Migration and Fertility: Competing Hypotheses Re-examined

Transcription

1 Max-Planck-Insttut für demografsche Forschung Max Planck Insttute for Demographc Research Konrad-Zuse-Strasse 1 D Rostock GERMANY Tel +49 (0) ; Fax +49 (0) ; MPIDR WORKING PAPER WP OCTOBER 2003 Mgraton and Fertlty: Competng Hypotheses Re-examned Hll Kulu (kulu@demogr.mpg.de) Ths workng paper has been approved for release by: Jan M. Hoem (hoem@demogr.mpg.de) Head of the Laboratory of Contemporary European Fertlty and Famly Dynamcs. Copyrght s held by the authors. Workng papers of the Max Planck Insttute for Demographc Research receve only lmted revew. Vews or opnons expressed n workng papers are attrbutable to the authors and do not necessarly reflect those of the Insttute.

2 Abstract Competng vews exst concernng the mpact of geographcal moblty on chldbearng patterns. Early research shows that nternal mgrants largely exhbt fertlty levels domnant n ther chldhood envronment, whle later studes fnd mgrants fertlty to resemble more closely that of natves at destnaton. Some authors attrbute the latter to adaptaton, but others clam that selecton of mgrants by fertlty preferences may be the cause. Moreover, the short-term fertlty-lowerng-effect of resdental relocaton has also been proposed and challenged n the lterature. Ths paper contrbutes to the exstng dscusson by provdng an analyss of the effect of nternal mgraton on fertlty of post-war Estonan female cohorts. We base our study on retrospectve event-hstory data and apply ntensty regresson for both sngle and smultaneous equatons. Our analyss shows that frst, the rsk of brth decreases wth ncreasng settlement sze and the decrease s larger for hgherorder partes. Second, t shows that mgrants, whatever ther orgn, exhbt fertlty levels smlar to those of non-mgrants at destnaton. Our further analyss supports the adaptaton hypothess. We fnd no evdence on strong selectvty of mgrants by fertlty preferences, although we observe elevated fertlty levels after resdental relocatons arsng from unon formaton. Keywords: fertlty, nternal mgraton, ntensty regresson, smultaneous equatons, Estona 2

3 Durng the past century, many natons wtnessed ncreasng populaton spatal moblty and ts concentraton n the major centres of the country (Woods 2003). Whle the peak of rural-urban mgraton n (economcally) more developed countres passed a long tme ago, and the level of urbansaton has reached close to ts maxmum today, the overall spatal moblty does not seem to exhbt any sgns of decreasng. Consderng Zelnsky s (1971) theory of the moblty transton, the dfferent forms of spatal moblty smply replace urbansaton when a country proceeds from transtonal to advanced stage n ts demographc development. Increasng nter-urban movement, trends of suburbansaton and counter-urbansaton experenced by many European and North Amercan countres n the past (Geyer and Kontuly 1996), gve support for Zelnsky s general argument, despte the fact that the theory had not been able to foresee all these trends (Cadwallader 1993). Needless to say, movng from one place to another s an mportant lfe event, accompaned by both short and long-term changes n an ndvdual s lfe. Whle the effect of mgraton on dfferent lfe domans of an ndvdual seems rather self-evdent, dfferent vews exst concernng the mpact of a new socal envronment on chldbearng preferences and behavour of mgrants. Earler research has proposed four partly complementary, partly contradctory hypotheses about how the patterns of fertlty mght appear followng mgraton (Hervtz 1985; Rundqust and Brown 1989; Lee 1992; Sngley and Landale 1998). The socalsaton hypothess reles on the premse that fertlty behavour of mgrants reflects the fertlty preferences domnant n ther chldhood envronment. Therefore, mgrants exhbt smlar fertlty levels to stayers at orgn and the convergence towards fertlty levels of populaton at destnaton occurs only n the next generaton (gven that the dfferences exst). The adaptaton hypothess, n contrast, premses on an ndvdual s resocalsaton possblty, and suggests that the fertlty behavour of mgrants, sooner or later, comes to resemble the domnant behavour at the destnaton envronment. The selecton hypothess, n turn, argues that changng behavour s not a queston, yet rather the fact that mgrants are a specfc group of people whose fertlty preferences are more smlar to those of people at destnaton than at orgn. Fnally, the dsrupton hypothess suggests that mmedately followng mgraton, mgrants show partcularly low levels of fertlty due to the dsruptve factors assocated wth the mgraton process. Ths paper contrbutes to exstng dscusson, provdng an analyss of fertlty of nternal mgrants n Estona. The objectves are as follows: Frst, to examne the fertlty dfferences between people who move and those who stay 3

4 at varous orgn and destnaton envronments. Second, to look at the role of varous factors proposed n the lterature n accountng for observed fertlty patterns of mgrants. The study uses retrospectve event-hstory data, and apples ntensty regresson for both sngle and smultaneous equatons, wth the am to arrve at a more comprehensve nsght nto the causes of fertlty behavour of mgrants. The paper proceeds as follows: Frst, we wll gve an overvew of prevous research, and specfy arguments of dfferent vews. Then, we wll brefly descrbe the study context, and ntroduce the data, methods and varables. Thereafter, we wll present the results of multvarate analyss. Ths leads us fnally to the dscusson on the role of mgraton n shapng an ndvdual s fertlty behavour. Vews on the mpact of mgraton on fertlty The rse of the socalsaton hypothess n nternal mgraton-fertlty lterature s largely assocated wth Goldberg s (1959, 1960) two studes. Goldberg s man nterest was to examne the socoeconomc dfferences n fertlty n urban areas, whch many prevous studes had found. Whle research had establshed an nverse relaton of fertlty to socoeconomc status whte-collar famles were smaller than blue-collar famles, Goldberg hypothessed that n realty ths relatonshp mght not be so smple as t appeared. Namely, the larger fertlty of blue-collar workers mght result from occupatonal selectvty of rural mgrants whose fertlty was expected to be hgher than that of natve urban resdents because of dfferent chldhood socalsaton. To test the hypothess, Goldberg examned fertlty of populatons of Detrot and Indanapols. Both studes showed that people wth a rural background exhbted sgnfcantly hgher levels of fertlty than natve (two-generaton) urbantes, and the overall socoeconomc dfferences n fertlty could be attrbuted prmarly to the fertlty behavour of rural mgrants and ther concentraton n lower socal and economc postons n the cty. Several other papers studyng the socoeconomc dfferences of fertlty n the U.S. gave ndrect support for the socalsaton hypothess. Inspred by Goldberg s research, Freedman and Slesnger (1961) analysed the data on the U.S. total populaton and found that a tradtonally observed negatve correlaton between, ether ncome or educaton wth fertlty, dsappears when we consder only a natve urban populaton. Thus, overall socoeconomc dfferences n fertlty wthn an urban populaton dd result from the dfferences among rural-urban mgrants and ther over-representaton n lower ncome and educatonal groups. Duncan s (1965) research confrmed 4

