MPRA Mnich Personal RePEc Archive Are all igrants really worse off in rban labor arkets: new epirical evidence fro China. Jason Gagnon and Theodora Xenogiani and Chnbing Xing OECD Developent Centre Jne 2009 Online at http://pra.b.ni-enchen.de/16109/ MPRA Paper No. 16109, posted 8. Jly 2009 02:38 UTC
OECD DEVELOPMENT CENTRE Working Paper No. 278 Are all igrants really worse off in rban labor arkets? New epirical evidence fro China by Jason Gagnon, Theodora Xenogiani and Chnbing Xing Research area: Poverty Redction and Social Developent Jne 2009
DEVELOPMENT CENTRE WORKING PAPERS This series of working papers is intended to disseinate the Developent Centre s research findings rapidly aong specialists in the field concerned. These papers are generally available in the original English or French, with a sary in the other langage. Coents on this paper wold be welcoe and shold be sent to the OECD Developent Centre, 2 re André Pascal, 75775 PARIS CEDEX 16, France; or to dev.contact@oecd.org. Docents ay be downloaded fro: http://www.oecd.org/dev/wp or obtained via e-ail (dev.contact@oecd.org). THE OPINIONS EXPRESSED AND ARGUMENTS EMPLOYED IN THIS DOCUMENT ARE THE SOLE RESPONSIBILITY OF THE AUTHORS AND DO NOT NECESSARILY REFLECT THOSE OF THE OECD OR OF THE GOVERNMENTS OF ITS MEMBER COUNTRIES CENTRE DE DÉVELOPPEMENT DOCUMENTS DE TRAVAIL Cette série de docents de travail a por bt de diffser rapideent après des spécialistes dans les doaines concernés les résltats des travax de recherche d Centre de développeent. Ces docents ne sont disponibles qe dans ler lange originale, anglais o français ; n résé d docent est rédigé dans l atre lange. Tot coentaire relatif à ce docent pet être adressé a Centre de développeent de l OCDE, 2 re André Pascal, 75775 PARIS CEDEX 16, France; o à dev.contact@oecd.org. Les docents pevent être téléchargés à partir de: http://www.oecd.org/dev/wp o obtens via le él (dev.contact@oecd.org). LES IDÉES EXPRIMÉES ET LES ARGUMENTS AVANCÉS DANS CE DOCUMENT SONT CEUX DES AUTEURS ET NE REFLÈTENT PAS NÉCESSAIREMENT CEUX DE L OCDE OU DES GOUVERNEMENTS DE SES PAYS MEMBRES Applications for perission to reprodce or translate all or part of this aterial shold be ade to: Head of Pblications Service, OECD 2 re André-Pascal, 75775 PARIS CEDEX 16, France 2 OECD 2009
TABLE OF CONTENTS ACKNOWLEDGEMENTS... 4 PREFACE... 5 ABSTRACT... 7 RÉSUMÉ... 7 I. INTRODUCTION... 8 II. INSTITUTIONAL BACKGROUND AND LITERATURE REVIEW... 11 III. DATA AND SUMMARY STATISTICS... 14 IV. MODEL SPECIFICATION... 21 V. BASIC EMPIRICAL RESULTS... 23 VI. DISCUSSION... 29 VII. CONCLUSIONS... 36 APPENDIX 1... 38 APPENDIX 2. BROWN DECOMPOSITION... 40 APPENDIX 3. GENDER ANALYSIS... 43 REFERENCES... 45 OTHER TITLES IN THE SERIES/ AUTRES TITRES DANS LA SÉRIE... 48 OECD 2009 3
ACKNOWLEDGEMENTS The Developent Centre wold like to thank the Beiing Noral University, in particlar Professor Li Shi, for their on-going assistance and collaboration on research in China. The athors wold like to thank and acknowledge the sefl feedback and coents received dring the 10-11 Jly 2008 experts eeting on igration and developent in Paris as well as those received fro participants of the Developent Centre s internal seinar series in October 2008. They wold like to eqally thank Vanda Legrandgérard and Estelle Loisea for their assistance in the editing of this working paper. Chnbing Xing wold like to thank the Developent Centre for inviting hi to spend two onths at the Centre in the Ser of 2008 as well as for their kind hospitality. He also extends his thanks to Li Shi for his encorageent and spport and for the financial spport fro the Ministry of Edcation of the People s Repblic of China (No. 08JC790008). 4 OECD 2009
PREFACE The gradal integration of financial arkets, the sprring of technological progress, the redction of transaction costs and the increase of igration flows in any regions of the world, all spillovers of globalization, have ipacted on eployent otcoes and will contine to do so in the ftre. Despite sccess in creating ore obs in the corse of econoic expansion, welfare gains vary widely between contries and individals. The creation of ore and better obs reains a key challenge and in light of this the OECD Developent Centre has prodced two voles with the obective of generating dialoge on innovative policy options. The first, Policy Coherence for Developent: Migration and Developing Contries, provided an overview on how international igration and the better anageent of people flows can contribte to developent in both host and hoe contries. The second, Is Inforal Noral? Towards More and Better Jobs in Developing Contries, provides evidence that inforal eployent is pervasive in the world, that it is persisting bt that it incldes a certain eleent of heterogeneity within the grop of workers who do work inforally. Many of the newly created obs have neither been able to provide a ini standard of living nor to provide adeqate social protection. Indeed, soe arge that any of the newly created obs are bad obs, locking people into a vicios circle of low pay, high risk and insfficient prospects to clib the social ladder. As one of the OECD s enhanced engageent contries, China offers an interesting exaple to stdy this phenoenon. Over the past decade, China has experienced aor econoic and social transforations. Greater integration into the world econoy and strctral refors have pshed it into a higher growth path. While sstained econoic growth has contribted to bringing extree poverty down, high and often growing disparities in eployent opportnities, copled with a liited coverage of social protection systes, have been associated with rapidly widening earnings and incoe ineqalities. Althogh China is coitted to fighting poverty and ineqality (as exeplified by the Chinese governent's focs on "econoic growth balanced with social harony and stability"), growing segentation in the labor arket has ade it increasingly difficlt to create a haronios labor arket, where obility is free and well-being sstained. This paper looks into the plight of labor igrants in rban China and akes a contribtion to the literatre by sing nationally representative data to deterine if and how labor igrants are discriinated in the labor arket. While rral igrants ay indeed be paid less in rban China, the analysis in this paper shows that this is in fact de to their individal han capital characteristics. However, it is shown that labor igrants are indeed OECD 2009 5
discriinated, not in pay, bt in the free choice of sector in which they work. Being barred ot of the foral sector, they are in effect denied fair access to health and edcational services. While a coplete cancelation of the Chinese hko syste ay not be feasible, this paper provides evidence of a ore pressing isse and one where policy ay have a role. Access to basic social services, regardless of eployent stats, is fndaental in ensring the well-being of workers and their failies. Additionally, other than redcing the nber of obs, an econoic slowdown risks to have a significant ipact on the qality of obs, potentially eroding the working conditions of soe specific grops and rendering the ore vlnerable. In China, this is evident in the potential assive flow of retrn igrants which will eventally deepen preexisting social strains. Therefore, providing portable social secrity shold also for part of the policy agenda. Research in this area will contine at the OECD Developent Centre, notably nder the work prograe entitled "Global Developent Otlook". Javier Santiso Director and Chief Developent Econoist OECD Developent Centre Jne 2009 6 OECD 2009
ABSTRACT The rapid and assive increase in rral-to-rban worker flows to the coast of China has drawn recent attention to the welfare of igrants working in rban regions, particlarly to their working conditions and pay; serios concern is raised regarding pay discriination against rral igrants. This paper ses data fro a rando draw of the 2005 Chinese national censs srvey to shed ore light on the discriination isse, by aking coparisons of earnings and the sector of work between rral igrants on one hand, and rban residents and rban igrants on the other. Contrary to poplar belief, we find no earnings discriination against rral igrants copared to rban residents. However, rral igrants are fond to be discriinated in ters of the sector in which they work, with a vast aority working in the inforal sector lacking adeqate social protection. JEL Classification: O15, R23, J24, J71 Keywords: Migration, China, Discriination, Inforal Eployent RÉSUMÉ L'agentation rapide et assive des oveents rrax-à-rbains d'ovriers vers la côte de la Chine a appelé à l'attention récente le bien-être des igrants travaillant dans des régions rbaines, en particlier à lers conditions et ler salaire de travail; la préoccpation a d atant pls agenté concernant la discriination de salaire contre les igrants rrax. Ce papier eploie des données d'n tirage aléatoire d recenseent national chinois de 2005 por éclaircir la qestion de la discriination, en faisant des coparaisons de revens et de secter de travail entre les igrants rrax d'ne part, et les résidents et igrants rbains de l'atre. Contraireent à la croyance poplaire, nos ne trovons acne discriination de revens contre les igrants rrax coparés ax résidents rbains. Cependant, les igrants rrax s'avèrent être distingés en teres de secter dans leqel ils travaillent, ne grande aorité d entre ex travaillant dans le secter inforel, caractérisé par n anqe d accès à ne protection sociale adéqate. Classification JEL : O15, R23, J24, J71 Mots clés : Migration, Chine, Discriination, Eploi inforel OECD 2009 7
