The Household Registration System and Rural-Urban Educational Inequality in Contemporary China

Transcription

1 P o p u l a t o ns t u d e sce n t e r R e s e a r c hr e p o r t X a oga ngwu T hehous ehol dreg s t r a t ons y s t em a ndrur a l Ur ba ne duc a t ona l I nequa l t y ncont empor a r ych na P o p u l a t o ns t u d e sc e n t e r Un v e r s t yo f M c h g a n I n s t t u t ef o r S o c a l R e s e a r c h

2 The Household Regstraton System and Rural-Urban Educatonal Inequalty n Contemporary Chna Xaogang Wu Assocate Professor of Socal Scence Hong Kong Unversty of Scence & Technology Clear Water Bay, Kowloon Hong Kong SAR, CHINA Emal: sowu@ust.hk Populaton Studes Center Research Report Aprl 2011 Paper presented n the Internatonal Socologcal Assocaton/Research Commttee on Socal Stratfcaton and Moblty (ISA-RC28) Sprng Meetng n Florence, Italy, May th, I am grateful for the a post-doctoral fellowshp from the Natonal Academy of Educaton/Spencer Foundaton and a General Research Fund from the Hong Kong Research Grants Councl (644208) to support ths project. Address all correspondence to Xaogang Wu, Socal Scence Dvson, Room 3377 Academc Buldng, Hong Kong Unversty of Scence and Technology (HKUST), Clear Water Bay, Kowloon, Hong Kong. Emal: sowu@ust.hk.

3 Household Regstraton System and Educatonal Inequalty n Chna 2 ABSTRACT Ths paper examnes the effect of the household regstraton (hukou) system, based on whch Chnese ctzens were desgnated as ether rural or urban status and enttled to dfferent lfe chances, on educatonal nequalty n contemporary Chna. Analyses of data from a natonal representatve survey n 2005 consstently reveal a sgnfcant educatonal gap between people of dfferent hukou status, even after controllng for other covarates. Furthermore, to rule out the selectvty effect of hukou assgnment based on educaton, endogenous swtchng regresson models are employed to demonstrate a causal mpact of hukou status on educatonal attanment. The paper thus calls for attenton to be pad to the hukou nsttuton n studyng educatonal stratfcaton and the reassessment of the role of the socalst state n generatng socal nequalty n contemporary Chna.

4 Household Regstraton System and Educatonal Inequalty n Chna 3 INTRODUCTION Educaton assumes a key role n stratfcaton processes n all modern socetes. Comparatve studes reveal that the assocaton between socal orgn and educatonal attanment has remaned unchanged n almost all ndustralzed countres, despte of the expanson of educatonal opportuntes (Shavt and Blossfeld 1993; Treman and Yp 1989). In developng countres educatonal outcomes can also be explaned largely by famly backgrounds (Buchmann and Hannum 2001). Ths persstent nequalty s commonly attrbuted to the stablty of nsttutonal structures that lnk chldren of dfferent socal backgrounds to educatonal opportuntes n a socety (Kerckhoff 1995). Students from dfferent backgrounds may have more or less access to educatonal resources that ether drectly from famles or ndrectly from famly nfluences on school organzatons (Bdwell and Fredkn 1988; Coleman 1988). Among varous macro-level nsttutonal forces that ntervene n the process of educatonal stratfcaton, the communty and the state have receved partcular attenton. In countres wth a decentralzed educatonal system (e.g., the Unted States), schools are more exposed to communty nfluence, provdng greater leverage for hgh-status famles to affect students opportuntes for learnng and educatonal attanment (Kerckhoff 1995, p 330). In other countres wth a centralzed educatonal system, where school organzatons, the nature of currculum, and the pace of nstructon are more unform, the shft of state educatonal polces, such as adoptng compulsory educaton free of charge, could exert greater mpact on educatonal opportuntes for chldren from all socoeconomc backgrounds. The ncreasng avalablty of educatonal opportuntes ntated by the state, nevertheless, seems to have lmted consequences n detachng educatonal attanment from socal orgns n developng countres (e.g., Smth and Cheung 1986).

5 Household Regstraton System and Educatonal Inequalty n Chna 4 In state socalst socetes, one may expect a dfferent story, as the government had played a more vsble and drect role by adoptng an egaltaran admsson polcy to promote socal equalty (Parkn 1971; Smkus and Andorka 1982). Evdence from former state socalst countres, however, has lent lttle support to the thess of educatonal equalty. Substantal educatonal nequalty contnues to exst n Chna (Deng and Treman 1997; Hannum and Xe 1994; Zhou, Moen, and Tuma 1998), Czechoslovaka (Mateju 1993), Hungary (Smkus and Andorka 1982; Szeleny and Aschaffenburg 1993), Poland (Heyns and Bateck 1993), and the Sovet-era Russa (Gerber and Hout 1995). The effects of socal orgn on educatonal stratfcaton, after the ntal declne, ncreased n the late years of socalsm, reachng the level at least comparable to that n western socetes (Shavt and Blossfeld 1993). Why has the socalst state faled to produce educatonal equalty as ts egaltaran deology orgnally promsed? What role the state/government has played n the educatonal restratfcaton n the late perod of socalsm? The new class theorsts pont out that the socalst bureaucrats, once establshed ther prvleges, tend to take steps to secure the educatonal advantages of ther own chldren (Djlas 1957). Accordngly, the state educatonal polces would favor chldren from advantaged famles, and the socalst state was but a tool wth whch the new elte manpulated to mantan the prvleges for ther own famles (Mateju 1993; Smkus and Andorka 1982; S. Szeleny and Aschaffenburg 1993). Alternatvely, Gerber and Hout (1995) suggested that the state dd have produced more equal opportuntes by expandng general secondary educaton n the Sovet-era Russa, but hgher educaton faled to keep pace. The resultng enrollment squeezed at unversty level hurt the dsadvantaged classes partcularly, who ganed the most n the attanment of general secondary educaton. Therefore, despte the ntent to reduce nequaltes, the state educatonal polcy roncally had produced stratfyng pressure of ts

6 Household Regstraton System and Educatonal Inequalty n Chna 5 own. Educatonal stratfcaton was greater at the end of the Sovet perod than at the begnnng, as observed elsewhere. Hence, the role of socalst state n creatng educatonal nequalty has been portrayed contradctorly n the prevous lterature. On one hand, the socalst government, vewed as representng the nterests of prvleged classes, s drectly responsble for producng educatonal nequalty at the late perod of socalst regmes. On the other hand, the government s vewed as a consstent promoter of educatonal equalty, but ts effort has just faled due to the offset by some other factors. Further assessment of ths ssue s hndered manly by the followng factors: (1) the measure of the state mpact n emprcal models s not avalable n current lterature; and (2) the lnkages between the state polcy, communty/resdence, and famly backgrounds are partcularly lmted. Most scholars tend to rely on cohort/perod varatons n educatonal attanment patterns to extrapolate the mpact of government polcy shfts, rather than to conduct drect evaluatons of specfc government programs/socal polces (e.g., Deng and Treman 1997; Zhou, Moen, and Tuma 1998). Moreover, whle the ntertwnng effect of poltcs and resdental place has been acknowledged to some extent (Hannum 1999; Zhou, Moen, and Tuma 1998), how these contextual factors ntervene n the process of transferrng famly advantages to chldren s educatonal outcome are yet to be nvestgated. In ths paper, I examne the effect of the household regstraton system on educatonal nequalty n Chna. As one of ts procedures for soldfyng admnstratve control on populaton mgraton, the Chnese communst government had establshed the system snce 1955, whch s stll n place today. All households were regstered n the locale where they resded and also were categorzed as ether agrcultural or non-agrcultural - or as rural or urban - households (Chan and Zhang 1999). In the pre-reform era, resdental moblty, partcularly from vllages to ctes,