5 prevously observed patterns and led hm to conclude that a modern fertlty pattern could be reached ether by non-rural rearng or by prolonged contact wth the educatonal system. Later, however, McGrr and Hrschman (1979) showed that sgnfcant socoeconomc dfferences dd exst among earler cohorts of rural-urban mgrants, but not among more recent generatons. Ther research also ndcated that despte decreasng socoeconomc dfferences, rural-urban mgrants had slghtly hgher fertlty n most educatonal groups. Surprsngly, however, from later lterature on fertlty of nternal mgrants, we can fnd very few studes dealng wth the socalsaton hypothess. One notable excepton s a study on fertlty of nter-regonal mgrants n Brazl by Hervtz (1985), where he found only lmted support for the hypothess. Meanwhle, many studes dealng wth fertlty of mmgrants have supported the man arguments of the socalsaton hypothess, although usng dfferent rhetorc (the assmlaton hypothess). Rosenwate s (1973) study showed that frst-generaton Italan-Amercans mantaned ther specfc fertlty behavour, whle the second generaton exhbted smlar behavour to natve Amercans. More recently, Stephen and Bean (1992) found smlar ntergeneratonal dfferences for Mexcan-Amercans, and Kahn s (1994) research showed no evdence of changes n fertlty behavour for most mmgrant groups n the U.S. Whle the socalsaton hypothess receved support manly n early nternal mgraton-fertlty lterature, the adaptaton hypothess seems to be wdespread and popular later on as well. Examples of early studes supportng the adaptaton hypothess are research by Myers and Morrs (1966), and Goldsten (1973). The former examned fertlty of nternal mgrants n Puerto Rco usng the census data on current resdence and place of brth. As opposed to domnant research at that tme, ther study showed that mgrants from rural to urban areas exhbted the same levels of fertlty as the natve urban populaton. Goldsten (1973) arrved at largely smlar results when examnng fertlty of rural-urban mgrants n Thaland. She found that the fertlty levels of mgrants, especally n the captal cty of Bangkok, were well below those of the non-mgrants n the rural areas from whch most of the mgrants came. Later, Hday (1978) showed that prevous fndngs also appled to fertlty of nternal mgrants n the Phlppnes. Whle all of these studes found the fertlty levels of mgrants to be more smlar to those of the populaton at destnaton than at orgn, the authors were stll careful to gve ther full support for the adaptaton hypothess, as t remaned unclear whether mgraton dd operate as a cause or an effect of low fertlty. 5

6 More recently, the adaptaton hypothess has been tested and supported by several authors. Farber and Lee (1984) examned the effect of rural-urban mgraton on fertlty n Korea. To control the possble preference selectvty, a model was constructed where they compared fertlty paths of ndvduals who had already mgrated (post-mgrants), and people who had not yet mgrated, but were known to mgrate later (pre-mgrants). The authors found two sgnfcantly dfferent paths, and concluded that rural-urban mgraton slowed down the fertlty rate of Korean women. Later Lee and Pol (1993) showed a sgnfcant rural-urban adaptaton n Mexco, but not n Cameroon, whch they attrbuted to the specfc context of Afrcan fertlty transton (the fertlty ncreasng effect of urban resdency due to reduced nfertlty). Brockeroff and Yang (1994), however, found support for the adaptaton hypothess n the Afrcan context as well. Ther comparatve study on fertlty of rural-urban mgrants n sx countres ndcated that mgrants rsk of concepton declned dramatcally n all countres around the tme of mgraton, and remaned low n the long run among most mgrant groups. Addtonal analyss has shown that the declne n mgrants fertlty could be largely attrbuted to a pronounced mprovement n the standard of lvng after mgraton and the ncreasng use of modern contraceptve methods. Brockeroff s (1995) subsequent study on fertlty of rural-urban mgrants n thrteen Afrcan countres confrmed prevous fndngs. Clear evdence supportng the adaptaton hypothess can lkewse be found n Hervtz s (1985) research on fertlty of nter-regonal mgrants n Brazl, and n a recent paper by Umezak and Ohtsuka (1998) on fertlty of rural-urban mgrants among the Anjangmu dalect group n Papua New Gunea. The selecton hypothess has been dscussed n many papers, but examned n only few studes. Myers and Morrs (1966), and Goldsten (1973), rased the queston of mgrant selectvty n the fnal sectons of ther papers. Stll, some studes at that tme also addressed the ssue. Macsco et al. (1970) compared fertlty of mgrants and non-mgrants n San Juan, Puerto Rco. Whle both groups exhbted sgnfcantly lower levels of fertlty than the rural populaton, the fertlty of mgrants was even lower than that of urban natves. The analyss showed that a hgher actvty rate and educaton level of mgrants explaned some of the dfferences, but not all. Ths led the authors to conclude that the rural-urban mgrants n Puerto Rco were more orented toward achevement and nnovaton, as were the stayers. Early marrage and chldren n rural areas mght be vewed as obstacles to upward moblty, and the response was to delay marrage and fertlty, favourng hgher educaton and mgraton to the captal cty (Zarate and Zarate 1975, 125). Hendershot s 6

