mmgrants who arrved between 990 and 2000 s estmated by applyng emgraton rates from the earler arrvals. Pror emgraton estmates based on the resdual method as well as other methods have been revewed n detal elsewhere Kraly 998; Woodrow-Lafeld 998; Mulder To synthesze ths lterature, we compare the results of varous studes n a table showng estmates 2

5 of the number of emgrants together wth the emgraton rates and the rate at whch mmgraton s offset by emgraton as would be mpled by the varous estmates Table. Even though the resdual estmates of the annual number of emgrants has ncreased over tme from 4, to 95, and then to 225 thousand, the assocated rates of emgraton relatve to the md-year foregn born populaton and rato of emgrants to newly-admtted legal mmgrants appears to have declned from.8% durng the 960s,.5% n the 980s, and 0.88% n the 990s. Among recently-arrved mmgrants, the emgraton rates are nconsstent across studes. Mulder s 2003 estmate of 2,000 emgrants per year among 990s arrvals mples a much lower emgraton rate than pror studes. Emgraton rates for Mexcan mmgrants are even less consstent, wth the resdual estmates showng much lower levels of emgraton than the estmates made by Massey and hs colleagues. Massey and hs colleagues estmate emgraton through the analyss of detaled accounts of trps by Mexcan mgrants to the Unted States. The detal n the data permt the dentfcaton of each separate trp as contrbutng n- and outmgraton, and thus capture many more emgrants and return mgrants over tme. A major drawback of the resdual method s that the estmates are senstve to dfferences n census coverage. or example, coverage error was hgher n the 990 Census than the 2000 Census Robnson et al., 993; Hogan 993; U.S. Census Bureau 200, so n many cases the expected populatons n 2000 turned out to be smaller than the enumerated populatons n 2000 and thus mpled a negatve emgraton rate an mpossblty. Ths problem s especally evdent for country-of-orgn groups that contan large proportons of unauthorzed mgrants, such as Mexco. or Mexcans, the expected populaton s sgnfcantly lower than the enumerated populaton for both the and decade Ahmed and Robnson 994; Mulder Ahmed and Robnson s 994 analyss suggests that census coverage of 3

6 mgrants from Mexco and Central Amerca may have been much lower n the 980 Census than n the 990 Census by whch tme many unauthorzed mgrants had legalzed under RCA. Ahmed and Robnson 994 handle the problem of dfferental undercount and negatve emgraton estmates by estmatng emgraton rates by race/ethncty not country-of-orgn whle excludng those country groups wth negatve rates. They use the race-specfc rates as proxes for rates of countres, for example, matchng Hspanc rates to countres sendng hgh proportons of Hspanc mmgrants. Mulder 2003 handles the problem by adjustng the census estmates for undercount. Ths results n a set of fnal emgraton estmates that are hghly senstve to the coverage estmates on whch the adjustments are based. A second drawback of the resdual method s that t depends on the accurate comparson of age- and year-of-entry cohorts across two censuses. The consstency of reportng on year of entry s of partcular concern. n the 990 Census, one-thrd of mmgrants who reported havng come to the U.S. between 985 and 990 s lkely to have been resdent n the Unted States pror to 985 Ells and Wrght 998. f a sgnfcant number of recently-arrved mmgrants understate the length of tme they have lved n the Unted States n one census, ths would lead to upwardly based estmates of recent arrvals n the frst census, and upwardly based estmates of the expected populaton ten years later. f reportng on year of entry were more accurate n the later census, t would appear that more recent arrvals emgrated than was n fact the case. A thrd weakness of the resdual method les n ts nablty to estmate emgraton for recently arrved mmgrants. Ths s evdent by the nconsstent and, n some cases, unrealstcally low emgraton resdual-based estmates for recently arrved mmgrants Table. The resdual method compares cohorts between two censuses and therefore produces emgraton estmates for those who were present n the country at the tme of the frst census. But the 4