7 Household Regstraton System and Educatonal Inequalty n Chna 6 must be approved. The regstraton status (hukou) serves as an mportant bass of assgnng jobs, benefts, and other lfe chances (ncludng educatonal opportunty) to Chnese ctzens (Wu and Treman 2004, 2007). Because one s hukou status s tagged to the famly, but nsttuted by the state, and closely related to household resdence, studyng the mpact of hukou nsttuton offers a unque perspectve to shed lght on how the state, resdental place and famly backgrounds ntersect wth each other n producng educatonal nequalty n developng countres. HUKOU, FAMILY BACKGROUND & RESIDENCE IN EDUCATIONAL STRATIFICATION Perhaps the most promnent structural feature of contemporary Chnese socety s the socal dvde between the rural and urban sectors, nsttutonalzed by the household regstraton (hukou) system snce 1955, under whch all Chnese households had to be regstered n the locale where they resded and also were categorzed as ether agrcultural or non-agrcultural (synonymously, de jure rural or urban ) households (Chan and Zhang 1999, pp ). The hukou system has provded an mportant admnstratve tool for the government to cope wth demographc pressures n the course of rapd ndustralzaton startng n the 1950s (Chan 1994). Under the hukou system, the majorty of the populaton was confned to the countrysde and enttled to few of the rghts and benefts that the socalst state conferred on urban resdents, such as permanent employment, medcal nsurance, housng, pensons, and educatonal opportuntes for chldren. The hukou system served as an mportant mechansm n dstrbutng resources and determnng lfe chances (ncludng educaton) n Chna (Wu and Treman 2004). Permanently movng from a vllage nto a town or a cty (or ndeed, movng from one place to another at the same level) had to be approved by the government and undertaken through lmted channels, among whch hgh educaton s the most mportant crteron for selecton (Wu and Treman 2004).

8 Household Regstraton System and Educatonal Inequalty n Chna 7 Hukou status, lke other famly background characterstcs, can be vewed prmarly as an ascrptve attrbute, snce t s assgned at brth on the bass of the mother s regstraton status (Chan and Zhang 1999). Those whose mothers have urban status automatcally acqure urban status themselves, whle those whose mothers have rural status must compete for urban status. Regstraton (hukou) status need not be dentcal to resdental locale. People wth rural hukou status could and can lve n ctes, as have ncreasngly large numbers of mgrants begnnng n the early 1980s. Smlarly, people wth urban hukou status could lve n rural areas, as do agrcultural techncans and school teachers. Table 1 presents the percentage dstrbuton of natonal populaton by resdence type and hukou status n Chna from 1982 to In 1982, when the economc reform just started, over 92 percent (=74.8/81.0) of rural hukou holders resded n rural areas, whereas about 77 percent (=14.6/19.0) of urban hukou holders ndeed resded n ctes. As of 2005, the hukou system has lost much of ts effectveness n restrctng rural-urban mgratons: only 71 percent (=52.5/73.9) of rural hukou holders stll lved n vllages whereas the rest 29 percent (=21.4/73.9) resded n urban areas and were often referred as rural mgrants or floatng populaton. On the other hand, about 93 percent (=24.2/26.1) of urban hukou holders lved n ctes and towns n 2005, although the percentage they account for the de facto urban populaton has declned from 70 percent (=14.6/20.8) n 1982 to 53 percent (=24.2/45.6) n 2005 because of ncreasng rural mgraton n urban areas (Lang and Ma 2004). Despte the great ease n spatal mgraton, the hukou change from rural to urban status remans restrctve and selectve. As Fgure 1 shows, whle the de facto urban populaton has ncreased from 20.8 percent n 1982 to 45.6 percent n 2005, the de jure urban populaton (wth urban hukou) ncreased only slghtly, from 19.0 percent n 1982 to 26.1 percent n 2005.

9 Household Regstraton System and Educatonal Inequalty n Chna 8 Table 1. Rural-Urban Resdence and Hukou Status n Chna, Hukou Status (De Jure) Resdence (De Facto) Agrcultural Non-agrcultural Total 1982 Rural 74.8% 4.4% 79.2% Urban Total 81.0% 19.0% 100.0% 1990 Rural 64.1% 2.5% 66.2% Urban Total 79.5% 20.5% 100.0% 2000 Rural 60.2% 3.0% 63.2% Urban Total 75.2% 24.8% 100.0% 2005 Rural 52.5% 1.9% 54.4% Urban Total 73.9% 26.1% 100.0% Sources: Mcro-data of Chna populaton censuses 1982, 1990, 2000 and mn-census The hukou contnues to be used as the man crteron for socal excluson of rural de jure resdents, who have moved nto ctes but are dened of access to welfare, benefts and opportuntes enjoyed by local urban permanent resdents (Wang, Zuo and Ruan 2001; Yang and Guo 1996; Zhao 1999). Recent studes have documented the dscrmnatons faced by rural mgrants n ctes wthout local urban hukou (Solnger 1999), rural mgrant chldren s lmted access to educatonal opportuntes (Lang and Chen 2007), and the mpact of hukou on ncome nequalty (Lu 2005), educatonal attanment (Wu and Treman 2004), and occupatonal moblty (Wu and Treman 2007).

10 Household Regstraton System and Educatonal Inequalty n Chna 9 Fgure 1. Percentage of De Facto and De Jure Urban Resdents n Chna, year %de facto urban resdents % de jure urban resdents Nevertheless, the effect of hukou system on educatonal nequalty s confounded by three factors. Frstly, the hukou status s closed ted to famly orgns, measured by father s occupaton and parents educaton, whch are found to be strong predctors of educatonal attanment n almost all socetes beng studed (Shavt and Blossfeld 1993; Treman and Yp 1989). 1 Secondly, the hukou status s hghly correlated wth resdence place (see Table 1), and the rural-urban nequalty n educaton exsts n many developng countres (Buchmann and Hannum 2001). 1 In socal stratfcaton lterature, fathers have occuped a more central place than mothers for two reasons. Frst, because maternal and paternal status characterstcs are hghly correlated, t s often assumed that mother s status would be of lttle help n explanng addtonal varance n educatonal and occupatonal outcomes (also to be shown wth data later n ths paper). Second, because a sgnfcant proporton of mothers were not workng outsde home, socoeconomc dfferences among employed mothers were not beleved to be as consequental as socoeconomc dfferences among fathers, who are the bread earners of the famles (e.g., Roos 1985; Hout 1988; DPrete and Grusky 1990; Grusky and DPrete 1990).

11 Household Regstraton System and Educatonal Inequalty n Chna 10 Fnally, the converson of rural-to-urban hukou status has been very selectve, even to date (Lu 2003: ). Educaton, party membershp, and mltary experence are three major factors that facltate hukou moblty n Chna, among whch educaton s the most mportant (Wu and Treman 2004). Matrculaton n a specalzed secondary (zhong zhuan) or tertary (dazhuan or ben ke) school carred wth t enttlement to urban status, not counted n the government quota (State Councl [1958] 1986). Hence, junor hgh school graduates wth a rural hukou had (and have) two strateges for securng an urban hukou va hgher educaton. The frst was to gan admsson to a specalzed secondary school (zhong zhuan), whch conferred urban hukou status mmedately upon admsson. The second was to gan admsson to an academc senor hgh school and then to a tertary nsttuton. Tertary educaton confers both urban hukou status and a hgh-status job. Accordng to Wu and Treman (2004: 367), only 11% of all respondents from rural orgns had successfully converted ther hukou status, and hgher educatonal attanment accounted for about half of all hukou moblty. Because of selectvty of urban hukou based on educatonal qualfcaton n Chna (Wu and Treman 2004), t would be dffcult to dstngush the educatonal consequence of the hukou and the effect of educaton on (urban) hukou attanment based on the cross-sectonal data. Past research has always attempted to lnk changng educatonal stratfcaton to the poltcal process n Chna. For example, usng the 1982 urban census data, Deng and Treman (1997) reported that the educatonal attanment of Chnese men was hghly egaltaran wth respect to socal orgns, and that t had become ncreasngly so over tme. The state nterventon n equalzng educatonal opportuntes culmnated durng the Cultural Revoluton ( ), when the advantages of chldren from cadre and ntellgentsa famles were substantally reduced. Based on the analyss of survey data collected n 20 ctes, Zhou, Moen and Tuma (1998) demonstrated that the shftng state polces have had great mpact on varyng mechansms of educatonal stratfcaton n urban Chna from 1949 to 1994.