I. INTRODUCTION A cobination of factors in recent years, notably the increasing rral-rban incoe gap 1 and the easing of internal igration restrictions (Cai, 2000), has led to a treendos increase in rral-to-rban igration in China, a reslt largely de to individals in search of work in the coastal regions. Indeed, the nber of igrants has increased significantly over the years fro an estiate of 68 illion in 1996 (Liang, 2001) to one of nearly 140 illion in 2003 (Hang and Zhan, 2005). According to nationally representative censs data fro 2005, igrant workers accont for ore than 20 por cent of the labor force in the rban labor arket 2. There is little dobt as to whether the inflx of igrants into rban China has contribted to econoic growth (Liang, 2001; Song and Zhang, 2003), bt concern prevails as to the iplications of the accopanying geo-deographic transforation on individal welfare. Particlar interest is raised on how igrants fare in the rban labor arket, especially in coparison to their rban resident conterparts. Research shows that igrants disproportionately take p obs in inforal sectors, are paid less, are less likely to be covered by rban social secrity systes and occasionally find it difficlt to get their settled salaries fro their eployers on tie and enforced (China Labor Blletin, 2008). The sitation is exacerbated by China s resident registration (hko) syste, as ost rral igrants retain their rral hko stats, despite the fact that they ay spend a significant aont of tie in rban areas. Hence, the potential for discriination on wages is evident for igrants with rral hko stats bt it ay also be present in the general ability to access foral sector obs. This paper sets ot to answer whether igrants are really worse off in rban labor arkets in China. Or starting point is a coon one fond in the research literatre on discriination: the notion that wage gaps ay owe to either differing levels of individal han capital broght to the labor arket or fro differing sets of skill-prices offered on this arket. Indeed, it is a relevant starting point as it carries notable policy iplications. If, for instance, rral igrants are paid less as a reslt of lower skills broght to the rban labor arket, reforing the rban labor arket will have little effect on their well-being, whereas providing the with edcation and appropriate skills will enable the to exploit better labor arket opportnities. However, if igrants have significantly lower wages copared to rban residents 3 with 1. The ost recently pblished ratio is estiated at 3.2:1 (China Statistical Yearbook, National Brea of Statistics of China, 2007) 2. In soe relatively developed coastal regions, sch as those of Gangdong and Fian, the share of igrants is greater than 50%. 3. For the reainder of the text, rban resident is eant to ean non-igrant rban resident. 8 OECD 2009
identical characteristics, the root case of wage differences cold be discriination. Reforing the labor arket, if this is the case, wold lead to an increase in welfare for igrants. Moreover, looking for differences within the sbgrops of igrants will provide additional insight into the deterinants of discriination. We first estiate a ltinoial logit odel to deterine the factors deterining the choice of eployent sector and then apply the reslts to a decoposition analysis to test for the existence of wage discriination. The Oaxaca-Blinder (OB) decoposition fraework (Blinder, 1973; Oaxaca, 1973) provides a sefl analytical instrent with which to face the qestion of discriination and is often applied to copte the explained (endowent) effect and the nexplained (discriination) effect in potential discriinatory cases of gender or black/white incoe differentials. We follow the literatre in this sense bt the coparison between the two grops, rral igrants and rban residents, is ltifaceted as it involves two diensions of potential discriination instead of one, and or reslts hinge on this dichotoy. The first diension is hko stats (rral vs. rban), while the second is igrant stats (igrant vs. resident). We therefore introdce another reference grop -naely rban igrants 4 - to distingish these two effects. As rban igrants are different fro rban residents only in ters of igrant stats, we interpret the positive nexplained figre of the OB decoposition as a prei associated with igration (or a selection effect); and as rban igrants and rral igrants differ only in hko stats, we interpret the nexplained figre of the OB decoposition as discriination against rral hko stats. As we are concerned by the possibility of segentation in the rban labor arket, another featre we take into accont is the distribtion of individals in the foral and inforal sectors as this will have an effect on incoe and well-being. This isse is also at the heart of policy preoccpations with respect to the working conditions of rral igrants and in particlar their access to good secre obs. In light of this we apply a Brown decoposition exercise (an extended version of the Oaxaca-Blinder decoposition) between and within the different grops and report the findings in Appendix 2 5. Secondly, becase sectoral segentation can have a sbstantial ipact on social secrity coverage despite having a possible liited ipact on incoe, we also apply an OB decoposition directly to differences in sectoral distribtion. The paper ses a recent nationally representative dataset, a one-fifth rando draw fro the 1 per cent censs data of China, to explore the isses raised above. Reslts show that ost of the incoe differential between rral igrants and rban residents can be explained by differences in individal characteristics. However by sing rban igrants as a control grop, we find that the presed inexistence of a discriination effect we obtain while coparing rral igrants and rban residents is the reslt of a conterbalance between a discriination effect against rral hko stats and a prei effect accred by igrants. As for sectoral distribtion, a Oaxaca-Blinder decoposition indicates that all igrants (both rral and rban) are 4. Migrants with an rban hko stats oving fro one rban region to another. 5. Cai and Wang (2006) also find discriination along gender lines in China. We therefore report gender dissagregated reslts in Appendix 3. OECD 2009 9
discriinated along sector choice 6. The extent of discriination is larger for rral igrants indicating a frther discriination against rral hko stats. In ters of incoe, rral igrants enoy a prei associated with igration and sffer discriination in rban labor arkets siltaneosly. The reason that they earn less when we copare the with rban residents is de to their lower levels of han capital. Increasing the edcation level of rral igrants and providing the with sefl skills will help increase their incoe and earning opportnities. As both rral and rban igrants face nfair treatent in sector choice, reforing the labor arket, especially by reoving sector barriers can help increase eployent in the foral sector and facilitate earnings obility for igrants. The paper is organised as follows: Section II provides a short discssion on the instittional backgrond and a literatre review on labor arket segentation in China. Section III describes the data. Section IV presents and discsses the odel specification and Section V reports or basic epirical reslts, where we copare rral igrants and rban residents. In Section VI, we present a ore profond analysis coparing different grops. The last section concldes. 6. More than 80% of the difference is nexplained 10 OECD 2009
II. INSTITUTIONAL BACKGROUND AND LITERATURE REVIEW Several articles provide an overview of the instittional backgrond of China s rralrban igration, with particlar ephasis on the hko registration syste; aong the Cai (2000), Zhao (2005), Deng and Gstafsson (2006) and de la Rpelle (2007). Despite the several refors to the syste since the 1970s, deliberate discriination of igrants in cities reained legal ntil very recently, in order to redce copetition of rral igrants in rban centres (Cai, 2000). Today, rban China faces the challenge of integrating two distinct labor force grops, and it is still nclear as to whether they are copleentary or copeting for the sae obs. Rral igrants, it has been shown, have higher ob obility in the rban labor arket and generally lower ob tenre rates than rban residents (Knight and Yeh, 2003). Althogh the two grops work in seeingly segented labor arkets, copetition ay be increasing as ore internal igration and labor arket refors gain pace (Knight and Yeh, 2004). Rral igrants in rban centres can be divided into two grops: those who have obtained an rban hko registration and those who have not. Migrants with an rban hko are registered officially as rban residents, a prereqisite to be covered by the rban social secrity syste and to gain access to varios fors of pblic assistance. Moreover, once registered as rban residents, peranent igrants forfit their rral resident stats, their right to agricltral land in their conity of origin as well as their voting rights on village affairs. Both anecdotal evidence and acadeic research (Deng and Gstafsson, 2006, for exaple) indicate that rral igrants who sccessflly obtain an rban hko registration are well integrated in rban society, at least after acclating experience in the rban labor arket over tie 7. Alternatively, any rral igrants retain their rral hko stats, whether deliberately or not, and ths retain rights on their rral land and a voice on political affairs in their village of origin. These igrants are less integrated in rban labor arkets copared to peranent igrants. They ay also find it difficlt to gain access to obs in foral rban sectors de to their registration stats (Zhao, 2005). As a reslt, non-rban-registered igrants are often paid less on average. Althogh they ay spend a significant aont of tie in rban areas, they ay additionally not be covered by the rban social secrity syste nor be entitled to varios other benefits. 7. Soe rral igrants can sccessflly obtain a peranent rban hko stats after leaving a rral area, and therefore are often deeed peranent igrants. It shold be noted that peranent igrants are different fro rban igrants with the latter having oved fro other rban areas and hence holding an rban hko. We do not consider peranent igrants in or paper. OECD 2009 11