7 emgraton rates for mmgrants who arrved durng the ntercensal perod are not derved from the data, but nstead, are assgned the rates that were calculated for earler arrvals. nally, the resdual method s that t does not permt the estmaton of emgraton by mgraton status because mgraton status s not dentfed n the Census. Massey and Snger 985 make a sgnfcant contrbuton by usng Mexcan Mgraton Project MMP data to estmate n-, out-, and net unauthorzed mgraton flows from Mexco to the Unted States. Ther work suggests that the unauthorzed flow from Mexco conssts of large proportons of temporary and crcular mgrants whose emgraton rates are lkely to be much hgher than legal permanent resdents. However, they do not prepare comparable statstcs for legal mmgrants or unauthorzed mgrants from countres other than Mexco. t remans mportant to compare the emgraton patterns of unauthorzed and legal mmgrants. One reason s that the resdual method for estmatng the number of unauthorzed mgrants depends on the emgraton patterns of legal mgrants, not all foregn born. We supplement resdual estmates for the 990s wth estmates based on a new approach we refer to as the CPS Matchng Method. Ths method takes advantage of the unque sample desgn of the CPS, whch follows housng unts but not necessarly ndvduals over a perod of one year and 4 months. ndvduals n the March CPS not successfully followed up n the March CPS n the followng year nclude those who ded, nternal mgrants, and emgrants. We use statstcal methods to estmate the proporton of emgrants among those not followed up. An advantage of the CPS Matchng method s that t does not depend on assumptons about dfferental census coverage or consstent reportng on year-of-entry snce n follows ndvduals, not groups, over tme. Another advantage s that t treats recently-arrved The Census Bureau and others produce estmates of the number and characterstcs of unauthorzed mgrants enumerated n the Census. However, these estmates rest n part on assumptons about foregn-born emgraton. Ther subsequent usage n the producton of emgraton estmates nvolves a certan amount of crcularty n logc. 5

8 mmgrants n the same manner as earler arrvals. t therefore s more lkely than the resdual method to produce comparable estmates across dfferent perod-of-entry groups. nally, the CPS Matchng Method permts us to estmate emgraton by legal status. We use the mgraton status mputatons developed earler by Clark and Passel. The mputaton procedure assgns foregn-born non-ctzens legal status on a probablstc bass gven place of brth, occupaton, sex, gender, welfare recpency, state of resdence, and other characterstcs. Once we assgn ndvduals as legal and unauthorzed, we analyze one-year follow-up rates separately by legal status n order to produce emgraton estmates by legal status. As wll become clear below, the CPS matchng method depends on the accuracy of certan assumptons, most sgnfcantly that emgraton among the thrd-or-hgher generaton s very low, and that mmgrants and the thrd-or-hgher generaton have smlar patterns of nonfollow-up due to resdual causes whle controllng for a number of socoeconomc factors. Although our assumptons could be subject to the same knd of scrutny as Mulder s assumptons about dfferental undercount, both sets of emgraton estmates would gan credblty f they were consstent wth each other. Our emgraton estmates are lkely to be larger than those based on the resdual method because the resdual method tends to mss emgrants who return to the Unted States. Massey and hs colleagues estmate that the average duraton of a Mexcan labor mgrant s frst trp to the Unted States s only 2 months and that one-thrd of these mgrants return to the Unted States n a second trp wthn ten years of the frst trp Massey, Durand, and Malone Eght-fve percent of entres nto the U.S. by unauthorzed Mexcan mgrants between 965 and 989 were offset by exts Massey and Snger 995. Ths type of crcular mgraton s not typcally captured by the resdual method, whch counts the number of net rather than total emgrants. 6

9 The number of net emgrants from 990 to 2000 s equal to the sum of the number of emgratons each year E y mnus the number of return trps to the U.S. n each year among those who emgrated between 990 and 2000 R y : E y 2000 y 990 y 2000 E y R y 990 y Takng the annual average, E E y R 0 y E 0 E y R y, and E y + R y. 0. So, E Ths shows that the dfference between estmates usng a ten-year nterval and a one-year nterval as does the CPS matchng method ntroduced here s equvalent to the average annual number of return trps. Data To estmate foregn-born emgraton rates for the late 990s and early 2000s, we use data drawn from the 998, 999, 2000, 200, 2002, 2003, and 2004 March Current Populaton Survey CPS fles. The March CPS samples offer several advantages n that they follow housng unts over tme and contan nformaton about nternal mgraton and nternatonal mmgraton. n addton, we have already developed methodology for mputng mgraton legal status for the foregn-born n the CPS. Thus the groundwork has already been lad that makes t possble to produce foregn-born emgraton estmates separately for legal and unauthorzed mgrants. An underused feature of the Current Populaton Survey s that t follows housng unts over a perod of 6 months or more. Most CPS respondents are elgble to be ntervewed eght tmes. The number of months a samplng unt has been n the sample s dentfed by the varable Month-n-sample, whch ranges from to 8. When most respondents frst enter the CPS sample, they are ntervewed four months n a row months-n-sample -4. Eght months later, 7