12 Household Regstraton System and Educatonal Inequalty n Chna 11 The state polces were not always conducve to educatonal equalty, but could also have led to educatonal nequalty. Under the state socalst regme, the centralzed state, wth the monopoly of almost all resources, was able to transfer resources among dfferent economc sectors, localtes, and socal groups on a much large scale and at a much faster pace than market economes could do (Zhou et. al.1998: 20; Wang 2008). Ths mples that the state has the capacty to create nequalty as well as equalty f needed for the runnng of the redstrbutve system. For example, the Chnese socalst redstrbutve economy desgnated a dstnctve pattern of spatal herarchy n resource allocaton, wth large ctes n the top and vllages n the bottom. Chldren n hgher-level ctes had better access to educatonal resources and lfe opportuntes, and enjoyed much more advantages n educatonal attanment than do ther counterparts n small ctes and towns (Zhou et. al. 1998). Hence, nequalty based on resdence place can be partally attrbuted to the state polces. Despte the clear evdence of state nterventon n educatonal stratfcaton wth respect to socal orgns and resdence n Chna, wthout the broader context of rural-urban dsparty, the role of the redstrbutve state n ths process s yet to be pnned down. Gven the mechansm of hukou moblty we have known (Wu and Treman 2007), any emprcal results based urban/nonfarm sample wth excluson of rural resdents (e.g. Deng and Treman 1997; Zhou et. al. 1998), are lkely subject to selecton bas. For nstance, the weak assocaton between socal orgn and educatonal attanment n urban Chna may be due to the government s successful effort to block the socal moblty of the rural majorty, rather than ts egaltaran polcy, and the temporal declne n the assocaton may reflect the tghtenng of the rural-urban segregaton system. Moreover, the educatonal advantages of those who resded n the largest ctes (Zhou et. al. 1998)

13 Household Regstraton System and Educatonal Inequalty n Chna 12 could have resulted from a sortng process,.e., hgher educated people are more lkely to permanently move and resde n large ctes. In short, wthout reference to the hukou system, the effect of famly background on educatonal attanment s lkely to be under-estmated among current urban resdents, whereas the effect of resdence locaton s lkely to be over-estmated among current rural and urban resdents. Scholars have also pad much attenton to the rural-urban dspartes n educatonal attanment n Chna. For example, Hannum (1999) examned the rural-urban gap n basc educaton n the context of poltcal change n Chna from ; Tsu (1997) nvestgated the mpact of economc reform on rural-urban educatonal nequaltes. In these studes, researchers have not explctly made dstncton between the effects of resdental locaton and of hukou enttlements. Table 2 presents the educatonal gap between rural and urban hukou holders for those of age between 25 and 29 years old, who supposedly have completed educaton. The dfference n the mean years of schoolng between the two groups seems to have been enlarged from 1982 to Table 2. Educatonal Gap between Rural and Urban Hukou Holders Age (25-29) Urban Rural Urban Rural Urban Rural Urban Rural Schoolng* % Senor H.S % College * Educaton s coded as years of schoolng: prmary=6, junor hgh school=9, senor hgh school =12, specalzed hgh school=14, specalzed college=15, unversty or above=16. Sources: Mcro-data of Chna populaton censuses 1982, 1990, 2000 and mn-census 2005.

14 Household Regstraton System and Educatonal Inequalty n Chna 13 Whle Chna s redstrbutve polcy may favor bg ctes over small ctes, urban areas over rural areas (Zhou et. al. 1998), the effects of resdental locaton on educatonal opportuntes cannot be solely attrbuted to the state force, snce the hstorcal and natural factors may also have contrbuted to the dspartes. Wth an approach mxng spatal factors wth nsttutonal factors, whether rural-urban educatonal nequalty s attrbutable to the government polces stll unanswered, as geographc nequalty exsts n any socety, partcularly n developng countres. Because of the great varance n the basc materal nputs such as textbooks, lbrares, and teacher tranng between rural and urban areas n less developed or developng) countres, such school effects have found to affect academc achevement, ndependently of famly background (Buchmann and Hannum 2001; Fuller and Clarke 1994; Heyneman and Loxley 1983). In ths paper, wth the data from a natonal probablty survey conducted n 2005, I nvestgate the effect of the household regstraton status on educatonal attanment for Chnese adults. Rather than focusng on the ftful shfts of government polces (extrapolated from the temporal varablty), I examne how ths nsttuton, whch has been relatvely stable snce ts nstallaton n the 1950s, ntervenes n the process of educatonal stratfcaton. To assess the causal relatonshp between hukou status and educatonal attanments, I re-conceptualze the substantve problems wth explct counterfactuals n causal nference. I treat the hukou system as a specfc polcy program n Chna s socalst experment and adopt a program evaluaton approach to assess ts effect on educatonal nequaltes between those who were assgned rural or urban hukou status.

15 Household Regstraton System and Educatonal Inequalty n Chna 14 DATA, VARIABLE MEASURES, AND MODELS Data The Chna General Socal Survey (2005) s the annual survey of the natonal representatve sample of adult populaton aged 18 or above n both rural and urban Chna except for Tbet. A mult-stage stratfed random samplng method s used n the survey. Frst, 125 prncpal samplng unts (PSU) were selected from 2,798 county or county-level dstrcts, stratfed by regon, rural and urban populaton and educaton levels; then 4 second-level samplng unts (SSU) are selected n each selected PSU; and 2 thrd-level samplng unts (TSU) were selected n each selected SSU; and Fnally, 10 households are selected n each selected TSU. One elgble person aged 18 or above was randomly selected from each sampled household to serve as the survey respondent. There are 10,372 completed ntervews n the sample, ncludng 6,098 from urban and 4,274 from rural areas. Data are weghted to represent the general populaton n Chna (Ban and L 2005). In ths paper, I restrct the sample to those born between 1937 and Varable and Measures The survey questonnare contans detal nformaton on respondents educaton, resdence place, and ther famly backgrounds. Respondents educaton nformaton ncludes the hghest educaton completed and hghest educaton attended. Unlke most other surveys conducted n Chna, educaton s measured wth great detals: from no schoolng, grade 1 to 6 of prmary school, grade 1-3 n junor hgh school, grade 1-3 n senor hgh school, vocatonal school/techncal school, specalzed secondary school, part-tme and full-tme 3-year college, part-tme and full-tme 4-year unversty, and graduate school educaton. Ths measure not only renders the codng of educaton nto the years of educaton wth greater precson, but also the

16 Household Regstraton System and Educatonal Inequalty n Chna 15 analyze school transton process n a dsaggregate way (e.g., to academc vs. vocatonal track n senor hgh school). To nvestgate the effect of hukou status on overall educatonal attanment, I use years of schoolng completed as the dependent varable n OLS regresson models. To examne how the hukou s effect changes n dfferent stages of school transton, I construct a set of logstc models on educaton progresson based on school attended and school completed (Mare 1980). Fnally, wth endogenous swtchng regresson models, I take nto account of the selecton effect on the years of schoolng completed between rural and urban de jure resdents. The ndependent varables nclude respondent s hukou status, resdence type, father s and mother s occupaton status when the respondent was at age 14, father and mother s educaton, as well as gender and cohort. Respondent s current hukou s the varable of central nterest n ths paper, coded n a dummy varable (rural hukou=1). There s no hstorcal nformaton on the respondent s hukou orgn or hukou change. 2 Related to the hukou status s the respondent s sze of resdence, whch s coded nto a 6- level spatal herarchy: 1=vllage, 2=townshp, 3=county-level cty, 4=prefecture cty, 5=provncal captal, 6=drect admnstratve muncpalty (Bejng, Shangha, Tanjn and Chongqng). They are converted nto a set of dummy varables n the regresson analyss. As an exploratory analyss, I employ both parents occupatonal status and educaton to measures of famly socoeconomc background. Father s and mother s occupaton when the respondent was at age 14 s an ndcator of famly socoeconomc status. Whle prevous research has shown that mothers occupaton has lttle mpact on chldren s educatonal attanment for the reasons lad out n footnote 2, the stuaton mght be dfferent n Chna gven the hgh rate of 2 Whle the Chnese General Socal Survey (CGSS 2003) dd have collected the hukou nformaton when the respondents were at age 14, t covered only those resdng n urban areas at the tme of the survey and therefore napproprate for the ssue to be addressed here.