The hko syste ths creates iportant distortions and increases ineqality in the rban Chinese labor arket (Whalley and Zhang, 2004) despite the fact that several papers point to igration within China as a natral echanis for rral-rban incoe convergence (Lin et al., 2004; D et al., 2004). In fact, althogh igrants have been oving to rban labor arkets for any years, the hko syste has ensred that the rban labor arket reains segented opening the possibility of discriination against those who are not registered in rban centres. Research also shows that igrants in China are positively self-selected on the basis of (both observed and nobserved) characteristics which increases the likelihood of yielding a positive otcoe in the labor arket 8, yet despite these favorable characteristics, evidence of discriination against rral igrants in the rban labor arket has been docented. Data fro the 2002 China Hosehold Incoe Proect (CHIP) shows that igrants theselves perceive to be discriinated against in rban labor arkets (Derger et al., 2008). Both casal observation and existing research (Meng and Zhang, 2001; UNDP, 2005) indicate that a significant share of igrants take p obs in the inforal sector, are paid less and are also less likely to be covered by rban social secrity systes (Wei, 2007). The lack of social secrity coverage is likely to contribte to an iportant decrease in welfare; a report by the China Labor Blletin (2008), for instance, reported that the crrent wage gap between rban and rral regions wold increase fro 3-fold to 6-fold in real ters, if we considered the benefits accred fro social secrity. Even worse is that igrants occasionally find it difficlt to get their settled salaries fro their eployers on tie and enforced 9. Despite extensive refors in ini wage legislation (see The 1994 Labor Law), the large nber of igrants working inforally ensres that the ini wage is not binding. Many epirical papers have analysed the rise in rban labor arket segentation and the disparity between the rral and rban sectors in China (see for instance Whalley and Zhang, 2004; Knight and Li, 2005). Althogh the relative disadvantages of rral igrants raises concern, little has been done to stdy the deterinants of their labor arket otcoes and better nderstand labor arket otcoe differentials between the and rban residents 10. 8. For instance, selection has been docented on the basis of level of edcation, age, health stats or gender (Kikchi et al., 2000; W, 2008). 9. The China Labor Blletin (2008) clais that in 2004 there were 114 997 labor dispte lawsits filed by igrants. 10. Aong the very few, Meng and Zhang (2001) find that edcated rban residents are ore likely to have a white-collar ob or to work in wholesale or retail trade occpations. Moreover, despite wage discriination against igrants (which can be as high as 50%), they find that 82% of the discriination is de to ineqality between sectors. Meng (2001) finds that igrants with higher levels of edcation and rban labor experience are ore likely to be self-eployed in the inforal sector. Shi and Zhang (2006) find that the retrn to edcation in the rban labor arket is arond 5.4%, and show that edcation is iportant in deterining higher wages for igrants in rban centres. Dérger et al. (2008), decopose annal earnings differences between rban residents and rral igrants into for categories (a sectoral effect, a wage effect, an hors worked effect and a poplation effect) and find that igrant workers have a coparative advantage in working in the private sector while the opposite holds for rban residents. Moreover, the poplation effect, the 12 OECD 2009
Another related qestion is whether the rban labor arket is segented with respect to the hko syste. Given that an individal s hko stats ay be correlated with the individal characteristics of a igrant sch as edcation level, work experience and ability, it is difficlt to tell whether poor perforance in the labor arket is de to these characteristics (e.g. low han capital) or hko stats. Many stdies on discriination se the Oaxaca-Blinder fraework to test for discriination in the labor arket (the nexplained share of the incoe differential). Many of the, however, se data fro different regions at different ties and ltiately derive different conclsions, in effect aking any coparison a difficlt task. For instance, Meng and Zhang (2001) find that 51 per cent of the wage differential between rban residents and igrants is de to nexplained factors (discriination, loosely speaking) while Dinh and Marer-Fazio (2004) find 25 per cent and Wang (2005) 43 per cent, each sing different datasets which focs on different regions of the contry. Deng (2007), sing the China Hosehold Incoe Proect (CHIP) data collected by the CASS 11 and which reasonably covers the contry, finds that 60 per cent of the incoe differential originates fro nexplained factors. As coented by Zhao (2005), The datasets that have been sed in existing research papers are qite varied, and in soe cases they are otdated. In any cases, it can be said that the reslts cannot be generalised as the data is not representative of the entire poplation of the contry. In this paper we wish to nderstand the reason for wage differences in rban labor arkets between igrants and rban residents. We se a Oaxaca-Blinder odel to decopose the difference between skill levels and skill-prices. Or research contribtes to the literatre by sing a ore representative dataset than previos stdies, the 1/5 th rando draw fro the 2005 1 per cent national censs, which allows s to distingish between work in the foral and the inforal sectors. This is a salient featre when stdying igration, as other saple datasets ay not flly captre all igrants, especially those working inforally. Discriination can appear in any fors. By introdcing a sectoral breakdown along inforal-foral labor arket segentation, we gain a ore coplete and realistic pictre of rral igrant labor otcoes in rban labor arkets. nderlying individal characteristics of rban residents and igrants, is significantly iportant, signaling that pre-arket rather than on-arket factors prevail. 11. Chinese Acadey of Social Sciences. OECD 2009 13
III. DATA AND SUMMARY STATISTICS III.1 The Data The data we se is a one-fifth rando draw fro the 1 per cent censs data of China adinistered by the National Brea of Statistics (NBS) in 2005. The saple size is arond 2.3 illion individals covering 31 provinces, nicipalities and atonoos regions. We se the sapling rle to extrapolate the total poplation in China, which is 1.29 billion, slightly lower than the 1.31 billion referenced in the China Statistical Yearbook (China Statistical Yearbook, National Brea of Statistics of China, 2006) 12. Generally speaking or data is representative of ainland China. This is confired by Figre A1 in Appendix 1, where we show the weighted saple shares by province copared to the corresponding shares we get fro the Chinese Statistical Yearbook (CSY). The difference we find is negligible. The 1 per cent censs data has great advantages for stdying igrant labor arket otcoes. An ordinary hosehold srvey ay be less likely to obtain a representative saple of igrants de to the floating natre of igrants and de to the sapling process. For exaple, srveys sapling igrants fro neighborhoods or conities ay nder-represent those who arrived recently and those who live at the constrction site collectively. Censs data does not sffer fro sch probles. Another advantage of or censs data is that, not only can we identify rral-to-rban igrants bt also rban-to-rban igrants, which allows s to have a ore detailed and coprehensive pictre of labor arket integration in the rban labor arket as it offers an alternative coparison grop to rral igrants. III.2 Definitions Two qestions in the qestionnaire are sed to identify igrants: (1) Where is yor hko registration location? and (2) How long have yo left yor hko registration location? We define igrants as those who have left their hko registration location for ore than half a year and disaggregate the into for categories according to the place where they were living at the tie of the srvey (city, town or village) and their hko type (rral or rban). Naely, we define these categories as follows: 12. There are at least two possible reasons for the difference. First, or data does not inclde Hong Kong, Maca, and Chinese Taipei, while the Chinese Statistical Yearbook does. Second, ilitary staff ay be nderrepresented in or saple. 14 OECD 2009
rral-to-rral: individals with rral hko stats who have oved to another rral area (village or town). (city). rral-to-rban: individals with rral hko stats who have oved to an rban area rban-to-rral: individals with rban hko stats who have oved to a rral area rban-to-rban: people with rban hko stats who have oved to another rban area. The saple sizes of these for types of igrants are 38.7, 159.5, 12.5, and 116.8 thosand respectively. Given the fact that or saple is a 1/5 th draw fro the 1 per cent national censs, we can estiate the corresponding totals for these for grops, which are 19.4, 79.7, 6.3, and 58.4 illion respectively 13. The total nber of igrants is approxiately 164 illion, nearly 12.5 per cent of the entire poplation. The shares of the above for types of igrants are 1.5 per cent, 6.1 per cent, 0.5 per cent, and 4.5 per cent of the total poplation 14. We also disaggregate igrants by the reason they give for leaving their hko registration location (see Table A2 in Appendix 1). Across all grops, ost igrants ove for work, and this is especially tre for rral-to-rban igrants (the share approxiately 61 percent). People also igrate for other reasons; for rral-to-rral igrants, the second largest reason is arriage (which acconts for 19 per cent), while for rral-to-rban igrants, the second ost iportant reason is to be with their failies (relatives). Marriage is also a aor reason for this type of igration. The nber of rban-to-rban igrants is also large, bt the share of those who ove for ob opportnities is relatively sall (only 20 per cent). As we are ainly concerned with rban labor arket otcoes, we focs on the following three grops: rral-to-rban igrants, rban-to-rban igrants, and rban residents (non-overs). In addition, we restrict or saple to those ot of school and aged 16 to 60 and those who igrate to look for work and for bsiness reasons. This paper also acconts for the evidence linked to the segented natre of the Chinese rban labor arket 15 by extending the analysis to foral and inforal eployent. Or data allows for two definitions of inforal eployent: (a) self-eployent and (b) the absence of a foral labor contract 16. As it will be shown, these two tally exclsive definitions allow for 13. The nweighted reslts are 14.2 illions, 66.4 illion, 3.8 illion and 43.7 illion respectively, sing p to 128 illion. 14. It is difficlt to copare the agnitde of igrants to other papers since different researchers se different criteria to identify igrants depending on data availablility. For exaple, Cai and Wang (2003) find there are 131 illion rral igrants sing the 2006 Chinese agricltral censs, which is in fact larger than the one we se (99.1 illion). The difference between this figre and or figre of 128 illion can be explained by coparing how igrants are defined. The definition of a igrant in their stdy is based on whether individals spent at least one onth otside their hoe conties. 15. For exaple, Meng and Miller (1995) ephasize segentation based on occpation while Derger et al. (2008) ephasize segentation based on ownership. 16. For the reainder of the text we will reference this grop as the no contract grop. OECD 2009 15