10 they are ntervewed agan for four months months-n-sample 5-8. or example, a respondent who s ntervewed n each month from January through Aprl n one year s elgble to be rentervewed n the same months n the followng year. Ths means that all respondents n the March sample wth a month-n-sample code of -4 are elgble to be followed up n March of the followng year. The March supplement ncludes an oversample of Hspancs that follows a slghtly dfferent ntervew schedule than the regular sample. All Hspancs n the October CPS sample are elgble to be ntervewed n March n addton to the eght months they would normally be ntervewed. or example, Hspancs who are ntervewed for the frst tme n October are elgble to be ntervewed not only n ovember, December, and January, but also n March. All respondents who appear for the frst tme n the March Demographc supplement, ncludng those n the oversample, are elgble to be followed up n the March sample n the followng year. A key feature of the CPS sample desgn one that s crtcal for our purposes s that t follows housng unts rather than ndvduals. f a respondent moves to a new address, the new occupants of the orgnal housng unt are ntervewed and the orgnal respondent s dropped from the sample. Ths feature of the CPS sample desgn permts us to use follow-up rates.e., the proporton of persons n the March sample who are successfully followed up n the March sample n the followng year as a bass for estmatng emgraton. Components of non-follow-up. Madran and Lefgren 999 estmate that 29 percent of those elgble for follow-up n the March CPS surveys were not successfully followed up. There are many reasons a person may not be followed up. Based on known rates of nternal mgraton and mortalty 8

11 derved from the CPS and CHS statstcs, Madran and Lefgren 999 estmate that 6.3 percent moved to another address n the Unted States and 0.86 percent ded, leavng.8 percent who were not followed up for other reasons. Of the resdual.8 percent, some may have moved to another country, whle others were not followed up due to non-response, refusals, codng error, or the nablty to contact a person n the housng unt. Thus, the proporton of elgble persons not followed up u, conssts of the proporton who mgrated wthn the Unted States m, the proporton who ded d, the proporton who emgrated e, and the proporton who were not followed up for other reasons r. or mmgrants, and for natves: u u m d r e a m Subtractng b from a and solvng for e yelds: e u u d r e b m m d d r c We are able to estmate non-follow-up rates u and u and nternal mgraton rates m and m drectly from the CPS data. The mortalty component s lkely to be small except n the older age groups Madran and Lefgren 999. We are able to estmate mortalty rates by natvty from the atonal Health ntervew Survey Pallon and Ares n addton, we make the smplfyng assumpton that emgraton among natves s neglgble.e., that the value of e s close to zero. ernandez 995 estmates that durng the 980s, roughly 48,000 U.S. born emgrated per year. Ths amounts to an annual rate of about.02 percent. However, estmatng resdual non-follow-up rates r and r s more challengng. Our strategy s to assume that r e 9

12 mmgrants and natves have dentcal non-follow-up rates after controllng statstcally for factors assocated wth attrton. We descrbe below how we use predcted values generated from multvarate logstc regresson models to approxmate e. Estmaton Strategy The probablty of non-follow-up for each person can be modeled wth logstc regresson as a functon of a vector of socoeconomc and demographc characterstcs: exp X 'β + exp X 'β. u X 'β, mmgrants u n exp Xn'β + exp Xn'β. Xn'β, natves and each component of non-follow-up can be smlarly modeled: m d r e X 'µ X 'δ X 'ρ X 'λ mmgrants m d r e n n n n Xn'µ Xn'δ Xn'ρ Xn'λ natves The lkelhood of non-follow-up can therefore be expressed as the sum of predcted values generated from each component model. or example, for mmgrants: u + X'µ + X 'δ + X 'ρ X 'λ 3 Averagng 3a across all mmgrants yelds the non-follow-up rate among mmgrants: 0