17 Household Regstraton System and Educatonal Inequalty n Chna 16 female labor partcpaton under the state socalst regme. They are both measured ISEI (Internatonal Socoeconomc Index of Occupatons), whch s a scale of occupatonal status, rangng n prncple from 0 to 100 (Ganzeboom, De Graaf, and Treman 1992). The Chnese Standard Classfcaton of Occupatons, used to code the occupaton data n the survey, closely matches the 1988 Internatonal Standard Classfcaton of Occupatons, so 1988-bass ISEI scores were assgned to the data. Both father s years of schoolng and mother s years of schoolng are ncluded n the model as contnuous varables. Gender s also coded as a dummy, wth men as 1 and female as 0. Fnally, to nvestgate the temporal varatons, the sample s coded nto 5 brth cohorts: , , , , and In the followng analyss, the second brth cohort ( ) s used as the benchmark because ths s the frst cohort whose schoolng experence has been fully exposed to the mpact of the hukou system, whch was establshed snce Methods and Models To assess the net effect of hukou status on educatonal attanment, I employ an OLS regresson model of the followng form: Y = β k 0 + β1h + β k X k + ε k = 2 (1) where subscrpts ndvduals, Y, denotes year of schoolng completed, and H ndcates current hukou status (1=urban, 2=blue-stamped, 3-rural). X k s a vector of other ndependent varables, β k s the estmated coeffcent of the respectve ndependent varable, and ε s the dsturbance term.

18 Household Regstraton System and Educatonal Inequalty n Chna 17 For school transton models, I estmate three bnary logt models p log 1 k = β + β H + β X (2) 0 1 p = 2 k k where p s the probablty of transton to junor hgh school and academc senor hgh school, gven the completon of prevous level of educaton. The notatons of the ndependent varables are the same as those n equaton (1). Because admssons to vocatonal school and college normally nvolved hukou changes, we do not model the mpact of current hukou status on the transtons to vocatonal school and to college (Wu and Treman 2004). The fundamental queston n ths paper s to assess the causal effect of hukou on educatonal attanment n Chna. Equaton (1) exemplfes a conventonal approach to assess the effect of socal poston/status (hukou) on the outcome varables (educatonal achevement), whch assumes that, once the other measured varables that affect the dependent varable are taken nto account, the sortng process of ndvduals nto the poston s ndependent of factors nfluencng the dependent varable (hukou attanment) per se. In other words, no other unmeasured varables could affect both the dependent varables and the lkelhood of obtanng hukou status. Such assumptons are often too strong to be tenable, as ndvduals may obtan the urban hukou status based on ther own acheved educaton (Wu and Treman 2004). In recent years when there are alternatve channels other than educaton and career advancement for gettng urban hukou status for those born n rural background (e.g., buyng an apartment n the cty to obtan an urban hukou), parents may manage to use ther economc and socal resources to get an urban hukou for ther chldren n ctes so as to afford them more educatonal opportuntes. On the other hand, famles stayng as rural status may dscourage ther chldren from attendng schools because of parents falure experence n socoeconomc moblty va the channel of

19 Household Regstraton System and Educatonal Inequalty n Chna 18 educaton. In ths regards, the assgnment of hukou s endogenous to hs/her educatonal attanment. 3 The educatonal outcome and hukou status may be smultaneously determned, leadng to a selecton bas even though the measurable ndependent varables are controlled for. Endogenous swtchng regresson models, whch smultaneously model the sortng of ndvduals nto dscrete postons and the effects of poston and other varables on outcomes, are sutable for assessng the effects of a socal classfcaton on a consequence of membershp n ths classfcaton (Wnshp and Mare 1992:338). They are especally useful n the stuaton here, under whch unmeasured determnants of hukou status are also lkely to affect educatonal outcomes. The models contan three equatons: an educatonal attanment equaton for people of rural and urban hukou regstraton status, respectvely, and a probt equaton predctng hukou status: 4 K Y1 = β 1k X k + ε1 k = 0 (3) K Y2 = β 2k X k + ε 2 k = 0 (4) where Y 1 and Y 2 denote educatonal outcomes for people of rural hukou status and people of urban status (postons 1 and 2), β 1k and β 2k are estmated coeffcents, and ε 1 and ε 2 are dsturbance terms for each group separately. 3 There are a number of such examples n the socal world: school admnstrators or parents may assgn students to dfferent tracks accordng to ther academc potentals (Gamoran and Mare 1989). Indvdual may choose to attend college on the bass of the perceved beneft relatve to that of otherwse (Wlls and Rosen 1979). Welfare recpents may self-select themselves nto socal programs because of the benefts of program partcpaton. 4 The notatons here are adopted from Mandela (1983), Wnshp and Mare (1988:135-46), and Gerber (2000).

20 Household Regstraton System and Educatonal Inequalty n Chna 19 In addton, a model s specfed for a latent score Z that ndexes the th ndvdual s lkelhood of assgnment to urban hukou status: Z K = γ k X k + η1y 1+ η 2Y2 + ξ k = 0 (5) where η 1, η 2 and γ k are parameter estmates and ζ s the dsturbance term. The dependent varables n (3) and (4) are ncluded n the selecton model because educatonal achevement may enter nto ther competton for urban hukou status. Several measures of famly background are used as the predcators for respondent s current hukou status: father s socoeconomc status (ISEI), father s educaton, educaton, whch are measured the same as descrbed n Equatons (1) and (2). Current resdence, gender, and cohort that affect educatonal attanment are also used to predct the current hukou status educatonal outcomes. In addton, I nclude mother s occupatonal status (ISEI) as a factor that may determne respondent s hukou status but not ther educatonal achevement, because one s hukou status usually follows mother s hukou status, whch s often assocated wth ther occupaton. The terms Y 1 and Y 2 n Equaton (5) can be substtuted by equatons (3) and (4) and a reduced-form verson of the model can be obtaned, K Z = π X + (6) k = 1 where π k =η 1 β 1k +η 2 β 2k +γ k and ε 3 =η 1 ε 1 +η 2 ε 2 + ζ. Whereas the dsturbance for the structural verson of the probt equaton (.e., ζ ) s un-correlated wth the dsturbances for educatonal k k ε 3 determnaton equatons ε 1 and ε 2, n the reduce-form verson, ε 3 s n general correlated wth ε 1 and ε 2. The two covarances σ 13 and σ 23 ndcate common but unobserved factors that affect both

21 Household Regstraton System and Educatonal Inequalty n Chna 20 hukou assgnment and educatonal determnaton. When σ 13 and σ 23 are non-zero, or the correlaton coeffcents ρ 13 and ρ 23 between error terms are non-zero, then ANACOVA estmaton of coeffcents n Equaton (1) wll be based. If σ 13 (or ρ 13 )<0, t suggests a negatve selecton nto rural hukou status, namely, people of rural status could do better than they actually do wthout selectvty; If σ 13 (or ρ 13 )>0, t suggests a postve selecton nto rural hukou status, namely, people of rural status mght do worse than they would do wthout selectvty. Smlarly, σ 23 (or ρ 23 )<0, t suggests a postve selecton nto urban hukou status, namely, people of urban hukou status could actually do better than they would have done wthout selectvty; If σ 23 (or ρ 23 )>0, t suggests a negatve selecton nto urban hukou status, namely, those of urban hukou status mght do worse than they would have done wthout selectvty. The three equatons are estmated smultaneously by maxmum lkelhood assumng that the error terms for the three equatons may be correlated and follow a trvarate normal dstrbuton. 5 Data are weghted to represent the Chnese general populaton and the clusterng effect on prncpal samplng unts county) s also taken nto account. DESCRIPTIVE STATISTICS Table 3 presents summary statstcs for varables ncluded n the analyss, for the entre sample and for the sample wth urban hukou status and rural hukou status, separately. Among 7799 cases wth completed nformaton, 3196 held urban hukou status and 4603 held rural hukou status. The sample statstcs are weghted to represent the hukou composton of the natonal populaton. As the table shows, the average schoolng for ths sample s 7.7 years, 41.7 percent have prmary school educaton, 31.8 percent have junor hgh school educaton, 18.9 percent 5 I use movestay- command n STATA to estmate the model.