soe degree of heterogeneity within inforal eployent. Both are characterised by the lack of social secrity coverage. Eployees with a foral contract, whether long-ter or short-ter, are defined as being forally eployed. Adding indstry and occpation dies in or regressions does not significantly alter or reslts. III.3 A Few Sary Statistics Sary statistics are reported in Tables 1 and 2. Or data show that rban residents (non-overs) are gender balanced, qite edcated, ostly all arried and that any are working in the foral sector, and particlarly in the pblic sector. Self-eployed workers are typically older, less edcated and ale, while showing higher incoe earnings than no contract eployees. Rral igrants are yonger, less edcated and ore likely to be ale. Again, we observe differences depending on the definition of inforal eployent we se. The self-eployed are older and less edcated while foral workers have siilar characteristics as the no contract eployees. Urban igrants are siilar to rral igrants in ters of age, gender balance and arriage stats bt have ch higher edcation levels and also earn the ost. In ters of inforal eployent, self-eployed workers earn ore than workers withot a contract for both types of igrants, bt the opposite is observed for rban residents, where waged eployees earn ore than the self-eployed. One possible explanation is the high nber of individals working in the pblic sector and foring the grop eployed in so-called iron rice bowl obs. Apart fro coparing average earnings, we also plot incoe distribtions for each grop (see Figres 1 to 4 below). When we copare the incoe distribtions of rban residents, rral igrants and rban igrants (Figre 1) that of rral igrants has the sallest dispersion. Taking the incoe distribtion of rban residents as a benchark, that of rral igrants falls disproportionally on the lower-edi part of the benchark distribtion. This is st the opposite for rban igrants. It has greater dispersion, and it falls disproportionally on the pper-half of the benchark distribtion. Althogh the distribtions have different shapes and positions, they are generally noral. This paper looks at the ean incoe differential as opposed to the incoe differential along the whole range of the incoe distribtion and attepts to explain the difference between these three incoe distribtions. In Figres 2 to 4, we disaggregate the incoe distribtion of each grop into three categories: foral eployent, self-eployent, and no contract eployees. There are noticeable differences between foral and inforal eployent, and between the self-eployed and the grop of no contract eployees. This is an indication that segentation exists between foral and inforal eployent, and there is soe degree of heterogeneity within the inforal sector. 16 OECD 2009
0 density: lhinc.2.4.6.8 0 density: lhinc.2.4.6.8 Figre 1. Distribtion for Urban Residents, Rral Migrants and Urban Migrants rban res rral igr rban igr -4-2 0 2 4 6 Ln of horly incoe distribtion Figre 2. Distribtions for Foral and Inforal Eployent, Urban Residents foral self ep no contr -4-2 0 2 4 6 Ln of horly incoe distribtion OECD 2009 17
0 density: lhinc.2.4.6.8 0 density: lhinc.2.4.6.8 1 Figre 3. Distribtions for Foral and Inforal Eployent, Rral Migrants foral self ep no contr -4-2 0 2 4 6 Ln of horly incoe distribtion Figre 4. Distribtions for Foral and Inforal Eployent, Urban Migrants foral self ep no contr -4-2 0 2 4 6 Ln of horly incoe distribtion 18 OECD 2009
Table 1. Sary Statistics Urban residents Rral igrants Urban igrants total foral self-ep no contr total foral self-ep no contr total foral self-ep no contr age edcation levels 39.9 40.7 39.1 37.1 30.5 29.6 35.8 29.1 32.5 32.4 36.8 30.5 Priary and below 0.10 0.11 0.15 0.06 0.21 0.17 0.31 0.22 0.06 0.04 0.10 0.06 nior iddle school 0.37 0.35 0.53 0.34 0.62 0.62 0.57 0.64 0.32 0.26 0.46 0.38 senior iddle school 0.31 0.32 0.28 0.32 0.15 0.18 0.11 0.13 0.36 0.36 0.34 0.36 College and above 0.22 0.22 0.04 0.29 0.02 0.03 0.01 0.01 0.26 0.34 0.10 0.20 # of people in the hosehold 3.65 3.61 3.88 3.71 4.75 5.31 3.17 4.89 3.47 3.48 2.95 3.71 gender 1.50 1.53 1.37 1.44 1.44 1.52 1.30 1.43 1.44 1.45 1.36 1.46 arital stats. 0.11 0.11 0.06 0.14 0.34 0.38 0.08 0.42 0.33 0.34 0.10 0.43 left the hko within 0.5-3 years 0.57 0.56 0.45 0.63 0.54 0.52 0.49 0.59 occpation Manager, officials, < 0.04 0.05 0.02 0.04 0.02 0.03 0.02 0.01 0.06 0.10 0.04 0.02 technician (professional) 0.23 0.28 0.04 0.23 0.02 0.03 0.02 0.02 0.15 0.19 0.05 0.13 adinistrative staff 0.15 0.16 0.01 0.17 0.03 0.05 0.01 0.03 0.09 0.12 0.01 0.08 service staff 0.26 0.18 0.64 0.27 0.34 0.25 0.64 0.29 0.45 0.33 0.71 0.47 related to ag, forestry, fishery ect. 0.04 0.06 0.04 0.02 0.02 0.04 0.02 0.01 0.01 0.01 0.02 0.00 related to anfactre/transport/ect. 0.27 0.27 0.26 0.28 0.56 0.60 0.30 0.63 0.25 0.25 0.17 0.29 eployent stats eployee 0.82 0.91 0.00 1.00 0.74 0.78 0.00 1.00 0.71 0.76 0.00 1.00 eployer 0.03 0.06 0.00 0.00 0.05 0.14 0.00 0.00 0.09 0.19 0.00 0.00 self-eployed 0.13 0.00 1.00 0.00 0.19 0.00 1.00 0.00 0.17 0.00 1.00 0.00 hosehold worker 0.02 0.03 0.00 0.00 0.03 0.08 0.00 0.00 0.03 0.05 0.00 0.00 ownership pblic sector 0.26 0.28 0.00 0.34 0.01 0.01 0.00 0.01 0.05 0.07 0.00 0.04 SOE 0.26 0.39 0.00 0.13 0.04 0.05 0.00 0.04 0.10 0.16 0.00 0.07 collective owned enterprises 0.05 0.06 0.00 0.06 0.03 0.05 0.00 0.03 0.04 0.05 0.00 0.05 Faily bsiness (registered) 0.19 0.07 0.82 0.15 0.35 0.22 0.79 0.28 0.35 0.21 0.89 0.27 private enterprises 0.12 0.10 0.00 0.23 0.34 0.36 0.00 0.47 0.32 0.33 0.00 0.45 other work nit 0.04 0.04 0.00 0.04 0.13 0.26 0.00 0.10 0.10 0.16 0.00 0.08 others 0.07 0.06 0.18 0.06 0.09 0.05 0.21 0.07 0.05 0.03 0.11 0.05 onthly incoe 1058 1188 848 902 973 1100 982 878 1527 1905 1231 1133 horly incoe 6.12 7.04 4.23 5.19 4.61 5.38 4.57 4.07 8.25 10.62 6.11 5.92 type of contract Fixed short ter 0.21 0.34 0.00 0.34 0.94 0.00 0.43 0.81 0.00 long ter contract 0.41 0.66 0.00 0.02 0.06 0.00 0.10 0.19 0.00 no contract 0.38 0.00 1.00 0.64 0.00 1.00 0.47 0.00 1.00 no neployent insrance 0.69 0.63 0.94 0.78 0.95 0.88 1.00 0.98 0.76 0.63 0.95 0.90 no pension 0.45 0.38 0.77 0.55 0.89 0.76 0.98 0.96 0.61 0.44 0.79 0.81 no edical insrance 0.46 0.41 0.81 0.52 0.85 0.71 0.94 0.93 0.64 0.48 0.84 0.83 OECD 2009 19
Table 2. Sary Statistics for Recent and Non-Recent Migrants Rral igrants Urban igrants Recent Non recent Recent Non Recent age edcation levels 28.3 33.5 30.7 34.5 Priary and below 0.19 0.25 0.05 0.06 nior iddle school 0.65 0.58 0.32 0.33 senior iddle school 0.15 0.15 0.36 0.35 College and above 0.02 0.02 0.27 0.25 # of people in the hosehold 5.27 4.06 3.60 3.31 gender 1.47 1.40 1.45 1.42 arital stats. 0.45 0.19 0.41 0.23 eployent stats eployee 0.80 0.65 0.75 0.67 eployer 0.03 0.07 0.08 0.11 self-eployed 0.15 0.25 0.15 0.19 hosehold worker 0.02 0.04 0.02 0.03 occpation Manager, officials, < 0.01 0.03 0.05 0.08 technician (professional) 0.02 0.03 0.15 0.15 adinistrative staff 0.03 0.04 0.08 0.09 service staff 0.32 0.38 0.46 0.43 related to ag, forestry, fishery ect. 0.02 0.03 0.01 0.01 related to anfactre/transport/ect. 0.60 0.51 0.25 0.24 ownership pblic sector 0.01 0.01 0.05 0.05 SOE 0.04 0.04 0.10 0.10 collective owned enterprises 0.03 0.03 0.04 0.03 self-eployed 0.32 0.41 0.34 0.36 private enterprises 0.37 0.30 0.32 0.31 other work nit 0.15 0.11 0.11 0.10 others 0.08 0.11 0.05 0.05 onthly incoe 901 1072 1371 1715 horly incoe 4.28 5.07 7.43 9.24 type of contract Fixed short ter contract 0.34 0.34 0.42 0.44 long ter contract 0.02 0.02 0.09 0.12 no contract 0.64 0.64 0.49 0.45 no neployent insrance 0.95 0.94 0.77 0.75 no pension 0.89 0.88 0.63 0.59 no edical insrance 0.85 0.85 0.66 0.62 20 OECD 2009
IV. MODEL SPECIFICATION We se two ain epirical strategies in this paper. First we eploy a ltinoial logit odel to infer siple correlations related to sector choice and also to generate sary statistics for conterfactal predictions. Three ltinoial logit odels are estiated, one each for rban residents, rban igrants and rral igrants, featring eployent sector as the dependent variable. Eployent sectors are defined as: = foral, self-eployent and no-contract. The foral odel is as follows: Pr( y i where ) 1 exp( X ) J i exp( X ) i ( 1,2,3) (1) X i is a vector of explanatory variables related to sector. Second, we se a Oaxaca-Blinder decoposition fraework acconting for wage differences between rral igrants and rban residents 17. The incoe differential between the two grops can be decoposed into two parts: one de to differences in individal skill levels (the so-called endowent effect) and the other de to the differences in the skill-prices individals face in the labor arket (the price effect). The Oaxaca-Blinder odel is estiated in two steps. First, wage regressions one for each grop (rral igrants and rban residents) are estiated, by assing wages for each grop can be defined as follows: W g i g g g g X i i (, ) g (2) where g i W refers to the incoe (in log for) of individal i where g, refers to g rban residents and rral igrants. X i is a vector of independent variables, inclding edcation, age, arital stats, gender, province dies, indstry dies and occpation g dies. is the intercept for grop g. bar g g g g If the odel is estiated sing an OLS odel, we can state W ˆ X ˆ, with the over on W and X refering to saple eans, and ˆ g, ˆ g the OLS estiates for g, g. Differencing ot the ean wages for both grops, the typical Oaxaca-Blinder odel is then as follows: 17. For the oent we only consider differences between rral igrants and rban residents, and let the analysis on rban igrants for later. OECD 2009 21