13 u X 'λ X 'ρ X 'δ X 'µ X ' β 4a where s the number of mmgrants. Our strategy s to estmate generatonal dfferences n each component of non-follow-up nternal mgraton, mortalty, and resdual non-follow-up whle controllng statstcally for compostonal dfferences. We accomplsh ths by comparng predcted values for mmgrants equaton 4a wth predcted values for natves assumng both groups had dentcal characterstcs. The non-follow-up rate of natves f they had the same characterstcs as mmgrants u * s obtaned by replacng the coeffcents n equaton 4a wth natve coeffcents: * u X 'λ X 'ρ X 'δ X 'µ X ' β 4b By subtractng equaton 3b from 3a and solvng for 'λ X, we estmate the foregnborn emgraton rate as: e 'λ X + + [ δ δ µ µ β β ' X X ' X ' X ' X ' ' X ] λ ρ ρ ' X X ' ' X + + 5a

14 At ths pont, we ntroduce a number of smplfyng assumptons. rst, we assume that f gven the same socoeconomc, health, and demographc characterstcs mmgrants would exhbt the same resdual non-response rates as natves. That s [ ' ρ 'ρ ] 0 X X. Also, we assume that emgraton among natves s neglgble for the reasons noted above. n other words, [ 'λ ] 0 X. Therefore, [ ' β X ' β + X ' δ X ' δ + X ' µ ' ]. e µ X X 5b where the frst two terms gves the predcted natvty dfference n non-follow-up, the second two terms equals the predcted dfference n nternal mgraton, and the last two gve the predcted dfference n mortalty. Equaton 4b s partcularly useful because all components can be estmated wth CPS and HS data. To obtan values for the components of equaton 5b, we frst estmate three sets of logstc regresson models, the frst predctng non-follow-up among those elgble to be followed up, the second predctng nternal mgraton.e., lvng at a dfferent address from the year before among those who were lvng n the Unted States the year before, and the thrd predctng the one-year probablty of dyng. The models of non-follow-up and nternal mgraton are estmated separately by sex and race/ethncty for mmgrants and the second generaton and nclude as ndependent varables dentcal sets of soco-demographc varables: age, educaton, housng tenure rent vs. own, type of resdence moble home, barracks, dorm, etc., school enrollment status, and general health status. The second generaton rather than all natves s used as the comparson group. The reason s to reduce the predcted natvty dfference n resdual non-follow-up. Because of 2

15 cultural and socal smlartes, mmgrants resdual non-follow-up rates are more lkely to be smlar to the second generaton than the thrd-or-hgher generaton s rates. Ths s especally clear n the case of blacks. Thrd-or-hgher generaton blacks nclude very few n the thrd generaton. Most are descendants of slaves whose famles have been n the Unted States for centures and dffer from black mmgrants and ther chldren n many mportant ways, ncludng perhaps, n ther rates of resdual non-follow-up. One problem wth usng the second generaton as the comparson group s that both mmgrant and second generaton chldren are chldren of mmgrants, often share the same households, and are therefore lkely to have smlar emgraton rates. Ths means that, n the case of chldren, the estmated emgraton rates of the second generaton are unlkely to be anywhere close to zero. f we were to use the methodology outlned above for chldren, we would almost certanly underestmate mmgrant chldren s emgraton rates. or ths reason, we treat chldren ages 0 to 4 dfferently by assgnng them the predcted values of ther parent on the assumpton that most chldren emgrate wth ther famles. or adults, we generate from each model predcted values of the lkelhood of nonfollow-up and nternal mgraton. our predcted values are calculated for each mmgrant: non-follow-up usng mmgrant coeffcents, 2 non-follow-up usng second generaton coeffcents, 3 nternal mgraton usng mmgrant coeffcents, and 4 nternal mgraton usng second generaton coeffcents. Predcted values of non-follow-up and nternal mgraton for each sex and race/ethnc group are derved from each groups correspondng models. or example, the predcted values for Mexcan males come from the Mexcan male models. n the case of Asans, there were not enough cases n the second or thrd generaton to obtan stable estmates, so we obtaned ther predcted values from models estmated on all race-ethnc groups combned. 3