7 (1971) smlar analyss showed lower mgrant fertlty at older ages, but hgher fertlty among younger mgrants to Manla n the Phlppnes. The author attrbuted the dfferences largely to changng mgraton streams: Whle n early stages of urbansaton, rural-urban mgraton was dffcult and therefore selectve, whereas n later stages t was less dffcult and therefore also less selectve (Zarate and Zarate 1975, 137). Many subsequent papers dscussed the ssue of mgrant selectvty (Goldsten and Goldsten 1981; Murphy and Sullvan 1985). Yet, the study by Courgeau (1989) on fertlty of rural-urban and urban-rural mgrants n France also provded clear evdence supportng the hypothess. The multvarate analyss of longtudnal data showed that mgraton to the cty sgnfcantly reduced a woman s fertlty, whereas mgraton to rural settlements ncreased t. However, further analyss revealed that mgraton to rural areas attracted women whose fertlty before the move was smlar to that of other women n the urban area, whle mgraton to urban areas attracted women whose fertlty was smlar to that prevalng n the urban areas. Thus, urban-rural mgrants adapted to the behavour domnant n the rural areas, whle rural-urban mgrants were a selectve group, accordng to ther fertlty preferences. Recently, Whte et al. (1995) found evdence supportng the selecton hypothess when analysng fertlty of nternal mgrants n Peru. More specfcally, the new resdents n larger locatons n general, and n the captal cty n partcular, were more lkely to arrve wth lower lfetme fertlty preferences. The dsrupton hypothess, assumng the short-term fertlty-lowerngeffect of mgraton event, has found drect or ndrect support n many studes. Goldsten s (1973) early analyss on mgrant fertlty n Thaland showed that fertlty levels of lfetme mgrants were not very dfferent from those of nonmgrants at destnaton, whle the fertlty of recent mgrants (those who have been lvng n a new destnaton less than fve years) was consderably lower. She attrbuted ths phenomenon to possble cohort changes or dsrupton effect, resultng from spousal separaton. In her later study, however, she and her collaborator tended to favour the dsrupton hypothess, although they alternatvely proposed that the phenomenon mght also be related to a low overall fertlty of mgrants that later caught up to the correspondng levels of urban populaton (Goldsten and Goldsten 1981). A few years later, Hervtz (1985) brought clear evdence supportng the dsrupton hypothess n hs study on mgrant fertlty n Brazl. More recently, Brockerhoff (1995) demonstrated a very low fertlty of urban-rural mgrants durng ther frst few years n ctes n several Afrcan countres, whch he attrbuted to the unmarred status of 7

8 mgrants and to hgh levels of spousal separaton among marred mgrants. Whte et al. (1995) analysed mgrant fertlty n Peru usng longtudnal data, and showed that resdental relocaton lengthened the brth nterval of mgrants. Recently, however, several authors studyng mmgrant fertlty have mplctly or explctly challenged the dsrupton hypothess. Sngley and Landale (1998) compared the rsk of the frst brth of several groups of Puerto Rcan women (born n Puerto Rco, but resdng n the U.S., resdng n Puerto Rco and the U.S.-born Puerto Rcans) usng longtudnal data. Ther analyss revealed that sngle women mgratng to the U.S. were much more lkely than ther non-mgrant counterparts n Puerto Rco to form unons and experence a concepton, ether n unons or outsde. The authors concluded that mgraton to the U.S. should be seen as a part of the famly buldng process for many Puerto Rcan women. Andersson (2001) arrved at very smlar conclusons when examnng mmgrant fertlty n Sweden. The analyss of rsk of the frst brth showed elevated levels of chldbearng durng the frst couple of years after mmgraton to Sweden. Moreover, the author found mgraton to trgger, rather than dsrupt the process of chldbearng, also for hgher brth orders. The study by Mulder and Wagner (2001) on famly formaton and home ownershp n West Germany and the Netherlands, n turn, demonstrated ncreasng rates of frst chldbrth shortly after a couple had moved to ther own house. To sum up, dfferent hypotheses have been proposed to predct and explan fertlty patterns of mgrants. Each of these hypotheses has receved support n the lterature, and has also been challenged. Each perspectve draws upon some theoretcal vew, assumng a varety of factors to be more mportant than others n shapng mgrant s chldbearng behavour. Socalsaton hypothess emphasses the crtcal role of the chldhood envronment. The norms and values domnant n a mgrant s chldhood envronment gude her/hs later actons n other places as well. To the contrary, the adaptaton hypothess assumes that what matters most n shapng a mgrant s fertlty behavour s her/hs current sococultural and economc envronment. Selecton hypothess also seems to emphasse the mportance of the chldhood envronment. However, norms and values dffer across populaton subgroups, and the mnorty later moves to locatons where values smlar to thers domnate. Fnally, the dsrupton hypothess argues that economc and psychologcal costs of resdental relocaton cannot be underestmated when studyng fertlty patterns of mgrants. Contradctory conclusons of studes often arse from dfferent tme, context and types of mgratons nvestgated. Varous methodologes used may also account for some dfferences. In ths context, some crtcal remarks on 8

9 domnant research methodologes are nevtable. Frst, most studes use crosssectonal (usually census) data whle studyng the effect of mgraton on fertlty. Longtudnal data have found only lmted use, despte ther domnant poston n many areas of populaton research. Clearly, the lack of nformaton on the precse tmng of mgraton, fertlty and other factors restrcts any causal nferences to be made about the mgraton-fertlty relatonshps. Second, most recent studes have successfully controlled for selecton of mgrants by varous socoeconomc factors, whle assessng the mpact of mgraton on fertlty. However, the possble unobserved selectvty of mgrants by fertlty preferences has been addressed only n a few studes (e.g. Courgeau 1989; Montgomery 1992; Mcheln 2002). Ths fact has further ntensfed the dffcultes when drawng conclusons about mgraton-fertlty relatonshps. Therefore, usng retrospectve event-hstory data, and controllng for unobserved selecton of mgrants when examnng the effect of varous factors on mgrant fertlty, should be seen as major contrbutons of ths paper. Before we ntroduce the data and methods, however, we wll brefly outlne the context of the research. Fertlty and mgraton trends n Estona The begnnng of the fertlty transton n Estona can be traced back to the md-19th century. The 1850s 60s sgnfes a perod when fertlty levels began to decrease and gradually approached levels characterstc of a modern fertlty regme. By the late 1920s, perod fertlty had already reached below replacement level n Estona (Katus 2000, ). Wth ts relatvely early fertlty transton, Estona (and neghbourng Latva) has been assocated wth the group of poneerng natons of demographc transton, along wth France, Swtzerland, Sweden and Norway (Katus et al. 2002, 143). Whle earler fertlty development n Estona followed patterns common n Western and Northern Europe, the post-wwii trends dffered. Estona dd not experence a post-war baby boom, and perod fertlty remaned below replacement level untl the md-1960s. Thereafter t slghtly ncreased among ts natve populaton, and stayed above replacement level untl the late 1980s, when a rapd fertlty decrease, characterstc of the post-socalst transton perod, began. Cohort fertlty, based on the Estonan Famly and Fertlty Survey, show relatvely stable and low levels among cohorts of natve people born from the 1910s to the 1940s, and a slght ncrease among the cohorts born n the late 1940s and the 1950s (Katus et al. 2002, 145). 9