W W ( ) ( X X ) ( ) X (3) The second ter on the right-hand side ( X X ) is the wage differential de to differing individal characteristics (sch as han capital) in the absence of discriination. The third ter, ( )X, easres the proportion of the relative wage differential de to discriination. Discriination is easred as the residal, or the nexplained difference in the regression coefficients. We also calclate a Brown et al. (1980) decoposition ( Brown ), which additionally considers foral and inforal sectors. The fll Brown odel is elaborated in Appendix 2. In addition to analysing incoe differentials sing a Oaxaca-Blinder decoposition techniqe, we also look at the differences with respect to sector choice. In particlar we decopose differentials in sectoral distribtions into endowent and price effects 18. For this prpose we estiate a linear probability odel (instead of a ltinoial logit odel), and then se the reslts to calclate a Oaxaca-Blinder decoposition based on sector of eployent. We do this by defining a dy variable for work in the foral sector (eqal to zero if it is work in the inforal sector). 18. Or sector-choice strctre effect. 22 OECD 2009
V. BASIC EMPIRICAL RESULTS As indicated, or epirical strategy contains several steps. First, we estiate the three ltinoial logit odels to deterine characteristics consistent with sector eployent and then se these reslts to predict conterfactal (and factal) sectoral distribtions. Second, we estiate incoe eqations for different sectors for both rban residents and rral igrants sing siple OLS regressions, and calclate Oaxaca-Blinder decopositions for each sector sing reslts fro the regressions. A Brown decoposition analysis is also perfored to ensre robstness and the reslts relating to it are reported in Appendix 2. V.1. Reslts of the Sector Choice Model Table 3 presents the arginal effects of the ltinoial logit regression for sector choice. For rban residents, age, edcation, gender, and arital stats all have significant effects on sector eployent. Taking foral eployent as the base category, the probability of being self-eployed takes an inverted U-shape with respect to age. Individals with higher levels of edcation, woen and narried individals are less likely to be self-eployed. The effects of these variables on the probability of being in the no contract category are qite different. First, as people age, the probability of not having a foral contract decreases and after the age of 35, the probability changes little with age. The ore edcated are also less likely to be withot a contract. The effect of college edcation on self-eployent and no contract work is qite different however. College edcation copared to senior iddle school sees to have only a arginal (and even negative) effect for people working withot a labor contract. As the nber of college gradates increased treendosly in the last several years, the pressre to find a good ob has also conseqently increased and or reslts ay reflect the crrent sitation of any college gradates. Finally, woen and narried individals are ore likely to work otside the protective confines of a foral contract. The signs of the coefficients for rral igrants are the sae as for rban residents (except when looking at gender) bt the agnitde of the coefficients differs qite a bit. The reslts in Table 3 indicate that the rban labor arket treats rban residents and rral igrants with identical individal characteristics differently, at least with respect to their choice of sector. To see this difference ore clearly, we se the ltinoial logit odel reslts for rban residents to predict the conterfactal sectoral distribtion for rral igrants. Fro Table 4 we see that if they were treated as rban residents based on observable characteristics, rral igrants wold have a different sectoral distribtion. Most iportantly, approxiately 10 per cent of rral igrants wold be reallocated fro the no contract category to foral eployent, whereas the share of the self-eployed wold not change ch. We can expect this to have an effect on the incoe differential between rral igrants and rban residents. OECD 2009 23
Table 3. Marginal Effects of Mltinoial Logit Regressions Urban residents Rral igrants Urban igrants Self-eployed No contract Self-eployed No contract Self-eployed No contract dp/dx dp/dx dp/dx dp/dx dp/dx dp/dx s.e. s.e. s.e. s.e. s.e. s.e. Age 16-20 oitted age: 21-25 0.022*** -0.144*** 0.059*** -0.046*** 0.053*** -0.061*** (0.005) (0.009) (0.006) (0.006) (0.015) (0.014) age: 26-30 0.025*** -0.193*** 0.094*** -0.071*** 0.091*** -0.123*** (0.005) (0.009) (0.007) (0.008) (0.015) (0.016) age: 31-35 0.019*** -0.220*** 0.115*** -0.088*** 0.115*** -0.128*** (0.005) (0.009) (0.007) (0.008) (0.015) (0.017) age: 36-40 0.019*** -0.227*** 0.132*** -0.103*** 0.124*** -0.136*** (0.005) (0.009) (0.007) (0.009) (0.016) (0.019) age: 41-45 0.017*** -0.224*** 0.142*** -0.100*** 0.125*** -0.173*** (0.005) (0.009) (0.007) (0.010) (0.016) (0.020) age: 46-50 0.009* -0.219*** 0.154*** -0.123*** 0.116*** -0.128*** (0.005) (0.009) (0.007) (0.012) (0.017) (0.022) age: 51-55 0.001-0.232*** 0.138*** -0.100*** 0.116*** -0.175*** (0.005) (0.010) (0.008) (0.015) (0.017) (0.026) age: 55-60 0.008-0.204*** 0.120*** -0.103*** 0.125*** -0.075** (0.006) (0.011) (0.010) (0.020) (0.020) (0.035) Priary and below oitted Jnior iddle school -0.033*** -0.013** -0.029*** -0.089*** -0.029*** -0.063*** (0.002) (0.005) (0.002) (0.005) (0.007) (0.018) Senior iddle school -0.089*** -0.082*** -0.050*** -0.177*** -0.073*** -0.169*** (0.002) (0.005) (0.004) (0.006) (0.008) (0.018) College and above -0.214*** -0.071*** -0.109*** -0.237*** -0.147*** -0.285*** (0.002) (0.005) (0.011) (0.014) (0.009) (0.019) narried -0.032*** 0.032*** -0.121*** 0.089*** -0.077*** 0.079*** (0.002) (0.004) (0.004) (0.006) (0.007) (0.010) Feale -0.017*** 0.029*** -0.049*** -0.001-0.014*** 0.029*** (0.001) (0.002) (0.002) (0.003) (0.004) (0.007) Observations 219712 94621 22214 Note: the base category is foral eployent. region dies not reported. *** p<0.01, ** p<0.05, * p<0.1 rban Table 4. Actal and Conterfactal Sectoral Distribtions based on Mltinoial Logit Regression Reslts actal Predicted based on logit rral igrants rban igrants rral igr rban igr Rral igrant as rban resident Urban igrant as rban resident Urban igrant as rral igrant non recent actal predicted actal predicted recent Recent as non recent Foral 0.585 0.348 0.517 0.457 0.571 0.440 0.345 0.351 0.359 0.535 0.501 0.524 Self-eploy 0.118 0.175 0.147 0.170 0.105 0.162 0.232 0.133 0.168 0.167 0.130 0.139 No contract 0.298 0.477 0.336 0.373 0.324 0.398 0.423 0.517 0.474 0.298 0.369 0.337 non recent recent Recent as non recent 24 OECD 2009
V.2. Incoe Deterination for Different Sbgrops We next consider incoe deterination for foral work, self-eployent and no contract eployees, for both rban residents and rral igrants. The ai of this exercise is to bring ot the price difference between the sectors and between rral igrants and rban residents. The reslts pertaining to the OLS regressions are presented in Table 5 and we highlight soe of the ore iportant reslts in by plotting incoe-age profiles and retrns to edcation in Figre 6. For rban residents (left panel of Table 5), the incoe-age profiles vary qite a bit depending on sector of work; for the foral eployent and the no contract eployees grops incoe levels increase with age, while for the self-eployed incoe first increases and then decreases with age. For rral igrants, however, all three grops show no significant differences in the incoe-age profile. In addition they are qite siilar to that of self-eployed rban residents. Their incoe level begins to decrease arond 25 to 30 years of age. As shown in Figre 6, the retrns to edcation are also different; for rban residents there are no significant differences in retrns to edcation between foral eployent and no contract work, whereas the self-eployed have the lowest retrns to edcation. The sae is tre for rral igrants, with the forally eployed having the highest retrn to edcation aong the. Beyond this, the rban labor arket also rewards gender and arital stats differently. Woen s earnings are significantly lower than those of en. The differential is ch higher for the self-eployed than for the forally eployed and no contract eployees. There are only slight differences in coefficients between rban residents and rral igrants along gender lines. Table 5. OLS Regression Reslts: Dependent Variable=Log(Horly Incoe) Urban residents Rral igrants Urban igrants foral Self-ep No contr foral Self-ep No contr foral Self-ep No contr Age 16-20 oitted age: 21-25 0.114*** 0.113** 0.093*** 0.073*** 0.177*** 0.117*** 0.117*** 0.006 0.130*** (0.016) (0.045) (0.013) (0.009) (0.036) (0.007) (0.027) (0.119) (0.023) age: 26-30 0.227*** 0.108** 0.157*** 0.128*** 0.151*** 0.138*** 0.276*** 0.096 0.191*** (0.017) (0.045) (0.014) (0.011) (0.036) (0.009) (0.029) (0.118) (0.027) age: 31-35 0.280*** 0.117** 0.219*** 0.123*** 0.167*** 0.119*** 0.296*** 0.067 0.175*** (0.017) (0.045) (0.015) (0.013) (0.037) (0.009) (0.031) (0.119) (0.030) age: 36-40 0.307*** 0.098** 0.239*** 0.098*** 0.140*** 0.094*** 