16 Predcted one-year probabltes of death for adults ages 5 and older are obtaned from the atonal Health ntervew Survey HS. The HS s lnked to the atonal Death ndex n order to ascertan whether and the age at whch respondents de. Usng a personyear fle, we estmate logstc regresson models of whether a person ded durng the year separately for mmgrants and natves. We do not estmate models for the second generaton apart from all natves due to data lmtatons of the HS. The ndependent varables nclude sex, age, race, ethncty, and general health. We use the coeffcents from the mortalty models to generate both mmgrant and natve predcted probabltes for mmgrants n the CPS data, as shown n the last two terms n equaton 5b. nally, we subtract the mmgrant predcted values from the second generaton predcted values natve predcted values n the case of mortalty as shown n the brackets n equaton 5b. We then average the results to obtan an estmate of e for all mmgrants and for mmgrant subgroups by age, sex, country-of-orgn, year-of-entry, and mgraton status.e., we average the bracketed porton of 5b for the dfferent groups. As noted above, chldren are assgned the predcted values of ther parent dentfed by the PARET dentfer n the CPS. Matchng CPS Surveys. To determne whether a respondent n the March CPS n one year t s successfully followed up the followng year t+, we match those elgble for followup 2 n the March CPSs wth respondents n the followng years CPS We use the methodology and STATA code developed by Madran and Lefstrom 999 for matchng cases across CPS fles, matchng on household dentfcaton number and person lne number. Because matched cases may not represent the same ndvdual due to codng errors on the person or household dentfcaton varables, we also requre consstency n sex or age 2 or most cases, those wth month-n-sample codes -4 are elgble for follow-up. or those n the Hspanc oversample n most years, month-n-sample s erroneously reverse-coded personal communcaton wth Greg Weyland. or these cases, we select months-n-sample 5-8 as elgble for follow-up. 4

17 person at year t can be no more than 2 years younger than the matched case n year t before consderng a case a true match. We do not take nto account consstency n race because of changes between 2002 and 2003 n the race queston now multple racal denttes are permtted. nternal mgraton. The CPS asks respondents whether he/she lved n a dfferent resdence one year before. We defne the nternal mgraton rate as the proporton of movers among those who reported havng lved n the Unted States one year before. Snce the nternal mgraton queston s retrospectve, we estmate nternal mgraton rates for year t from respondents n the CPS n year t or example, the proporton of nternal mgrants n 998 s estmated from the 999 March CPS. Ths s dfferent from how we calculate follow-up rates, whch are the proporton of CPS respondents n year t who are followed up n year t+. Return Mgraton. oregn born who lved abroad one year before but who reported havng come to stay n the U.S. more than 2 years before are defned here as return mgrants. Mgraton Status. [comng soon]. Results As shown n Table 2, we estmate an annual foregn born emgraton rate of 2.39 percent. or a populaton of 30 mllon foregn born as n the 2000 March CPS, ths translates nto roughly 76 thousand emgrants per year. At the same tme, we estmate a return mgraton rate of 0.87 percent, or 260 thousand. Subtractng return mgraton from total emgraton yelds an annual net emgraton rate of.52 percent, or 456 thousand per year. Males are nearly twce as lkely to emgrate as females 3.3 versus.65 percent and are more lkely to return to the Unted States but not enough to offset ther hgher emgraton rates. 5