10 Whle fertlty levels remaned rather stable n post-war Estona, the trends n populaton mgraton were far more dynamc. Besdes the ntensve mmgraton from other parts of the Sovet Unon (Kulu 2003), ntensve urbansaton also charactersed the post-war perod. In 1939, 33% of the Estonan populaton lved n urban areas. The share of urban populaton ncreased to 56% n 1959 and to 71% n 1989 (Tammaru 2001a, 580). The ctes grew both as a result of (nternal) rural-urban mgraton and mmgraton that was overwhelmngly destned for urban areas. Whle external mgraton fed the urban growth untl the late 1980s, trends n nternal mgraton were dfferent. Snce the late 1970s, urban-rural mgraton ncreased, and n the followng decade, rural areas n Estona wtnessed a postve net mgraton for the frst tme n the country s hstory. The mgraton turnaround has been attrbuted to ncreasng nvestments by the state n agrcultural producton, and also to the changng preferences of people (cf. Marksoo 1992, 134). Durng the 1990s, the concentraton n major centres agan became the domnant trend among the workng age populaton, and was later accompaned by ncreasng sub-urbansaton. The share of the urban populaton, however, decreased from 71% n 1989 to 67% n 2000, mostly as a result of emgraton of Russans and other ethnc mnortes (Tammaru 2001a). The fertlty of nternal mgrants n Estona has not been studed, nor s much known about regonal varaton n fertlty levels and urban-rural dfferences. In ths context, a follow-up of the recent census (2000) data on the Estonan natve populaton s hghly nformatve. We see sgnfcant dfferences that exst n fertlty levels of populaton across settlement herarchy (Fgure 1). Fertlty n rural areas s clearly above replacement levels n all brth cohorts who have already passed, or are about to reach the end of ther chldbearng ages. Fertlty n the urban populaton, however, remans below replacement level, comprsng about 80% and 65% of rural levels n towns and the captal cty of Tallnn, respectvely. (Very smlar pcture also apples for the mmgrant populaton.) It s also strkng that fertlty dfferences are rather stable, and do not change much across the cohorts, as one mght assume. Data and research populaton The data for our study come from the Estonan Famly and Fertlty Survey. The Estonan FFS was carred out n 1994 among 5,021 women born between (see Katus et al. 1995), usng the 1989 census as a sample frame. The share of those surveyed was 81% (5,021 from 6,212), or 86%, leavng asde over-coverage (those who had ded or had left Estona n ). A 10

11 comparson between women who were surveyed wth those who were not, and the total female populaton by major socodemographc varables, showed that there were no sgnfcant dfferences (Katus et al. 1995, 18 21). As part of the Europe-wde FFS program, the survey was based on the collecton of event hstores. All major demographc events that had taken place n the lfe of the respondent were dentfed (to the accuracy of a month), ncludng brths, coresdental unons and resdental changes snce age 14. In the FFS program, collecton of mgraton hstores was optonal, and only a few countres mplemented ths module. The hgh response rate, the multple retrospectve hstores collected, and the hgh qualty of the collected retrospectve nformaton n the Estonan FFS provde a good bass for studyng the effect of mgraton on fertlty n more detal. Our research populaton conssts of the natve female populaton born from We focus on only natve people because we wsh to have a homogeneous populaton when testng competng hypotheses. In total, there are 1918 natve women n the data set. However, we exclude 43 women who have not ndcated the date of ther unon dssoluton for some reason or other. Therefore, our fnal research sample conssts of 1875 natve females born from We study the mpact of mgraton on ther frst, second and thrd conceptons (leadng to brths). There are 1556, 1005 and 358 such events, respectvely. We defne mgraton as a resdental change that crosses the border of an urban settlement or rural muncpalty, and lasts for more than three months. We go beyond tradtonal urban-rural-dchotomy and dstngush three types of settlements of orgn and destnaton of mgraton: 1) rural areas (less than 2,000 nhabtants); 2) small and medum-sze towns (2, ,000); and 3) a large cty or the captal (the captal beng Tallnn, wth more than 400,000). The dstrbuton of the sample populaton across settlement-type s as follows: 46% of women lved n rural areas at age 14, 36% n small towns, and 18% n the captal cty. The correspondng fgures at the tme of ntervew were 37%, 38% and 25%. The share of mgrants was 80%, 74% and 54%, respectvely. We splt the data-set by concepton epsode, followng general logc of event-hstory data set up. Indvduals are at rsk snce age 14 (for the frst concepton) or prevous brth (for the second and thrd conceptons). The fnal censorng takes place at ntervew (actually, nne months before) or at age 45. Resdental epsodes outsde Estona are excluded from our analyss. If concepton occurs smultaneously (n the same month) wth mgraton and/or unon formaton, we use the sequence of events as follows: mgraton, unon formaton and concepton. Thus, we assgn smultaneous conceptons to the 11