22 Household Regstraton System and Educatonal Inequalty n Chna 21 have senor hgh school educaton, and only 7.6 percent have tertary educaton. The ratos of transton to junor hgh school gven the completon of prmary school educaton, to academc senor hgh school gven the completon of junor hgh school educaton, are 86 percent and 52.2 percent, respectvely. Table 3. Descrptve Statstcs for Selected Varables, 2005 Separated by Hukou Status Varables Total Urban hukou Rural hukou Years of schoolng (4.434) (4.013) (3.820) Educatonal level Prmary S Junor hgh school Senor hgh school College or above School transton rato to junor hgh school to academc hgh school to vocatonal school Sex (male=1) Brth cohort Father s ISEI at age (11.88) (14.74) (7.948) Father s schoolng (4.457) (4.876) (3.777) Mother s ISEI at age (7.937) (11.28) (2.249) Mother s schoolng (3.853) (4.497) (2.892) Resdence Vllage Townshp County-level cty Prefecture-level cty Provncal captal Drect admn muncpalty N (un-weght) Notes: Data are weghted. Fgures n the parentheses are standard devaton for contnuous varables.

23 Household Regstraton System and Educatonal Inequalty n Chna 22 To contrast the observed characterstcs between people of dfferent hukou status, I calculate the descrptve statstcs separated by hukou status. The educatonal nequalty between people of urban hukou status and people of rural hukou status are notably large. The average schoolng s 10.3 years for those who hold urban hukou status (de jure urban resdents), and about 6 years for those who hold rural hukou status, wth a gap of over 4 years that s consstent wth the natonal statstcs n Table 2 (though only for a specfc cohort). In terms of school transton ratos, 93 percent of urban hukou holders and 78 percent of rural hukou holders can successfully proceed to junor hgh school educaton after completng prmary school educaton; 67 percent of urban hukou holders and 29.9 percent of rural hukou holders can successfully proceed to academc senor hgh school after completng junor hgh school. To show the temporal trends n such a gap, Fgure 2 plots the mean years of schoolng for the two groups by brth cohorts from 1937 to Wth the educatonal expanson, the overall educatonal attanment has ncreased for both rural and urban resdents, but rural-urban gap remaned roughly constant (wth a dfference of about 4 years). Based on the gap n average years of schoolng, one may not dscern the dfferentals n school transton rates at dfferent levels of educaton. In Fgures 3 and 4, I plot the ratos of transton to junor hgh and academc senor hgh school by brth cohorts and hukou status. The rural-urban gap n transton to junor hgh school has been narrowed for younger cohorts, suggestng the mplementaton of the 9 years of compulsory educaton n the 1980s and 1990s have benefted the rural people when the rato n urban areas has already reached saturaton. However, beyond the compulsory educaton, the gap n the lkelhood of transton to senor hgh school seems to be enlargng for the younger cohorts. Analyss of the census data has reported smlar fndngs (Wu 2010).

24 Household Regstraton System and Educatonal Inequalty n Chna 23 Fgure 2. Trends n Educatonal Attanment between De Jure Urban and Rural Resdents, Chna, brth year rural hukou urban hukou Fgure 3. Trends n Rato of Transton to Junor Hgh School between De Jure Urban and Rural Resdents, Chna, 2005, brth year rural hukou urban hukou

25 Household Regstraton System and Educatonal Inequalty n Chna 24 Fgure 4. Trends n Rato of Transton to Senor Hgh School between De Jure Urban and Rural Resdents, Chna, brth year rural hukou urban hukou As ponted out earler, hukou status s also correlated wth other famly background characterstcs. The rural-urban gap n father s socoeconomc status (ISEI) s 7.86 ponts ( ), n father s average schoolng s 2.67 years ( ), and n mother s average schoolng s 2.59 years ( ). The spatal segregaton by hukou status s also evdent. Among people who held rural hukou status, 90.7 percent lve n vllages, 4 percents lve n townshps and county seats, and only about 5 percent lve n ctes of county-level or above. On the other hand, among those of permanent urban hukou status, only 6.9 percent resde n vllages. In the followng, I wll frst employ OLS regresson models to examne the effect of hukou status on educatonal attanment, and then use bnary logt models to nvestgate the determnaton of transton rato to junor hgh school and academc senor hgh school. Fnally, to assess the causal effect of hukou status, I use the swtchng regresson models to take nto account of the selectvty of hukou assgnment based on observable and unobservable characterstcs.

26 Household Regstraton System and Educatonal Inequalty n Chna 25 REGRESSIONS RESULTS Model 1 n Table 4 reports the ordnary least squares (OLS) regresson estmates for models of years of schoolng completed, wth hukou status as a dummy varable, resdence, father s occupaton ISEI, and father s educaton, mother s educaton, gender, and brth cohorts as controlled covarates. Table 4. Coeffcents for Models of Educatonal Attanment and School Transton, Chnese Urban and Rural De Jure Resdents, 2005 Model 1 Model 2 Model 3 Years of schoolng To junor hgh school To Academc Senor Hgh. school Rural hukou *** (0.197) *** (0.148) Resdence [vllage omtted] Townshp 0.747** (0.235) (0.234) County-level cty 1.370*** 1.337** (0.281) (0.429) Prefecture cty 1.106*** (0.304) (0.276) Provncal captal 1.350*** 0.766** (0.283) (0.264) Drect ad muncpalty 2.042*** 0.853** (0.286) (0.316) Father ISEI 0.020*** 0.037* (0.004) (0.014) Father schoolng 0.183*** 0.060*** (0.019) (0.015) Mother ISEI (0.004) (0.006) Mother schoolng 0.125*** 0.066** (0.017) (0.023) Gender (male=1) 1.684*** 0.559*** (0.100) (0.110) Cohort:[ omtted] *** *** (0.191) (0.156) *** 1.461*** (0.148) (0.147) *** 1.202*** (0.171) (0.176) *** 1.528*** (0.211) (0.204) Constant 5.068*** (0.286) (0.385) (Pseudo) R N Notes: data are weghted; standard errors are adjusted for the clusterng on samplng unt. * p<.05 ** p<.01 *** p< *** (0.171) 0.391* (0.193) 0.415* (0.183) 0.513** (0.195) 0.785*** (0.221) 0.721*** (0.192) 0.007* (0.003) 0.031* (0.013) (0.004) 0.058*** (0.013) (0.081) * (0.220) 0.497** (0.177) * (0.173) (0.170) *** (0.239)

27 Household Regstraton System and Educatonal Inequalty n Chna 26 As observed n other socetes, famly background plays an mportant role n respondent s educatonal attanment. Father s occupaton status and educaton and mother educaton have sgnfcant effect on respondents years of schoolng completed. An ncrease n father s ISEI by 10 pont n scale brngs about 0.2 years ncrease n chldren s schoolng; an addtonal year of father s educaton and mothers educaton brngs about 0.18 years and 0.13 years ncrease n chldren s educaton respectvely, holdng constant of the other factors. Confrmng wth the prevous lterature n socal stratfcaton (also see footnote 2), 6 we found that mother s occupatonal status, measured by ISEI, has no sgnfcant effect on chldren s schoolng and the lkelhood of school transtons. The effect of respondent s resdence locaton n the Chnese spatal herarchy on educatonal attanment s promnent. Roughly, those who resde n the central-governmentadmnstratve muncpaltes (e.g. Bejng, Shangha) on average receved 2.04 years more schoolng than those who resde n vllages, other thngs beng equal. The net advantage vares from 0.7 to 1.4 years for those who lve n small towns, county-level ctes, prefecture level ctes, and provncal captals (p<.001). 6 In contrast to the typcal status attanment study, psychologsts, economsts, and educatonal socologsts tend to focus on young chldren rather than on adults and examne dfferences n ntellectual ablty and scholastc accomplshment rather than dfferences n completed educaton or occupatonal attanment (Schachter 1981; Blau and Grossberg 1992; Dronkers 1989). Intally, these studes suggested that a reducton n the tme mothers were spendng at home would have negatve effects on ther chldren s early achevements. In response to ncreasng female labor force partcpaton, whle some economc studes showed that workng mothers fnancal resources contrbute to chld outcomes (Desa, Chase-Lansdale, and Mchael 1989), socologsts have not found frm evdence to support the clams that occupatonal status among employed mothers affects the educatonal attanment of adults when they grew up (typcally at age 14) (Doornk and J. Dronkers 1996). In ther classc comparson of the status attanment processes of men and women, Treman and Terrell (1975:177) reported n a footnote that among adult women who grew up n dual-earner households, mother's occupatonal status has nsgnfcant effect on hghest grade of schoolng completed, net of the nfluence of parental educaton and father's occupaton. The evdence from Chna seems to be consstent wth the conventonal wsdom n socal stratfcaton lterature.