18 Emgraton and return mgraton rates tend to be relatvely hgh for younger mmgrants, declne wth age, and n the case of emgraton but not return mgraton, ncrease n the oldest age group 65+. Takng emgraton and return mgraton together, net emgraton appears hghest among chldren and workng aged adults. et emgraton dps for teenagers and young adults age 5-24 and mddle-aged adults 45-64, but ncreases somewhat for the elderly 65+ perhaps a result of retrees returnng to ther countres of orgn. Among selected country or regon of orgn groupngs, mmgrants from Mexco, nda, Afrca, orth Amerca, and Europe appear to have the hghest emgraton rates and mmgrants from Chna the lowest. Of the hgh-emgraton groups, the relatvely hgh return mgraton rates of Mexcans stand out, reflectng the crcular mgraton patterns commonly observed for ths group. On balance, net mgraton rates appear lowest for groups from Southeast Asa Chna, Phlppnes, other Asa and Central and South Amerca, and hghest for mmgrants from south Asa nda, Afrca, Europe, and orth Amerca Canada. Because of ther hgh return mgraton rates, Mexcans fall between these groups. Emgraton rates vary wth mgraton/legal status n expected ways. Refugees many of whom cannot return to ther countres of orgn, and mmgrants who have naturalzed who have made the greatest legal/poltcal commtments to lvng n the Unted States exhbt the lowest emgraton rates both total and net. Legal non-ctzens have the next hghest emgraton rates. Unauthorzed mgrants, many of whom engage n crcular mgraton for temporary work, exhbt relatvely hgh emgraton and return mgraton rates. nally, legal non-mmgrants prmarly those on temporary work and student vsas exhbt by far the hghest emgraton rates. n general, emgraton rates are hghest for recent arrvals and declne wth tme n the Unted States. Return mgraton rates, as expected, are relatvely low for both recent and earler 6

19 arrvals 0 to 4 years and 0 or more years, but hgher for those who have lved n the U.S. for a moderate length of tme 5 to 9 years. How do these estmates compare wth resdual-based estmates? We compare our net emgraton rates wth the resdual-based estmates prepared by Mulder 2003 Table 3. Mulder s estmates pertan to the decade. To make a far comparson wth our estmates, we derve net emgraton rates from Mulder s results and then apply these rates to the 2000 populaton to obtan estmates of the number of emgrants for Our estmates are very smlar to Mulder s n the case of earler arrvals n the U.S. ten or more years. However, we estmate many more emgrants among recent arrvals.8 versus 0.32 percent, or a dfference of 20 thousand emgrants. n lght of the methodologcal problems assocated wth resdual-based estmates of emgraton among recent arrvals, ths suggests that the CPS-matchng method, whle comparable wth resdual-based estmates for earler arrvals, offers an mproved method for estmatng emgraton among recent arrvals. We also fnd that the CPS-matchng method tends to produce hgher emgraton estmates for countres or regons sendng many unauthorzed, temporary, and recently arrved mmgrants, such as Mexco a source of unauthorzed and crcular mgrants, nda a source of many temporary mgrants on HB vsas, and Afrca a relatvely new source of mmgraton. Conclusons and urther Work Refnements. We plan to explore the senstvty of the foregn-born emgraton estmates to varous assumptons about the level of U.S.-born emgraton. The method we outlne above assumes no emgraton for the second generaton. f emgraton for the second generaton were not neglgble, ths would mean that the foregn-born emgraton estmates are too low. We wll 7

20 re-estmate a range of foregn-born emgraton estmates under varyng assumptons about second generaton emgraton rates. A plausble range of assumptons about U.S.-born emgraton wll be drawn from the research lterature e.g., ernandez 995. Also, we plan to produce more detaled emgraton estmates by regon/state of resdence and martal status, and we plan to produce estmates for the U.S. born mnor chldren of the foregn-born. 8