12 destnaton envronment, whch, as we wll see later, turns out to be a reasonable soluton. We also buld a mult-epsode data-set for mgraton, whch we need for later smultaneous analyss. The rsk of mgraton starts at age 14, or at the prevous mgraton. In total, there are 3063 mgraton events n our data-set: 1183 of whch were destned to rural areas, 1255 to small towns and 625 to the captal cty. Methods and analytc strategy We use ntensty regresson or (multvarate) ndrect standardsaton (Hoem 1993) as a research method. We estmate several models n order to further examne varous hypotheses proposed n the lterature. We begn wth a smple model where we look at the effect of mgraton on concepton, controllng for only baselne duraton (tme snce age 14 or prevous brth), partnershp status (n case of the frst conceptons) and unon duraton. The results outlne possble dfferences between mgrants and non-mgrants at varous places of orgn and destnaton, and therefore gve us prelmnary evdence about how and whether mgraton shapes fertlty behavour. Thereafter, we also nclude n our analyss other background varables of ndvduals to further control for demographc and socoeconomc selectvty of mgrants when assessng the mpact of mgraton on fertlty. Our basc model can be formalsed as follows: (1) ln ( ) y ( t ) z k ( u k t ) α j x j µ t β w ( t ), k j l l l where µ (t) denotes the ntensty of concepton (frst, second or thrd) for ndvdual, y(t) denotes a pecewse lnear splne that captures the mpact of baselne duraton on the ntensty. The z k (u k + t) denotes the splne representaton of the effect of a tme-varyng varable that s a contnuous functon of t wth orgn u k. The x j represents the values of a tme-constant varable and w l (t) represents a tme-varyng varable whose values can change only at dscrete tmes. After havng outlned the basc dfferences between people who moved and those who stayed at varous orgn and destnaton envronments, we look next at the possble role of unobserved selectvty accountng for dfferences between mgrants and non-mgrants (whch we expect to fnd). We have bult a smultaneous-equatons model for that purpose, whch jontly estmates three equatons for fertlty, and three equatons for mgraton (accordng to destnaton of resdental change). Both processes have ther (person-specfc) 12

13 heterogenety terms, and allowng correlaton between the resduals we dentfy possble endogenety of mgraton n the fertlty process and control for the unobserved selectvty when analysng the effect of mgraton on fertlty (Lllard 1993). The model can be formalsed as follows: ln µ ln µ ln µ (2) ln µ ln µ C1 C2 C3 R S L ( t ) y ( t ) y ( t ) y ( t ) y ( t ) y ln µ ( t ) y R S L C1 C2 C3 ( t ) ( t ) ( t ) ( t ) ( t ) ( t ) C1 z C1 k ( u k t ) α j x j j C2 C2 z k ( u k t ) α j x j j C3 C3 z k ( u k t ) α j x j j R R z k ( u k t ) α j x j j βl S S S z k ( u k t ) α j x j j βl L L L z k ( u k t ) α j x j j βl C1 k j l l l C2 β k j l l w l C3 β w k j l l l R w k j l l k j l l k j l l w β w w ( t ) ε ( t ) ε ( t ) ε ( t ) ε ( t ) ε ( t ) ε where µ C1 (t), µ C2 (t), µ C3 (t) denotes the ntenstes of the frst, second and thrd conceptons, respectvely, and µ R (t), µ S (t), µ L (t) represents the rsks of mgraton to rural, small urban and large urban destnatons n the competng rsk framework. ε C and ε M are person-specfc heterogenety terms for fertlty and mgraton processes, respectvely, and are assumed to have a jont bvarate normal dstrbuton. The dentfcaton of our model s attaned through wthnperson replcaton: many people have gven several brths, and some people have also made several moves (see Lllard et al. 1995, 446). However, there s reason to beleve that the nature of selectvty may depend on the destnaton of mgraton. As the lterature revewed showed, larger ctes may attract people who prefer smaller famles for some reason or other, whle mgrants to rural settlements may desre many chldren. Therefore, we have to extend our smultaneous-equatons model allowng separate heterogenety terms for each mgraton equaton: ln µ ln µ ln µ (3) ln µ ln µ C1 C2 C3 R S L ( t ) y ( t ) y ( t ) y ( t ) y ( t ) y ln µ ( t ) y R S L C1 C2 C3 ( t ) ( t ) ( t ) ( t ) ( t ) ( t ) C1 z C1 k k ( u k t ) α j j x j j l C2 C2 z k ( u k t ) α j x j j C3 C3 z k ( u k t ) α j x j j R R z k ( u k t ) α j x j j βl S S S z k ( u k t ) α j x j j βl L L L z k ( u k t ) α j x j j βl C2 k j l l l C3 β k j l l w l R w k j l l k j l l k j l l w β w C1 l β w w l ( t ) ε M M ( t ) ε ( t ) ε M ( t ) ε C ( t ) ε ( t ) ε where ε R, ε S and ε L are heterogenety terms for mgraton to rural, small urban and large urban areas, respectvely. Agan, allowng correlaton between (person-specfc) resduals of fertlty and mgraton equatons, we test L R S C C C C C,, 13

14 endogenety of mgraton n the fertlty process, and thus elmnate a possble bas whle estmatng the effect of resdental change on fertlty. To dentfy the model, we have no need of nstruments at ths tme, as some people have made several moves towards the same destnaton. So far our modellng strategy has focussed mostly on testng three basc hypotheses n general, and the queston of adaptaton versus selecton n partcular. However, the dsrupton hypothess also needs examnaton. In order to clarfy the effect of mgraton event on fertlty, we have to make our statc models more dynamc. Therefore, n our fnal models, we allow the ntensty of concepton at destnaton to vary over tme snce arrval n the settlement, nstead of assumng a constant rsk. Techncally, ths s acheved by usng the lnear splne representaton to capture the effect of tme at destnaton on fertlty. We estmate our dynamc models both separately and jontly n order to see the dfference as a result of possble selecton. Explanatory varables and hypotheses We do have equatons for both fertlty and mgraton processes. In fertlty equatons, varables reflectng an ndvdual s resdental hstory hold a central poston. In the analyss we nclude a (tme-varyng) varable showng an ndvdual s current resdence, and a varable ndcatng resdence at age 14 (for mgrants). Although the chldhood settlement may have some role n later fertlty behavour, we draw upon recent nternal mgraton-fertlty lterature and hypothesse the followng: mgrants to larger settlements exhbt lower levels of fertlty than stayers at orgn, and mgrants to smaller places, n turn, have hgher fertlty than stayers at larger settlements. If the hypothess s supported, then fndng an answer to the queston of adaptaton versus selecton becomes a major task of our further analyss. We also expect to fnd evdence supportng the dsrupton hypothess, although as some prevous studes have shown, the rsk followng mgraton may depend on the type of mgraton (ths we can control to some extent by means of smultaneous analyss). In addton, we nclude a varable showng the number of mgratons to capture the nterm experence of mgrants. We control for several demographc and socoeconomc varables when testng dfferent hypotheses concernng the mpact of mgraton on fertlty. Our duraton varables are age and tme snce prevous brth (for the second and thrd conceptons). We expect the probablty of the frst concepton to be the hghest n the early twentes, as prevous studes have shown relatvely early chldbearng of post-war Estonan women (Vkat 1994; Katus et al. 2002,