0.304*** -0.005 0.142*** (0.017) (0.046) (0.015) (0.014) (0.037) (0.010) (0.034) (0.121) (0.033) age: 41-45 0.307*** 0.064 0.236*** 0.065*** 0.098** 0.040*** 0.279*** -0.007 0.097*** (0.017) (0.046) (0.015) (0.016) (0.038) (0.012) (0.036) (0.122) (0.036) age: 46-50 0.318*** 0.017 0.262*** 0.064*** 0.064 0.012 0.182*** -0.012 0.087** (0.017) (0.046) (0.015) (0.020) (0.040) (0.015) (0.039) (0.125) (0.039) age: 51-55 0.363*** 0.004 0.311*** 0.095*** 0.054-0.003 0.197*** 0.041 0.003 (0.018) (0.047) (0.016) (0.026) (0.043) (0.018) (0.046) (0.130) (0.048) age: 55-60 0.379*** -0.053 0.309*** -0.092** -0.042-0.065*** 0.267*** -0.096-0.038 (0.019) (0.050) (0.019) (0.036) (0.052) (0.025) (0.065) (0.145) (0.061) feale -0.132*** -0.225*** -0.164*** -0.136*** -0.249*** -0.159*** -0.164*** -0.198*** -0.145*** (0.003) (0.009) (0.004) (0.006) (0.011) (0.005) (0.012) (0.028) (0.013) Priary and below oitted nior iddle school 0.178*** 0.121*** 0.186*** 0.158*** 0.171*** 0.150*** 0.214*** 0.187*** 0.171*** (0.009) (0.013) (0.009) (0.009) (0.012) (0.006) (0.039) (0.046) (0.029) senior iddle school 0.366*** 0.225*** 0.374*** 0.368*** 0.277*** 0.314*** 0.444*** 0.324*** 0.360*** (0.009) (0.014) (0.009) (0.011) (0.017) (0.008) (0.038) (0.048) (0.030) College and above 0.691*** 0.529*** 0.731*** 0.775*** 0.472*** 0.632*** 0.888*** 0.638*** 0.751*** (0.009) (0.023) (0.010) (0.018) (0.059) (0.019) (0.039) (0.060) (0.033) narried 0.001-0.061*** -0.034*** -0.015-0.001-0.055*** 0.003 0.153*** -0.032* (0.006) (0.020) (0.008) (0.009) (0.022) (0.007) (0.017) (0.052) (0.019) R-sqared 0.424 0.147 0.423 0.291 0.148 0.235 0.435 0.205 0.389 N 128509 25832 65371 32947 16536 45138 11482 3272 7460 Note: province dies and constants not reported. Indstry and occpation controlled. OECD 2009 25
Figre 5. Ln(incoe)-Age Profile 26 OECD 2009
Figre 6. Retrns to Edcation V.3. Oaxaca-Blinder Decoposition Reslts Using the reslts fro the OLS exercise on incoe, we trn to the Oaxaca-Blinder decoposition for rral igrants and rban residents. We start by analysing the entire saple and then to the sb-saples based on foral and inforal sector definitions. The reslts are shown in the first coln of Table 6. If sector choice is not taken into accont, nearly 100 per cent of the incoe differential between rral igrants and rban residents can be attribted to differences in their characteristics (the endowent effect). We then copare rral igrants and rban residents with respect to foral and inforal eployent. The reslts are nearly the sae. Within foral eployent, 83 per cent of the incoe differential is de to the endowent effect. For the selfeployed, it s 92 per cent, and for no-contract eployees, 160 per cent. The decoposition reslts for no-contract eployees show that if rral igrants had the sae level of han capital as their rban resident conterparts and were paid as if they are, their incoe wold not only be higher than their actal incoe, bt also higher than the incoe of rban residents. Loosely speaking, the differential between rral igrants and rban residents is ainly cased by the differences in han capital levels they bring to the rban labor arket. However, we shold be very carefl in interpreting these reslts. In particlar, or reslts OECD 2009 27
indicate that han capital levels are very iportant, bt it does not ean that the differences in retrns to han capital (both in ters of incoe and sector choice) are not iportant. Table 6. Oaxaca-Blinder Decoposition for Incoe Differential (based on OLS regression reslts) rban residents vs. rral igrants rral vs rban igrants Urban residents vs. Urban igrants recent vs. non recent igrants rral rban Indstry and occpation controlled difference 0.238-0.488-0.250 0.107 0.183 Explained (%) 101.7 59.0 16.5 53.2 50.6 Unexplained (%) -1.7 41.0 83.5 46.8 49.4 foral eployent Difference 0.252-0.604-0.351 0.121 0.179 Explained (%) 82.6 67.2 38.8 59.3 55.7 Unexplained (%) 17.4 32.8 61.2 40.7 44.3 self eployed Difference -0.096-0.236-0.333 0.077 0.086 Explained (%) 92.3 13.4 33.6 11.8 14.0 Unexplained (%) 7.7 86.6 66.4 88.2 86.0 no contract Difference 0.153-0.302-0.149 0.118 0.188 Explained (%) 160.0 56.0-40.0 59.6 46.6 Unexplained (%) -60.0 44.0 140.0 40.4 53.4 recent igrants Difference -0.448 Explained (%) 58.5 Unexplained (%) 41.5 non recent igrants Difference -0.524 Explained (%) 57.5 Unexplained (%) 42.5 28 OECD 2009
VI. DISCUSSION VI.1. Are Urban Residents the Right Reference Grop? It is sefl at this point to go back to the two criteria sed to identify rral igrants. The first is hko type, with rral igrants having rral hko stats and rban residents having an rban hko stats. The second criteria is whether srveyed individals left their registered hko location 19. It is possible that rral igrants obtain a prei for oving (de to positive self-selection into igration), bt also be discriinated against de to their rral hko stats. These two opposing effects ay prodce a close-to-zero nexplained effect as was inferred in the previos section. The advantage of or dataset is that it not only allows s to identify rral igrants bt also rban igrants. By coparing the incoe differential between these two grops, we can attept to separate ot the pre hko effect as both grops are in fact igrants. The last two colns in Table 7 report the reslts of the ltinoial logit regression reslts for sector eployent for rban igrants. Taking foral eployent as the base category, the probability of being self-eployed increases with age and the rate of increase is larger in earlier years. On the contrary, the probability of being withot a contract decreases with age. Edcation is negatively related to the probability of being in inforal eployent. The effect of higher edcation on the probability of inforal eployent is higher for rban igrants than for rral igrants. As shown in Table 8, a conterfactal prediction shows that if rban igrants were treated as rral igrants at least with respect to choice of sector, there wold be less rban igrants in the foral sector (fro 51.7 per cent to 44 per cent), and ore in self-eployed and no contract grops (fro 14.7 per cent to 16.2 per cent, and fro 33.6 per cent to 39.8 per cent respectively). OLS regression reslts on incoe deterinants are presented in the last three colns of Table 5. Highlighting soe of the reslts fro these regressions in Figres 5 and 6 (age and edcation coefficients), it becoes obvios that rban igrants have a higher retrn to age than rral igrants, at least for foral eployent and no-contract inforal eployent. Most iportantly, rban igrants have the highest retrns to edcation for each level of edcation in all three sectors. The second coln of Table 9 reports overall Oaxaca-Blinder decoposition reslts for rral and rban igrants by sector. Ignoring the differences in sector distribtion, the reglar decoposition reslts indicate that nearly 60 per cent of the incoe differentials between these 19. This twofold criteria akes or research different fro ost papers that focs on the incoe differential between two natrally divided grops (based on gender or skin color for exaple). OECD 2009 29
two grops can be explained by difference in their characteristics and the reaining 40 per cent is de to differences in skill-prices. However we find significant heterogeneity across sectors. In the foral sector, differences in endowents can explain 67 per cent of the incoe differential whereas in the no-contract grop (self-eployent grop) the share goes down to 56 per cent (13.4 per cent). It is interesting that the explained and nexplained shares of the incoe differential between rral and rban igrants do not change ch if we consider only recent igrants or only the non-recent igrants. By coparing rral igrants with rban igrants instead of rban residents, we obtain qite different reslts. Rral igrants fare worse than the rban igrants not only becase they have low levels of han capital bt also becase they are treated differently de to their rral hko stats 20. These reslts indicate that igrants do receive a prei for igrating, and this holds tre both for rral and rban igrants. VI.2. Preis for Migrants: Urban Residents verss Urban Migrants To evalate the existence and agnitde of a igrant prei, we copare rban igrants with rban residents. Both grops have rban hko stats with the only observable difference being that the first grop is ade p of igrants. Hence if we find that rban igrants have higher incoe levels not only becase of their higher han capital levels, bt also becase of the different skill-prices, this ight provide evidence of a prei for igration. The siple Oaxaca-Blinder decoposition in coln 3 of Table 6 indicates that the prei not only exists bt that it is also iportant. The siple overall decoposition shows that 83.5 per cent of the incoe difference is nexplained. For the decopositions by sectors, the nexplained shares are 61.2 per cent, 66.4 per cent, and 140 per cent for the foral sector, selfeployed, and no-contract eployees respectively. Althogh rban igrants have large preis in incoe deterination within each sector, they see to detain less of an advantage to soe extent in the choice of sector. The reslts shown in Table 4 indicate that if treated as rban residents in sector allocation, rban igrants shold figre ore proinently in the foral sector (57 per cent instead of 52 per cent), and fewer in the self-eployed grop (10.5 per cent instead of 14 per cent) 21. This iplies that if rban igrants were treated as rban residents in ters of sector eployent, they wold oreover have higher incoe levels. This negative effect is overcoe by the large igrationrelated preis igrants receive in ters of earnings. 20. Clearly, agricltral hko stats has varios iplications. For exaple, we se age as a proxy for potential experience. Bt what really atters is rban labor arket experience, which is not available in the data. For rral igrants, age is definitely a bad proxy. 21. This also explains the negative percentages we obtained for the inter-sectoral differences in the Brown decoposition analysis (See Table A3 in the second appendix). 30 OECD 2009