21 References Ahmed, Bashr and J. Gregory Robnson Estmates of Emgraton of the oregn-born Populaton: Populaton Dvson Techncal Workng Paper o. 9. U.S. Census Bureau: Washngton, DC. Ells, M. and R. Wrght When mmgrants are not mmgrants: Countng Arrvals of the oregn Born Usng the U.S. Census. nternatonal Mgraton Revew 32: ernandez, Edward W Estmates of the Annual Emgraton of U.S. Born Persons by usng oregn Censuses and Selected Admnstratve Data: Crca 980. Populaton Dvson Techncal Workng Paper o. 0. U.S. Census Bureau: Washngton, DC. Hogan The 990 Post-Enumeraton Survey: Operatons and Results, JASA: Journal of the Amercan Statstcal Assocaton 88423:,047-,060. Kraly, Ellen Percy Emgraton: mplcatons for U.S. mmgraton Polcy Research. Pp n Mgraton Between Mexco and the Unted States: Bnatonal Study, Vol. 2 Research Reports and Background Materals. Mexcan Mnstry of oregn Affars and the U.S. Commsson on mmgraton Reform: Mexco Cty and Washngton, DC. Madran, Brgtte C. and Lars John Lefgren A ote on Longtudnally Matchng Current Populaton Survey CPS Respondents. Techncal Workng Paper 227. atonal Bureau of Economc Research: Cambrdge, MA. Massey, Douglas S. and Audrey Snger ew Estmates of Undocumented Mexcan Mgraton and the Probablty of Apprehenson. Demography 32: Massey, Douglas S., Jorge Durand, and olan J. Malone Beyond Smoke and Mrrors: Mexcan Mgraton n an Era of Economc ntegraton. Russell Sage oundaton: ew York. Mulder, Tammany oregn-born Emgraton from the Unted States: 990 to Paper presented at the 2003 annual meetngs of the Populaton Assocaton of Amerca, Mnneapols, M. Pallon, Alberto and Elzabeth Aras Paradox Lost: Explanng the Hspanc Adult Mortalty Advantage. Demography 43: Passel, Jeffrey and Rebecca Clark mmgrants n ew York: Ther Legal Status, ncomes, and Taxes. The Urban nsttute: Washngton, DC. Robnson, J. Gregory, Bashr Ahmed, Prths Das Gupta, and Karen A. Woodrow Estmaton of Populaton Coverage n the 990 Unted States Census Based on Demographc Analyss, JASA: Journal of the Amercan Statstcal Assocaton 88423:,06-,079. 9

22 U.S. Census Bureau ESCAP October 200 Report on Adjustment decson. Warren, Robert and Jeffrey Passel A Count of the Uncountable. Demography 24: Warren, Robert and Jennfer M. Peck oregn-born Emgraton from the Unted States: 960 to 970. Demography 7: Woodrow-Lafeld, Karen Vewng Emgraton at Century s End. Pp n Mgraton Between Mexco and the Unted States: Bnatonal Study, Vol. 2 Research Reports and Background Materals. Mexcan Mnstry of oregn Affars and the U.S. Commsson on mmgraton Reform: Mexco Cty and Washngton, DC. 20

23 Table Emgraton Estmates n Pror Research Annual umber Tme Perod of Emgrants Emgraton Emgrants of Estmate n thousands Rate per Arrvals All oregn Born Warren and Peck 980* Ahmed and Robnson 994* Mulder 2003* Recent Arrvals: n U.S. Less Than 0 Years Warren and Peck 980* Borjas and Bratsberg Mulder 2003* Recent Arrvals: n U.S Years Mulder 2003* Mexcan oregn Born Ahmed and Robnson 994* Mulder 2003* Massey and Snger , Recent Arrvals: n U.S. Less Than 0 Years Mulder 2003* Massey, Durand, and Malone 2002a Massey, Durand, and Malone 2002b Recent Arrvals: n U.S Years Mulder 2003* * Resdual estmate a unauthorzed Mexcan Mgrants b legal Mexcan Mgrants 2

24 Table 2 Emgraton Estmates of the oregn Born, Crca 2000 Based on CPS Matchng Method Annual Emgraton Rate Emgraton Rates % Annual Return Mgraton Rate Annual et Emgraton Rate Populaton Estmates for 2000 n thousands 2000 Populaton* Annual o. Emgrants Annual o. Return Mgrants Annual o. et emgrants All , Male , emale , Age , , , , , , Country/regon of Orgn Mexco , Central & South Am , Carbbean , Chna , Phlppnes , nda , Other Asa , Afrca Europe , orth Amerca Mgraton Status aturalzed , Legal mmgrant , Unauthorzed Mgrant , Refugee , Legal on-mmgrant Years n U.S , , , * calculated from 2000 March CPS usng Census 2000 weghts. 22

25 Table 3 Comparson of Emgraton Estmates Based on the Resdual Method vs. the CPS-Matchng Method CPS- CPS- Resdual Matchng Resdual Matchng Method a Method Dfference Method a Method Dfference All foregn born Recent Arrvals 0-0 yrs Longer-term resdents Country/regon of Orgn Mexco Central & South Am Carbbean Chna Phlppnes nda Other Asa Afrca Europe orth Amerca a Mulder 2003 et Emgraton Rate % Annual o. et Emgrants,