15 155). The ntenstes of the second and thrd conceptons rse rapdly n the frst sx months after a brth, and reman at hgh levels durng the subsequent year or two, before they begn to decrease (Katus et al. 2002, 158; cf. Lllard 1993, 675). Second, we nclude n our analyss, an ndvdual s partnershp status and unon duraton for those n unon. We expect unon formaton and marrage to sgnfcantly rase the probablty of concepton, and the rsk to gradually decrease wth duraton of unon (Bazan et al. 2003, 157). The next varable represents (calendar) tme to capture the mpact of changng context. We hypothesse slghtly ncreasng fertlty snce the late 1960s, but a sharp decrease n fertlty levels n the 1990s (Katus 2000; Phlpov 2002, 7). We nclude educatonal enrolment, employment status, and level of educaton to control the effect of an ndvdual s socoeconomc characterstcs. We assume the probablty of concepton to be hgher when an ndvdual has completed her/hs studes (Sngley and Landale 1998, 1459; Bazan et al. 2003, 157). We expect employed women to have a hgher ntensty of concepton (for the frst concepton, at least) compared to those who are nactve n the context of full (and compulsory) employment. The possble fertlty patterns across educatonal groups are more dffcult to predct. Earler studes showed an nverse relatonshp between educaton and fertlty, whle more recent research has demonstrated a relatvely hgh rsk for second and thrd brths for educated women. However, that seems to dsappear when controllng for relatve age of educated women, ther partners characterstcs and other (usually) unobserved factors (Hoem et al. 2001b; Kravdal 2001; Kreyenfeld 2002). We hypothesse an nverse relatonshp between fertlty and educaton regardng frst concepton, and a relatvely hgh rsk for second concepton for educated women (Katus et al. 2002, 177). However, that may vansh n the course of smultaneous analyss. Fnally, we assume the number of sblngs to be postvely related to an ndvdual s fertlty (Hoem et al. 2001a, 46; Bazan et al. 2002, 39), and women belongng to the Russan (hstorcal) mnorty n Estona to exhbt slghtly lower levels of fertlty (Sakkeus 2000, 278). Whle our paper focuses mostly on the mpact of mgraton on fertlty, the analyss also allows us to examne the determnants of mgraton n post- WWII Estona. All varables n the fertlty equaton may enter nto the mgraton equaton (f needed), as our models are dentfed through wthnperson replcaton. Our baselne varables are age and tme snce prevous resdental change (for the second and hgher mgratons). In keepng wth the lterature, we assume the ntensty of mgraton to reach ts peak at late adolescence, when the majorty of cohorts complete ther (secondary) educaton and contnue ther studes (often elsewhere), or enter nto the labour 15

16 market (Katus et al. 1999, 16; Sjöberg and Tammaru 1999, 828; see also Ma and Law 1997, 237). Concernng the effect of tme snce prevous mgraton, we expect ncreasng rsk durng the frst few years, and decreasng ntenstes thereafter (Gordon and Molho 1995). Second, we hypothesse sngles and dvorced people to be more moble than those already n a unon (cf. Mulder and Wagner 1993, 73; Newbold and Law 1995, 125). Thrd, we assume the presence of chldren n a famly to sgnfcantly decrease the probablty of urban-ward mgraton, yet possbly to ncrease the propensty to move to rural destnatons (Courgeau 1989, 140). The next varable represents (calendar) tme, and we hypothesse an ncreasng mgraton rsk towards rural areas n the 1980s, and decreasng overall ntenstes n the early 1990s as a result of economc hardshp arsng from post-socalst transton (Marksoo 1992; Kulu and Bllar 2003). Sxth, we assume ncreasng mgraton ntenstes after studes have been completed, and also a hgher rsk when an ndvdual s out of the labour market (cf. Fscher and Malmberg 2001, 265). Seventh, we hypothesse ncreasng moblty as t correlates to an ndvdual s rsng level of educaton. Ths s a result of ncreasng optons due to educaton and a larger dsperson of jobs of more educated ndvduals (Courgeau 1985, 159; Newbold 1999, 266). Next, there s also reason to beleve that the presence of sblngs rases the probablty of the frst mgraton (home-leavng), at least (Courgeau 1989, 136). We also assume that ethnc Russans move less than Estonans do, to rural areas n partcular (Kulu and Bllar 2003). Fnally, we hypothesse decreasng mgraton ntensty wth ncreasng settlement sze for non-mgrants. The pattern for mgrants s lkewse expected to depend on resdence at age 14 and on the number of prevous moves (Kulu 2002). Results: mpact of mgraton on fertlty We began our modellng by runnng a set of models to examne the effect of varous destnaton envronments on fertlty of mgrants wth dfferent orgns (e.g. a woman wth a rural background n a small town, wth a large cty orgn n a rural settlement etc.). As we found no sgnfcant varaton among mgrants wth dfferent orgn (resdence at age 14 or prevous resdence) when lvng n the same (destnaton) envronment, we decded to collapse the categores of orgn and leave only the destnaton of mgraton n our man analyss, wth one excepton: the resdence at 14 s ncluded n the models for the thrd brth. Thus, there are sx resdental categores n most cases: non-mgratonal and mgratonal epsodes n rural, small urban and large urban areas, respectvely. 16