VI.3. Does Dration of Migration Matter? We can think of the igrant prei referenced above as a net average effect for igrants as a whole. However, preis for igrants ay differ sbstantially and the heterogeneity ay coe not only fro the type of hko they have, bt also their igration dration or otherwise their rban labor arket experience. For instance, the Harris and Todaro (1970) fraework sggests that igrants ay first enter the inforal labor arket while they wait and perhaps gather experience for an opportnity at a foral sector ob. Unfortnately, it is very difficlt to find a coparable proxy for rban labor arket experience for all three grops. In order to evalate the assiilation effect, we apply or ethod to decopose the incoe differentials between recent and non-recent igrants. For both rral and rban igrants, the dration of their igration episode is iportant. More than 50 per cent of igrants (rral and rban) have less than 3 years of local rban labor arket experience (we call this grop recent igrants). It sees tre for both rral and rban igrants, that the self-eployed tend to have longer igration dration, and the no contract eployees tend to be ore recent igrants. To see the effect of dration ore clearly, we split rral igrants and rban igrants into recent and non-recent sb-grops (Table 2). To get a better nderstanding of the tie diension of igration and obtain decoposition reslts, we estiate ltinoial logit odels for eployent sector for recent and non-recent igrants separately and OLS regressions on incoe deterinants for different sectors. The ltinoial logit reslts for sector of eployent are reported in Table 7 and OLS incoe deterination reslts in Table 8. Recent and non-recent igrants show different patterns both in ters of incoe deterination and sector choice, and this is the case for both rral and rban igrants. The Oaxaca-Blinder decoposition reslts, reported in Table 6, show that for rral igrants (coln 5), 47 per cent of the differential between recent and non-recent igrants is nexplained. In the foral sector and for the no-contract eployee grop however, the nexplained shares are lower than the overall percentages, which are arond 40 per cent. For the self-eployed, 88 per cent of the difference between recent and non-recent igrants is nexplained. This eans that the assiilation effect is ore evident for the self-eployed. This is an expected reslt as setting p a bsiness in an rban area reqires financial and social capital which ay take tie to acclate. The case for rban igrants is siilar. For sectoral distribtion analysis we trn to or conterfactal predictions (Table 4). If recent igrants were treated as non-recent igrants, there wold be ore recent igrants in foral sectors or self-eployent, and this is tre for both rral and rban igrants. Nevertheless the difference between the actal and conterfactal sector distribtions is not very large, and this is closely related to the fact that actal sector distribtions of recent and nonrecent igrants are siilar. As a reslt, a Brown decoposition analysis shows that the share of the inter-sectoral difference in the total difference is relatively low: -6.3 per cent and 6.8 per cent for rral and rban igrants, respectively (Table A3 in Appendix 2). Most of the difference is de to intra-sectoral differentials, 58.6 per cent and 46.1 per cent of which can be explained by differences in characteristics. OECD 2009 31
Table 7. Marginal Effects for Mltinoial Logit Reslts (Migration Dration) Rral igrants Urban igrants Non Recent Recent Non Recent Recent Self-eployed No contract Self-eployed No contract Self-eployed No contract Self-eployed No contract dp/dx dp/dx dp/dx dp/dx dp/dx dp/dx dp/dx dp/dx s.e. s.e. s.e. s.e. s.e. s.e. s.e. s.e. Age 16-20 oitted age: 21-25 0.024-0.072*** 0.044*** -0.029*** 0.065-0.063 0.050*** -0.056*** (0.018) (0.015) (0.005) (0.007) (134.035) (35.751) (0.014) (0.017) age: 26-30 0.082*** -0.106*** 0.063*** -0.041*** 0.120-0.119 0.078*** -0.118*** (0.019) (0.016) (0.006) (0.009) (247.253) (64.895) (0.015) (0.020) age: 31-35 0.119*** -0.126*** 0.072*** -0.051*** 0.157-0.129 0.091*** -0.117*** (0.019) (0.016) (0.006) (0.010) (324.047) (93.749) (0.016) (0.023) age: 36-40 0.147*** -0.151*** 0.082*** -0.050*** 0.167-0.123 0.098*** -0.139*** (0.019) (0.017) (0.006) (0.011) (345.079) (104.560) (0.016) (0.025) age: 41-45 0.163*** -0.137*** 0.087*** -0.062*** 0.165-0.171 0.100*** -0.159*** (0.019) (0.018) (0.006) (0.013) (341.120) (87.419) (0.017) (0.028) age: 46-50 0.179*** -0.174*** 0.098*** -0.066*** 0.154-0.121 0.092*** -0.118*** (0.020) (0.020) (0.007) (0.017) (318.297) (93.774) (0.017) (0.031) age: 51-55 0.158*** -0.167*** 0.087*** -0.018 0.162-0.158 0.088*** -0.184*** (0.021) (0.023) (0.008) (0.021) (335.117) (89.152) (0.019) (0.037) age: 55-60 0.163*** -0.198*** 0.051*** -0.000 0.206-0.094 0.046* -0.036 (0.024) (0.030) (0.011) (0.029) (426.133) (147.299) (0.026) (0.051) Priary and below oitted Jnior iddle school -0.041*** -0.071*** -0.021*** -0.097*** -0.031-0.047-0.027*** -0.084*** (0.005) (0.007) (0.003) (0.007) (64.458) (41.861) (0.010) (0.027) Senior iddle school -0.081*** -0.143*** -0.035*** -0.189*** -0.087-0.137-0.062*** -0.206*** (0.007) (0.009) (0.004) (0.008) (179.533) (118.145) (0.010) (0.027) College and above -0.210*** -0.153*** -0.054*** -0.276*** -0.177-0.238-0.122*** -0.334*** (0.022) (0.021) (0.011) (0.018) (366.213) (228.109) (0.011) (0.028) arit -0.158*** 0.122*** -0.090*** 0.053*** -0.085 0.098-0.069*** 0.055*** (0.009) (0.009) (0.004) (0.008) (174.859) (41.425) (0.008) (0.015) r3-0.065*** 0.013** -0.037*** -0.013*** -0.018 0.014-0.011** 0.041*** (0.004) (0.005) (0.002) (0.004) (36.313) (10.786) (0.005) (0.010) Observations 40080 54541 10407 11807 Note: the base category is foral eployent. region dies not reported. *** p<0.01, ** p<0.05, * p<0.1 32 OECD 2009
Table 8. OLS Regression Reslts for Migration Dration: Dependent Variable=Log (Horly Incoe) rral igrants rban residents Non recent recent Non recent recent foral Self ep No contr foral Self ep No contr foral Self ep No contr foral Self ep No contr Age 16-20 oitted age: 21-25 0.064*** 0.245*** 0.118*** 0.065*** 0.137*** 0.110*** 0.023-0.25 0.047 0.116*** 0.022 0.129*** (0.024) (0.072) (0.016) (0.009) (0.041) (0.007) (0.069) (0.320) (0.058) (0.029) (0.126) (0.025) age: 26-30 0.109*** 0.206*** 0.138*** 0.114*** 0.100** 0.119*** 0.113-0.245 0.091 0.313*** 0.117 0.182*** (0.027) (0.072) (0.018) (0.013) (0.043) (0.011) (0.070) (0.313) (0.060) (0.033) (0.129) (0.032) age: 31-35 0.097*** 0.192*** 0.098*** 0.111*** 0.137*** 0.110*** 0.150** -0.274 0.069 0.311*** 0.077 0.164*** (0.028) (0.072) (0.019) (0.015) (0.044) (0.012) (0.072) (0.314) (0.064) (0.039) (0.131) (0.036) age: 36-40 0.060** 0.170** 0.070*** 0.098*** 0.099** 0.092*** 0.145** -0.339 0.009 0.329*** -0.022 0.161*** (0.029) (0.072) (0.020) (0.017) (0.045) (0.013) (0.073) (0.314) (0.066) (0.043) (0.134) (0.041) age: 41-45 0.021 0.1 0.017 0.073*** 0.100** 0.032** 0.101-0.393-0.003 0.340*** 0.062 0.080* (0.031) (0.073) (0.021) (0.021) (0.047) (0.015) (0.075) (0.315) (0.070) (0.046) (0.137) (0.047) age: 46-50 -0.012 0.07-0.039 0.108*** 0.06 0.028 0.046-0.457-0.021 0.186*** 0.114 0.077 (0.036) (0.075) (0.025) (0.026) (0.050) (0.019) (0.078) (0.318) (0.073) (0.052) (0.143) (0.051) age: 51-55 0.032 0.09-0.070** 0.127*** 0.003 0.014 0.038-0.359-0.07 0.225*** 0.113-0.05 (0.043) (0.078) (0.030) (0.034) (0.057) (0.023) (0.086) (0.320) (0.085) (0.062) (0.151) (0.063) age: 55-60 -0.167*** -0.016-0.094** -0.045-0.101-0.083*** 0.214** -0.591* -0.077 0.149 0.307-0.092 (0.057) (0.084) (0.041) (0.048) (0.078) (0.032) (0.106) (0.327) (0.101) (0.091) (0.209) (0.080) Feale -0.192*** -0.283*** -0.207*** -0.094*** -0.200*** -0.129*** -0.201*** -0.240*** -0.163*** -0.130*** -0.158*** -0.124*** (0.010) (0.015) (0.008) (0.007) (0.016) (0.005) (0.018) (0.040) (0.022) (0.016) (0.039) (0.017) Priary and below oitted Jnior iddle school 0.153*** 0.171*** 0.156*** 0.155*** 0.165*** 0.135*** 0.182*** 0.072 0.169*** 0.254*** 0.373*** 0.173*** (0.014) (0.016) (0.009) (0.011) (0.017) (0.007) (0.053) (0.062) (0.045) (0.057) (0.069) (0.039) Senior iddle school 0.376*** 0.277*** 0.324*** 0.352*** 0.271*** 0.298*** 0.413*** 0.196*** 0.356*** 0.483*** 0.503*** 0.360*** (0.017) (0.023) (0.013) (0.013) (0.026) (0.010) (0.053) (0.066) (0.046) (0.056) (0.071) (0.039) College and above 0.791*** 0.412*** 0.724*** 0.744*** 0.499*** 0.544*** 0.874*** 0.558*** 0.766*** 0.918*** 0.755*** 0.745*** (0.029) (0.087) (0.031) (0.023) (0.080) (0.025) (0.054) (0.085) (0.051) (0.057) (0.086) (0.043) narried -0.013-0.001-0.075*** -0.002 0.011-0.032*** -0.011 0.128-0.053* 0.039* 0.181*** -0.009 (0.016) (0.035) (0.012) (0.011) (0.029) (0.009) (0.024) (0.082) (0.030) (0.023) (0.065) (0.025) R-sqared 0.314 0.151 0.253 0.263 0.148 0.218 0.438 0.241 0.405 0.428 0.193 0.37 N 13819 9312 16949 19128 7224 28189 5570 1736 3101 5912 1536 4359 Note: province dies and constants not reported. Indstry and occpation controlled. OECD 2009 33