28 Household Regstraton System and Educatonal Inequalty n Chna 27 There s also a sgnfcant gender gap n educatonal attanment. Men on average receve 1.7 years more educaton than do women, and educatonal expanson has certanly benefted those n the younger cohort. For example, compared to those born between (about the tme when the hukou system was establshed and tghtened), people born n the next two successve cohorts have receved about 1.6 years more educaton, and those born between 1977 and 1985 have enjoyed an even greater advantage by 2.6 years. Even after controllng the effects of all these factors, whch are hghly correlated wth hukou status, a sgnfcant rural-urban gap can stll been seen. People of rural status receve 2.3 years of schoolng less than ther counterparts of urban hukou status, and the dfference s statstcally sgnfcant (p<.001). Model 2 presents estmated coeffcents for bnary logt models of transton to junor hgh school gven the completon of prmary school educaton, to academc senor hgh school gven the completon of junor hgh school. Confrmng the general patterns of educatonal stratfcaton found n other countres (Shavt and Blossfeld 1993), results n Models 2 and 3 show the logt effects of socal orgns (famly background) on educatonal transtons tend to be stronger at the begnnng of an educatonal career and declne n subsequent transtons, namely, socal selecton s most pronounced n the early stage of educatonal transtons (also see Mare 1980). Ths holds true for resdence locaton, father occupaton ISEI, mother s educaton, gender and cohort, and father s educaton to some extent. Almost all ndependent varables (except for mother s ISEI and townshp resdence) are sgnfcant predctors of the lkelhood of transton to junor hgh school after completng prmary school educaton, but only a few varables have sgnfcant effects on the lkelhood of transton to academc hgh school.

29 Household Regstraton System and Educatonal Inequalty n Chna 28 Take the effects of three famly background characterstcs as examples. Father s ISEI, educaton, and mother s educaton seem to have more mpact n earler stages of school transtons. One addtonal pont n father s ISEI wll ncrease the odds of transton to junor hgh school by 3.77 percent (=e ), to academc senor hgh school by 0.7 percent (=e 10* ) (p<.05), One-year ncrease n father s educaton brngs about 6.2 percent (=e ) ncrease n the odds of transton to junor hgh school, and 3.1 percent ncrease n the odds of transton to academc senor hgh school. One-year ncrease n mother s educaton brngs about 6.8 percent (=e ) ncrease n the odds of transton to junor hgh school, and 6.0 percent (=e ) ncrease n the odds of transton to academc senor hgh school. Even after controllng for other varables, the effect of hukou perssts. The rural-urban odds rato of transton to junor hgh school and to academc senor hgh school are 0.4 (=e ) and 0.53 (=exp ), respectvely. Scholars have provded some explanaton to the wanng effect of famly background on dfferent stages of school transtons. As young pupls get older and advance to hgher levels, they are ncreasngly able to decde on what they want and famly (parental) resources become less mportant. In many countres, chldren from dsadvantaged backgrounds encounter very severe selecton barrers n early stages of transtons: only the brghtest chldren from those backgrounds can make t to the hgher levels, whereas chldren from advantaged famles progress nto secondary and tertary schools wth great ease. Consequently, school success s less and less affected by famly socoeconomc background (Mare 1980; also see a summary n Blossfeld and Shavt 1993). Ths pattern also holds true wth regard to the effect of hukou status. Results from regresson analyses above should be treated cautously here, as educaton (not only tertary, as t always s) may be used as the crteron for the converson of hukou from rural to urban status (Wu and Treman 2004). Therefore, current hukou status mght be a cause but also a result of educatonal attanment. A typcal the endogenety problem arses here. It s for ths reason that I now turn to the swtchng regresson analyss.

30 Household Regstraton System and Educatonal Inequalty n Chna 29 RESULTS FROM ENDOGENOUS SWITCHING REGRESSION MODELS Because the effect of hukou status on educatonal outcomes could result from the selecton process, to establsh the causal lnk between hukou status and educatonal nequalty, I use endogenous swtchng regresson models to take nto account of the selecton bas and examne the causal effect of hukou status on the years of schoolng completed. As specfed n the precedng secton, the selecton equaton s a probt equaton (Eq. 5), but the functon outcome equatons (Eq 3. and Eq. 4) are OLS regresson models. For smplcty, I only present estmated results from the general model wthout mposng any restrctons. If σ 13 = σ 23 =0, then t suggests that no selectvty exsts at all and the ascrpton model can ft the data well. OLS regresson results wll not be based. Otherwse, the hukou system may serve as an mportant agent n generatng rural-urban nequalty n educaton. Table 5 presents the estmated coeffcents. Model 4 n Table 5 s the selecton model predctng the assgnment of urban hukou status to the respondents. Agan, I nclude all ndependent varables prevously mentoned, and mother s occupaton ISEI status n the equaton. Except for father s educaton, all other varables ncludng resdence, father s occupaton ISEI, mother s ISEI and educaton, gender and cohort are sgnfcant predctors of whether the respondent has rural hukou status. Ths fndng also verfes that hukou status s hghly ted to resdence and famly s socoeconomc background. It s worth partcularly mentonng that mother s ISEI sgnfcantly affects the lkelhood of havng rural hukou status (p<.001). The hgher ther mothers occupatonal ISEI, the less lkely the chldren hold rural hukou status. Model 5 and Model 6 present the estmated coeffcents for OLS regresson on educatonal attanment for rural hukou and urban hukou, respectvely.

31 Household Regstraton System and Educatonal Inequalty n Chna 30 Table 5. Parameters for Endogenous Swtchng Regresson Model on Educatonal Attanment n Urban and Rural Chna, 2005 (N=7712) Model 6 Model 6a Model 6b Varables Probt model for rural hukou assgnment Resdence [vllage omtted] Townshp *** (0.171) Educatonal attanment: rural hukou Educatonal attanment: urban hukou 2.466*** (0.662) (0.369) County-level cty 2.823*** (0.181) 3.207*** (0.897) (0.372) Prefecture cty 2.910*** (0.166) 2.997*** (0.787) (0.382) Provncal captal 2.564*** (0.247) 3.849*** (0.679) (0.407) Drect admn. muncpalty 3.252*** (0.169) 4.096*** (1.057) 1.152*** (0.360) Father ISEI when R was *** (0.003) 0.022*** (0.007) 0.024*** (0.005) Father schoolng (0.009) 0.200*** (0.028) 0.162*** (0.020) Mother schoolng *** (0.012) 0.169*** (0.028) 0.103*** (0.021) Gender (male=1) *** (0.051) 2.001*** (0.129) 1.273*** (0.142) Cohort : [ omtted] (0.069) *** (0.229) (0.355) (0.078) 1.538*** (0.195) 1.861*** (0.217) *** (0.089) 1.324*** (0.211) 1.804*** (0.252) ** (0.161) 2.266*** (0.283) 2.988*** (0.241) Mother s ISEI when R was *** - - (0.010) Constant 3.274*** (0.260) 2.704** (0.311) 5.984*** (0.462) ρ 13, ρ * (0.174) 0.103* (0.045) Wald test of ndep. equaton χ 2 (d.f) (1) Notes: Fgures n the parentheses are robust standard errors adjusted for clusterng effect on samplng unts. Data are weghted. * p<.01 ** p<.01 *** p<.001 We exclude mother s ISEI as the ndependent varable n the two equaton based on the emprcal fact that mother s ISEI affects chldren s hukou status but not educatonal attanment drectly (see footnote 2 and also Table 4). Contrastng the correspondng coeffcents between