17 The epsode s non-mgratonal f an ndvdual has not moved snce age 14. Mgratonal epsodes are defned accordng to destnaton of mgraton, whatever the orgn of mgrants. Let us now present the results of our man analyss. In the frst model, we look at the effect of mgraton on concepton, controllng only for baselne duraton, partnershp status (n case of the frst concepton) and unon duraton. We see that the ntensty of the frst concepton of resdents of rural settlements, both non-mgrants and mgrants, and those of small towns, does not dffer sgnfcantly (Tables 1 and 2, model 1). The major dvdng lne runs between ths pool of people and people lvng n a large cty, whatever ther orgn. Natves n the captal cty have 34% and mgrants have a 28% lower rsk of the frst brth than the natve rural populaton, for example. The results on the second concepton are dfferent. Here, the major dvson exsts between resdents of rural and urban areas. Non-mgrants n small towns have 42% and those n large ctes have a 45% lower rsk of the second brth than the natve rural populaton. Agan, mgrants exhbt smlar levels of rsk to the nonmgrants at destnaton. Therefore, mgrants from rural to urban areas also have a sgnfcantly lower ntensty as compared to stayers n rural areas, whle urban to rural mgrants have a hgher fertlty rate than non-mgrants n urban areas. The results on the thrd concepton have ther specfc character as well. The mpact of settlement herarchy s now clearly present the larger the settlement the lower the rsk of the thrd brth. Natves n small towns have 34% and those n large ctes have a 58% lower rsk of the thrd brth as compared to a non-mgrant populaton n rural areas. At frst t seems that mgrants have an even lower rsk of the thrd concepton than non-mgrants at destnaton. However, further analyss shows that the dfferences are not sgnfcant. Resdence at age 14 also shapes the patterns of the thrd brth. Surprsngly, however, mgrants who lved at small towns at age 14 exhbt the lowest ntensty levels, whatever ther later resdence, whle the rsk for those socalsed n a large cty seems to be the hghest. Fnally, our analyss ndcates that the number of prevous mgratons also matters. People who have moved twce or more have a hgher rsk of concepton than those who have mgrated only once. However, the dfferences are sgnfcant only n regards to the frst two brths. Next, we have ncluded n our analyss all background varables of ndvduals to further control for demographc and socoeconomc selectvty of mgrants when assessng the effect of mgraton on fertlty. The dfferences outlned above decrease slghtly, but reman sgnfcant (Tables 1 and 2, model 17

18 2). Thus, our analyss supports prevous fndngs that dfferences between resdents of varous settlements grow wth ncreasng party, and mgrants (whatever ther orgn) exhbt fertlty levels smlar to those of non-mgrants at destnaton, wth a possble mnor excepton for the thrd brth. Whle our results gve lmted support for the socalsaton hypothess (regardng the thrd brth), t s now clear that our major task s to clarfy why the fertlty of mgrants s smlar to that of natves at destnaton. Does ths result from mgrants adaptaton, or rather further selectvty of mgrants that s unobserved n ths case? In the next stage, we nclude person-specfc resduals nto fertlty and mgraton equatons, allowng for correlaton between heterogenety terms. The model ft mproves sgnfcantly (the value of lkelhood rato test statstc (LR) s wth 3 degrees of freedom, p-value s < 0.01). The standard devaton of resduals s sgnfcantly dfferent from zero n both cases (Table 1, model 3). Moreover, the correlaton coeffcent s postve (0.39) and sgnfcant. Thus, mgrants (some of them, at least) have unobserved characterstcs that ncrease ther probablty of chldbearng. Controllng for ths unobserved selectvty, however, does not change our prevous results substantally. Only the mpact of the number of mgratons dsappears, and the dfference between rural natves and other groups (ncludng rural mgrants) ncreases slghtly. Thus, what we have establshed s the fact that some mgratons are drectly related to the chldbearng process, and/or that strong postve selecton of mgrants by fertlty preferences operates towards some destnaton. As a result, the overall fgures also follow ths pattern. We should contnue allowng possble selecton to vary across destnaton of mgraton. To further examne the ssue of selectvty, we nclude n our analyss a person-specfc resdual for fertlty equatons and separate heterogenety terms for each mgraton equaton, allowng for correlaton between the resduals. Agan, the model ft mproves sgnfcantly, as compared to the prevous one (LR = wth 7 degrees of freedom, p-value s < 0.01). The standard devatons of all four heterogenety terms are sgnfcantly dfferent from zero (Table 1, model 4). More nterestngly, whle correlatons between the (personspecfc) resdual of fertlty equatons and that of mgraton to rural and small towns are postve, the correlaton wth resdual of equaton for the large urban destnaton s not dfferent from zero. Therefore, the unobserved selectvty of mgrants, whatever ts meanng, operates towards rural and small urban destnatons. However, the coeffcents of our man nterest do not change as sgnfcantly mgrants exhbt rather smlar fertlty levels to the natve populaton at destnaton. Stll, comparng the current results wth that of 18