VI.4. Decoposing Sectoral Distribtion Differentials Or analysis has not, ntil now, revealed any aor role played by sectoral segentation. The largest inter-sectoral share we fond in the Brown decoposition analysis was 39.5 per cent (rban residents vs. rral igrants in Table A3 and described in Appendix 2) with all other coparisons less than 10 per cent. We arge that these reslts do not iply that discriination against igrants in ters of sectoral choice is niportant. Even a sall degree of discriination can be qite significant, given the sbstantial differences that exist in ters of social secrity coverage, working conditions and pay between the foral and the inforal sectors. To show this we estiate linear probability odels (not reported) and se the reslts to calclate Oaxaca-Blinder decopositions based on sector choice. The decoposition reslts are reported in Table 9. In the first two colns, we consider the broad definition of inforal eployent, inclding both self-eployed and no-contract eployees and copare rral igrants with rban residents. It is clear that the sectoral distribtion differential is large. The fraction of inforal eployent for rral igrants is 33.7 percentage points higher than that for rban residents. The decoposition reslt shows that only 17 per cent of the difference can be explained by differences in characteristics, while 83 per cent reains nexplained. This is in contrast to the reslts we had derived both with the Oaxaca-Blinder and Brown decoposition reslts for incoe differentials, which indicated the doinant role of the endowent effect and a inor role played by sector segentation. The reslts here show that there is a significant share of differential cased by discriination in ters of sector choice, which eans igrants ay be even worse off in ters of social secrity coverage and working conditions, even conditional on individal characteristics. Interestingly (and likely by coincidence) the nexplained share of the sector distribtion differential between rban igrants and rban residents is also 83 per cent. However, taking into accont that the overall sector distribtion differential for these two grops is saller than that between rral igrants and rban residents (16.6 as opposed to 33.7 percentage points), the extent of discriination against rban igrants is saller. It shold be noted that rban igrants enoy the highest average incoe level, and they are better (not worse) off conditional on their characteristics in ters of incoe. However, the reslts here show that they are still discriinated in ters of sector choice. What follows natrally is to copare rral igrants and rban igrants. The reslts are also as expected. As these two grops are both igrants, differences in characteristics can explain a larger share (arond 50 per cent) of the sector-choice differential. Still, half of the differential is de to nexplained factors. We interpret this as discriination against rral hko stats. We also consider recent and non-recent igrants separately bt the reslts do not change ch. The final decoposition exercise for these two grops is to copare recent and non-recent igrants for each grop separately. The reslts indicate a large share of discriination against recent igrants. As differentials between recent and non-recent igrants are not very large, the decoposition reslts is of inor iportance for s. However, becase we se a broad definition 34 OECD 2009
of inforal eployent, the sall differentials ay be cased by coposition change within the broad inforal eployent definition. In the next for colns of Table 9, we consider self-eployent and no-contract eployees separately. The general pattern is siilar to the one fond in the first two colns, bt with a slight variation. There are at least two points worth entioning. First, de to the sall fraction of self-eployent in all three grops, the difference between grops is relatively sall, especially when we copare non-recent rral and rban igrants. Second, when we copare recent and non-recent igrants, there are larger sector distribtion changes than nder the broad inforal definition. For both rral and rban igrants, the fraction of self-eployed is larger for non-recent igrants than for recent igrants, and the fraction of no-contract eployees is saller. Table 9. Oaxaca-Blinder Decoposition Reslts for Sector Choice (based on Linear Probability Model) Different definitions of inforal eployent Self-eployed + no contract Self-eployed No contract rral igrants vs rban residents difference 0.337 100 0.087 100 0.250 100 explained 0.059 17 0.034 39 0.025 10 nexplained 0.278 83 0.053 61 0.226 90 rban igrants vs rban residents difference 0.166 100 0.057 100 0.109 100 explained 0.028 17-0.005-9 0.033 30 nexplained 0.138 83 0.062 109 0.076 70 rban igrants vs rral igrants Total difference -0.171 100-0.030 100-0.141 100 explained -0.087 51-0.007 23-0.080 57 nexplained -0.084 49-0.023 77-0.061 43 recent igrants difference -0.189 100-0.065 100-0.124 100 explained -0.099 53-0.038 59-0.061 49 nexplained -0.090 47-0.027 41-0.063 51 non recent igrants difference -0.153 100-0.005 100-0.149 100 explained -0.072 47 0.013-267 -0.084 57 nexplained -0.082 53-0.017 367-0.064 43 recent vs non recent igrants rral igrants difference 0.014 100-0.087 100 0.101 100 explained -0.007-52 -0.059 67 0.052 51 nexplained 0.021 152-0.028 33 0.050 49 rban igrants difference 0.049 100-0.027 100 0.076 100 explained 0.014 29-0.024 91 0.039 51 nexplained 0.035 71-0.002 9 0.037 49 OECD 2009 35
VII. CONCLUSIONS In this paper we se a nationally representative saple to investigate how rral igrants fare in the rban labor arket in China. This paper is different fro the existing literatre in several iportant ways. First, or data is nationally representative and is better at captring a representative saple of rral igrants. Second, we distingish aong different grops of igrants instead of only coparing rral igrants and rban residents. In particlar, we add rban igrants in or analysis, and this allows s to separate the rral hko effect fro the igrant prei effect. Third, we consider sectoral segentation in ters of foral and inforal eployent, which is an iportant diension of labor arket otcoes of rral igrants in Chinese cities. The ain finding in this paper is in stark difference to those in the existing literatre. When we copare rral igrants with rban residents, nearly 100 per cent of the difference can be explained by differences in characteristics, which eans the skill-price effect is alost negligible. The coparison between rral igrants and rban igrants gives a different pictre however; 40 per cent of the incoe differential is nstified. Given the fact that hko stats is the only difference between these two grops conditional on other personal characteristics (and igrant stats), this exercise is ore appropriate for s to detect discriination against rral hko stats. By coparing rban igrants with rban residents, we find significant igrant preis. We conectre that the no discriination effect we derive fro the coparison between rral igrants and rban residents ay be the reslt of a conterbalance between the discriination effect against rral hko and the prei effect for igrants. Another reslt, based on the Brown decoposition analysis, is that sector segentation plays a inor role in explaining the incoe differential. Sectoral segentation, however, ay be iportant in ters of social secrity and working conditions and therefore the sector distribtion differential in its own right is worth stdying (and this is overlooked by any researchers). A Oaxaca-Blinder decoposition indicates that igrants (both rral and rban) are discriinated in sector choice (ore than 80 per cent is nexplained). The extent of discriination is larger for rral igrants indicating a frther discriination against rral hko stats taken the agnitde of the differential. The decoposition for differences between sector distribtions copleents the incoe decoposition in a very iportant way. There are of corse liitations in the approach taken in this paper. A first difficlty has to do with choosing the appropriate reference grop. We take a step forward by sing rban igrants as an additional reference grop. However, this ethod is also not withot probles as rral igrants and rban igrants ay be different in nobservable characteristics other than their hko stats, even conditional on the characteristics we do control for. Another difficlty is 36 OECD 2009
easreent error which is especially salient when we are coparing rban residents (igrants) with rral igrants. Finally, age (even potential experience) is a poor proxy for rban labor arket experience for rral igrants. The policy iplications of or reslts are clear. In ters of incoe, rral igrants enoy igrant preis and sffer discriination at the sae tie. Generally speaking, however, the reason they ay be worse off when copared to rban residents is de to their lower levels of han capital. Increasing edcation levels of rral igrants, and providing the with training and relevant rban labor arket skills will help increase their earning opportnities. As both rral and rban igrants face nfair treatent in sector choice, reforing the labor arket, notably reoving barriers to obility between sectors, ay help increase foral eployent and well-being. Discriination against rral hko stats is evident in or stdy and ideally a coplete cancelation of the syste wold eventally lead to a ore eqal treatent on the labor arket. This, however, ay not be practically feasible and in fact reains a central focs of debate in China. What is perhaps ore pressing is to ensre that igrants have access to basic social services, even in cases where they are eployed inforally. Presently those withot sch coverage face exorbitant costs for health services and in sending their children to rban schools. Frtherore, for igrants who systeatically ove for obs, obtaining rban social secrity coverage is ftile as social secrity systes are for the ost part non-portable and expensive. In ters of schooling and childcare, any igrants leave their children back hoe in the rral parts of China, in effect ptting ore pressre on hosehold ebers left-behind and adding to the already existing social strain cased by igration. OECD 2009 37
APPENDIX 1 Table A1 saple size calclated national poplation Poplation fro CSY 2 585 481 1 292 740 500 1 307 560 000 ale 50.15 50.15 51.53 feale 49.85 49.85 48.47 Table A2 for types of igrants Why did yo leave yor hko registration location? rral-rral rral-rban rban-rral rban-rban For ob or bsiness 49.45 60.81 18.53 19.65 Job change 0.4 0.61 7.92 6.22 Eployed 0.11 0.16 2.66 1.52 Training 1.26 4.05 1.55 4.01 Move hose (change living place) 2.01 2.75 4.12 22.5 Marriage 18.77 5.18 11.83 9.14 Move with relatives 11.8 15.32 16.38 15.41 Move to live with relatives or friends 10.5 6.74 14.55 8.97 Teporary hko change 1.06 0.37 9.62 4.08 On a bsiness trip 0.37 0.37 0.26 0.24 Others 4.28 3.64 12.58 8.25 Weighted saple size 38 724 159 497 12 545 116 790 Unweighted saple size 28 495 132 840 7 646 87 315 Note: categories 2 and 3 are ainly for those working in the pblic sector. 38 OECD 2009
Figre A1. National Representativeness of the Saple OECD 2009 39