32 Household Regstraton System and Educatonal Inequalty n Chna 31 Models 5 and 6 suggests that educatonal stratfcaton process dffers between those of dfferent hukou status. For nstance, the effect of resdence sze on educatonal attanment for people of urban hukou s much smaller than that for people of rural hukou. The effects of father s occupaton ISEI, father s and mother s educaton on educatonal attanment are all greater for rural hukou resdents than for urban hukou holders. The rural-urban educatonal gap to some extent can be attrbuted to ther dfferences n famly socoeconomc and cultural resources that can facltate educatonal attanment. Gender gap s also greater among rural hukou holders than among urban hukou holders. What s the role of hukou n educatonal attanment n Chna? As shown n the bottom row of Table 5, ρ 13 <0, suggestng a sgnfcantly negatve effect of selecton nto rural hukou status. Ths s to say, those who stay as rural status are punshed to stay, as the brghtest has been able to move out of rural status. People of rural hukou status would do better f there were a random assgnment of rural hukou status. On the other hand, the sgnfcant and postve ρ 23 also suggests a negatve selecton nto urban hukou status, n other words, wthout selectvty, current urban resdents would do better than they actually have acheved now. To elaborate the hukou effect more fully, I calculate the expected years of schoolng that respondents would attan under alternatve hukou assgnment. I use the parameter estmates for rural hukou and urban hukou reported n Table 5 and apply equatons (3)-(5) for a hypothetcal ndvdual who has the average value of the total sample on the selected ndependent varables and a probablty of assgnment to rural hukou status equal to the sample proporton of ndvduals wth rural hukou status. Table 6 summarzes the calculatons. The column of table compare hukou tracks/regmes and the rows denote alternate groups of people: people actually holdng rural hukou, people actually holdng urban hukou, and all people combned.

33 Household Regstraton System and Educatonal Inequalty n Chna 32 Table 6. Summary of Hypothetcal Effects of Assgnment to Alternatve Hukou Status on Educatonal Attanment Track Populaton: Rural hukou Urban hukou Rural hukou Urban hukou All Notes: effects are estmated holdng constant of other varables at total sample means and assumng a probablty of assgnment to rural hukou s The last row of Table 6 provdes estmated years of schoolng completed that would be observed f people were randomly assgned to hukou status, namely, these estmates are free of selecton bases and the dfference between hukou types can be attrbuted to the structural effect of hukou per se. The rural rural hukou and urban-urban hukou cells denote the scenaros that people have actually experenced. The rural urban hukou urban-rural hukou cells denote hypothetcal years of educaton that the respondents would have were they assgned to a dfferent hukou from the one they actually had. There s clear evdence ndcatng a causal effect of hukou on educatonal attanment: urban hukou produces hgher years of schoolng than rural hukou for the populaton as a whole and, hypothetcally, for the rural majorty who were assgned urban status. For the total populaton, the advantage of urban hukou n years of schoolng s =1.66 years, an estmate of the dfference n educatonal achevement that would be observed were people wth the same characterstcs randomly assgned to dfferent hukou status. If the current rural hukou holders, who have 7.17 average years of schoolng, were assgned to urban hukou status, they would have attaned 9.96 years of schoolng on average, wth a dfference of 2.79 years. The

34 Household Regstraton System and Educatonal Inequalty n Chna 33 dfference can be nterpreted as the cost of rural hukou on educatonal opportunty for rural resdents. On the other hand, f current urban hukou resdents were assgned to rural hukou status, they would do even slghtly better by 0.07 years. In other words, the hukou stratfcaton has hurt rural people wthout necessarly beneftng urban people. The effect of hukou (1.66) s smaller than the actual contrast between urban and rural hukou resdents ( =2.18). Relatve to a system n whch people are randomly assgned to dfferent hukou tracks, the actual system has produced larger dfferentals n years of schoolng completed between people of dfferent hukou status. Ths s manly because rural people are negatvely selected nto rural hukou status, but the selectvty s neglgble for the urban hukou track. For the same reason, the exstng hukou trackng also slghtly decreases the populaton average years of schoolng by about 0.7 years. 7 In summary, the above llustratons ndcate the exstence of a causal effect of hukou status on educatonal attanment for the rural majorty. The exstng hukou system enlarges the educatonal gap between those who actually hold rural and urban hukou status and produces lower educatonal achevement than under random assgnment. People of rural hukou status are partcularly hurt by the system of hukou dstncton. They could ncrease ther educatonal attanment substantally f more of them were assgned to urban status nstead. SUMMARY AND CONCLUSIONS Ths paper examned the role of the household regstraton system (hukou) on educatonal stratfcaton n rural and urban Chna. Wth the data from a natonal probablty sample survey conducted n 2005, I found that the regstraton (hukou) status, desgnated by the state polcy, affects educatonal attanment and the ratos of school transtons. People of rural hukou status 7 Assumng the proporton of people n urban hukou status s fxed at 0.404, the actual achevement s 0.404* *7.17=8.05, and under random assgnment t s 0.404* *8.08=8.75.

35 Household Regstraton System and Educatonal Inequalty n Chna 34 are substantally dsadvantaged n years of schoolng attaned, and such dsparty has been consstent over tme. One s hukou status also plays an mportant role n determnng the lkelhood of transtons to junor hgh school and academc senor hgh schools. One should be cautous to establsh the causal lnk between hukou status and educatonal attanment because educaton can also be used as the crteron for urban hukou assgnment for those orgnally from rural backgrounds (Wu and Treman 2004). Wth swtchng regresson models that account for selectvty based on measured and unmeasured characterstcs, the causal effect of hukou on educatonal attanment can stll be dentfed, albet wth small magntude. Although hukou status s hghly correlated wth other famly background characterstcs, such as father s occupatonal ISEI, educaton, resdental place, t taps addtonal and sgnfcant dsadvantages n educatonal attanment for those of rural orgn. Moreover, the effects of famly background on educaton vary by hukou status, wth the effects greater among people of rural hukou status. Thus, the hukou system than dstngushes two classes of ctzens seems to strengthen the nequalty based on famly backgrounds. Thus scholars ought to pay more attenton to the unque socalst nsttuton and socal categores defned by hukou status n studyng socal stratfcaton and moblty n contemporary Chna. The Chnese hukou system provdes a unque case to examne how the socalst state constructed opportuntes for certan socal groups and blocked moblty of the others (rural majorty). Such a strong polcy nterventon n stratfcaton has rarely been seen n human hstory (probably the only comparable case s South Afrca under aparthed regn. see Sedman [1999]). The fndngs n ths paper challenge the conventonal percepton of socalst egaltaran state and shed new lghts on the role of the state n generatng nequalty, callng for a new theorzaton of the stratfcaton process under the state socalst regme.