19 separate modellng (model 2), we notce that the coeffcents for mgrants to rural and small urban areas are upward based, although slghtly, n the sngleequaton model. Interestngly, the same holds regardng frst, second and thrd conceptons, and apples lkewse to non-mgrant groups. To sum up, the smultaneous analyss largely supports our prevous fndngs. Furthermore, t shows that unobserved selecton of mgrants s a realty (n some cases), but ts mpact to refute the adaptaton hypothess s not very strong 1. So far, we have assumed a constant fertlty rsk for mgrants at destnaton. Next, we extend our second and fourth models, allowng the ntensty of concepton to vary over tme snce arrval n the settlement. Ths strategy enables us not only to examne the dsrupton hypothess n more detal, but also to gan further nsght nto the selectvty ssue. We focus only on the rsk of the frst concepton, as possble changes n tme are expected to be most colourful here (and the number of events suffcent for more detaled analyss). We present our results n the graph n order to assst n nterpretaton of the results. The results of separate modellng are presented frst. We notce dfferent tme patterns for mgrants to rural and small urban destnatons, on the one hand, and for mgrants to large urban destnatons, on the other hand (Fgure 2). In the former, the rsk after mgraton s very hgh (mgrants to rural areas have about 44% hgher rsk than rural non-mgrants, and those to small towns have an even hgher rsk, at 95%, than natves n small towns), whch then quckly decreases. In the latter, the ntensty s a very low rght after mgraton (37% lower than that of urban natves) and then ncreases. We also see that the rsk of the frst concepton mostly changes durng the frst half of a year, wth no sgnfcant changes later. (The model ft, however, mproves only slghtly (LR = 15.0 wth 9 degrees of freedom, p-value s < 0.10), pontng to the fact that not all parameter estmates are sgnfcantly dfferent from zero.) How could we nterpret the observed patterns? Clearly, the results tell us that some (or even many) mgratons to rural and small urban areas are drectly related to chldbearng (and famly formaton) n regards to women. Ths s not surprsng, consderng the recent fndngs by studes on mmgrants fertlty. However, ths s only one sde of the con. On the other sde s that ths concluson apples much less for mgratons to large urban destnatons. On the contrary, a relatvely low rsk rght after mgraton seems to gve support for the dsrupton effect: the settlng-n n a large cty takes some tme, and chldbearng s postponed n most cases, although only for a perod of a few months. Whle the nature of unobserved selectvty found n prevous analyses s also becomng clear for us, the results of smultaneous analyss gve further valuable nformaton. We see that the rsk of the frst concepton after 19

20 mgraton reduce sgnfcantly, when controllng for unobserved selectvty of mgrants, although t remans relatvely hgh n regards to mgrants to small towns (Fgure 2) 2. Meanwhle, the longer-term fertlty patterns of movers do not change very much mgrants n dfferent destnatons stll exhbt rather smlar fertlty levels to the non-mgrants there. The rsk of concepton for mgrants to rural areas s relatvely low, but the dfference (compared to natves at destnaton) s not sgnfcant. Thus, whle some mgratons are drectly related to chldbearng (and famly formaton), we fnd no evdence to conclude that certan areas attract people wth the fertlty behavour domnant there. Rather, mgrants tend to adapt to fertlty levels prevalent at destnaton, and sometmes postpone chldbearng for a perod of tme n order to overcome economc and psychologcal costs arsng from a resdental relocaton. Results: mpact of other varables Let us now dscuss the effect of other varables on fertlty. The results are for the most part as expected. The rsk of the frst concepton s the hghest n the early twentes, confrmng the relatvely early start of chldbearng of post-war Estonan women (Vkat 1994; Katus et al. 2002, ) (Table 1). The baselne ntenstes for the second and thrd conceptons largely follow patterns shown n other studes: they rse rapdly durng the frst sx months after prevous brth and then decrease (cf. Lllard and Wate 1993, 666; Hoem et al. 2001a, 46). As expected, both unon formaton (n case of frst brth) and marrage sgnfcantly ncrease the probablty of concepton. However, the rse s an extreme upward surge and the subsequent decrease rather steep, whch ponts to a concentraton of many conceptons (frst two, at least) n the begnnng of the unon, contrary to patterns found n other countres (Bazan et al. 2002, 39). Regardng changes over tme, the rsk of the second brth (at least) rose n the 1970s and the 1980s, and the ntenstes of all partes decreased n the 1990s as expected (Katus 2000; Phlpov 2002, 7). The lower rsk of concepton durng studes also corresponds to expectaton. Lkewse, t s not a surprse that there s a hgher rsk of the frst concepton durng employment, n the context of a planned economy where nactvty before chldbearng mght ndcate possble health problems. Our analyss supports prevous fndngs on the relatvely hgh rsk of a second brth of hghly educated women, showng them as carrers of twochld norm n post-war Estona (Katus et al. 2002, 177). Also, a hgher ntensty of the thrd concepton among less educated women corresponds to prevous fndngs. The analyss, however, does not confrm the nverse 20

21 relatonshp between educaton and the frst brth found n most studes. (Our further analyss reveals that ths pattern holds regardng younger brth cohorts, and also when separate (person-specfc) resduals are allowed for equatons of the frst and second-thrd brths. The ssue, however, requres more detaled treatment, whch goes beyond the scope of ths study.) The analyss also supports the role of sblngs and ethnc orgn n shapng fertlty patterns. As expected, the larger the number of sblngs, the hgher the fertlty. Belongng to the Russan mnorty, n turn, decreases the probablty of the second and (obvously also) the thrd conceptons. Fnally, we also tested the effect of varables showng parental dvorce, whether a prevous chld was conceved wth the same partner or not, and whether a prevous chld was born n the current resdence. None of these varables had a sgnfcant mpact on fertlty, and we excluded them from our man analyss. However, very relgous women exhbted hgh levels of the thrd brth ntenstes, as was expected (Hoem et al. 2001a, 46), but we also excluded ths varable, as there were too few cases and events n the most nterestng group. Results: determnants of mgraton Next, we brefly dscuss determnants of nternal mgraton of post-war Estonan female cohorts. The analyss largely supports fndngs of prevous studes, although some dfferences can also be outlned. As expected, the ntenstes of mgraton to all destnatons are the hghest at late adolescence, and thereafter gradually decrease (Tabel 3) (Katus et al. 1999, 16; Sjöberg and Tammaru 1999, 828). Our analyss also shows ncreasng moblty of mgrants durng the frst three to four years after resdental relocaton, and a subsequent decrease. The fact that people n a unon have a lower rsk of mgraton to ctes than sngles, n addton to the hgher moblty of separated people, corresponds to expectatons (cf. Mulder and Wagner 1993, 73). We also fnd the presence of chldren to sgnfcantly decrease the probablty of movng to urban areas n general, and to the captal cty n partcular, but we have not found ths to affect the moblty towards rural settlements, as shown n some other studes (Courgeau 1989, 140). Stll, further analyss ndcates that chldren (especally the second or a subsequent chld) ncrease the probablty of movng to rural areas from smaller towns (but not from a large cty). Decreasng mgraton ntenstes towards urban areas n the 1980s, and also to rural areas n the 1990s, are consstent wth the socoeconomc changes n Estona durng the late socalst and early transtonal perods (Marksoo 1992; Kulu and Bllar 2003). 21