36 Household Regstraton System and Educatonal Inequalty n Chna 35 REFERENCES Ban,Yanje and L Lulu Chna s General Socal Surveys 2003 (CGSS2003): A Methodologcal Report. HKUST Survey Research Center. Bdwell C.E., and N. E. Fredkn The Socology of Educaton Pp n The Handbook of Socology, edted by Nel Smelser. Newbury Park, CA: Sage Blau, Peter M, and Danqng Ruan Inequalty of Opportunty n Urban Chna and Amerca. n Research n Stratfcaton and Moblty Vol. 8: 3-32, edted by Arne.L. Kalleberg. Greenwch: JAI Press. Blau, Francne D. and Adam J. Grossberg Maternal Labor Supply and Chldren's Cogntve Development. Revew of Economcs and Statstcs 64: Blossfeld, Hans-Peter and Yoss Shavt Persstent Barrers: Changes n Eduatonal Opportuntes n Thrteen Countres. Pp 1-23 n Persstent Inequalty, edted by Yoss Shavt and Hans-Peter Blossfeld. Boulder, CO, Westvew. Buchmann, Clauda and Emly Hannum Educaton and Stratfcaton n Developng Country: A Revew of Theores and Research. Annual Revew of Socology 27: Chan, Kam Wng Ctes wth Invsble Walls. Hong Kong: Oxford Unversty Press. Coleman, James S Socal Captal n the Creaton Human Captal Amercan Journal of Socology 94 (suppl.) S 95-S120. Deng, Zhong and Donald Treman The Impact of the Cultural Revoluton on Trends n Educatonal Attanment n the People s Republc of Chna. Amercan Journal of Socology 103: DMaggo, Paul and John Mohr Cultural Captal, Educatonal Attanment, and Martal Selecton. Amercan Journal of Socology 90 (6): DPrete, Thomas A. and Davd B. Grusky Structure and Trend n the Process of Stratfcaton for Amercan Men and Women. Amercan Journal of Socology 96: Djlas, Mlovan The New Class: An Analyss of the Communst System of Power. New York: Frederck A. Praeger Publsher. Doornk, M. van and Jaap Dronkers The Effects of Mother's Pad Occupaton on The Well-beng of Chldren n Dutch Secondary Educaton. Paper presented at the European Conference on Educatonal Research, Sevlle (Span), Sept 25-28, Dronkers, Jaap Workng Mothers and the Educatonal Achevements of ther Chldren. Pp n The Socal World of Adolescents, edted by K. Hurrelmann and U. Engel. Berln, Germany: Walter de Gruyter. Fuller, Bruce, Prema Clarke Rasng School Effects Whle Ignorng Culture? Revew of Educatonal Research 64: Gamoran, Adam and Robert Mare Secondary School Trackng and Educatonal Inequalty: Compensaton, Renforcement, or Neutralty? Amercan Journal of Socology 94: Ganzeboom, Harry. B. G., Paul de Graaf, and Donald J. Treman An Internatonal Scale of Occupatonal Status. Socal Scence Research 21:1-56. Grusky, Davd B. and Thomas A. DPrete Recent Trends n the Process of Stratfcaton. Demography 27:

37 Household Regstraton System and Educatonal Inequalty n Chna 36 Hannum, Emly and Yu Xe Trends n Educatonal Gender Inequalty n Chna: Research n Socal Stratfcaton and Moblty 13: Hannum, Emly Poltcal Change and the Urban-Rural Gap n Basc Educaton n Chna, Comparatve Educaton Revew 43(2) Hannum, Emly Educatonal Stratfcaton by Ethncty n Chna: Enrollment and Attanment n the Early Reform Years. Demography 39: Heyns, Barbara, and Ireneusz Bateck Educatonal Inequalty n Postwar Poland. Pp n Persstent Inequalty, edted by Yoss Shavt and Hans-Peter Blossfeld. Boulder CO: Westvew Press. Hout, Mchael More Unversalsm, Less Structural Moblty: The Amercan Occupatonal Structure n the 1980s. Amercan Journal of Socology 93: Heyneman Stephen P, Loxley Wllam A The Effect of Prmary School Qualty on Academc Achevement across Twenty-nne Hgh- and Low- ncome Countres. Amercan Journal Socology 88: Gerber, Theodore P. and Mchael Hout Educatonal Stratfcaton n Russa Durng the Sovet Perod. Amercan Journal of Socology 101: Gerber, Theodore P Membershp Benefts or Selecton Effects? Why Former Communst Party Members do Better n Post-Sovet Russa. Socal Scence Research 29: Kerckhoff, Alan Insttutonal Arrangements and Stratfcaton Processes n Industral Socetes. Annual Revew of Socology 15: Lang, Za and Yu Por Chen The Educatonal Consequences of Mgraton for Chldren n Chna. Socal Scence Research 36: Lang, Za., and John Z. Ma Chna s Floatng Populaton: New Evdence from the 2000 Census. Populaton and Development Revew 30(3): Lllard, Lee A. and Robert J. Wlls Intergeneratonal Educatonal Moblty: Effects of Famly and State n Malaysa. Journal of Human Resources 29: Lu, Zhqang Insttuton and Inequalty: The Hukou System n Chna. Journal of Comparatve Economcs 33: Maddala.G.S Lmted Dependent and Qualtatve Varables n Econometrcs. Cambrdge: Cambrdge Unversty Press. Mare, Robert D Socal Backgrounds and School Contnuaton Decson. Journal of Amercan Statstcal Assocaton 75: Mare, Robert D. and Chrstopher Wnshp Endogenous Swtchng Regresson for the Causes and Effects of Dscrete Varables. Pp n Common Problems/Proper Solutons: Avod Errors n Quanttatve Socal Research, edted by Scott Long, Newbury Park: Sage Parsh, Wllam L De-stratfcaton n Chna. pp n Class and Socal Stratfcaton n Post-Revoluton Chna, edted by James L. Watson. New York: Cambrdge Unversty Press. Parkn, Frank Class Inequalty and Poltcal Order: Socal Stratfcaton n Captalst and Socalst Socetes. New York: Praeger. Post, Davd Educatonal Stratfcaton, School Expanson, and Publc Polcy n Hong Kong. Socology of Educaton 63: Powers, Danel A, and Chrstopher G. Ellson Interracal Contact and Black racal Atttudes: The Contact Hypothess and Selectvty Bas. Socal Forces 74: Roos, Patrca A Gender and Work: A Comparatve Analyss of Industral Socetes. Albany, NY: State Unversty of New York Press.

38 Household Regstraton System and Educatonal Inequalty n Chna 37 Sakamoto, Arthur, and Mechu Chen Inequalty and Attanment n a Dual Labor Market. Amercan Socologcal Revew 56: Schachter, Frances Fuchs Toddlers Wth Employed Mothers. Chld Development 52: Sedman, Gay Is South Afrca Dfferent? Socologcal Comparson and Theoretcal Contrbutons from the Land of Aparthed. Annual Revew of Socology 25: Shavt, Yoss and Hans-Peter Blossfeld Persstent Inequalty: Changng Educatonal Attanment n Thrteen Countres. Boulder, CO: Westvew Smth, Herbert, and P.L. Cheung Trends n the Effects of Famly Background on Educatonal Attanment n the Phlppnes. Amercan Journal of Socology 9: Solnger, Dorothy J Contestng Ctzenshp n Urban Chna: Peasant Mgrants, the State, and the Logc of Market. Berkeley, CA: Unversty of Calforna Press. Treman, Donald J. and Kermt Terrell Sex and the Process of Status Attanment. Amercan Socologcal Revew 41: Treman, Donald, Mathew McKeever, and Eva Fodor Racal Dfference n Occupatonal Status and Income n South Afrca, 1980 and Demography 33: Tsu, Ka-yuen Economc Reform and the Attanment n Basc Educaton n Chna. The Chna Quarterly 149: Unger, Jonathan Educaton under Mao. New York, NY: Columba Unversty Press. Weber, Max. [1946]. Class, Status and Party. Pp n Max Weber: Essays n Socology, translated by Hans H Gerth and C. Wrght Mlls. Cambrdge, MA: Oxford Unversty Press. Wlls, Robert J. and Sherwn Rosen Educaton and Self-selecton. The Journal of Poltcal Economy 87(5): S7-36. Wu, Xaogang Economc Transton, School Expanson And Educatonal Inequalty In Chna, Research n Socal Stratfcaton and Moblty 28: Wu, Xaogang and Donald Treman The Household Regstraton System and Socal Stratfcaton n Chna, Demography 41: Wu, Xaogang and Donald Treman Inequalty and Equalty under Chnese Socalsm: The Hukou System and Intergeneratonal Occupatonal Moblty. Amercan Journal of Socology 113(2): Zhou, Xueguang, Nancy Brandon Tuma, and Phylls Moen Stratfcaton Dynamcs under State Socalsm: The Case of Urban Chna, Socal Forces 74(3) Zhou, Xueguang, Phylls Moen, and Nancy Brandon Tuma Educatonal Stratfcaton n Urban Chna: Socology of Educaton 71:

39