World Development Vol. 71, pp. 79–93, 2015� 2013 Elsevier Ltd. All rights reserved.

0305-750X/$ - see front matter

www.elsevier.com/locate/worlddevhttp://dx.doi.org/10.1016/j.worlddev.2013.11.002

Rural–Urban Migration, Household Vulnerability,

and Welfare in Vietnam

LOC DUC NGUYEN, KATHARINA RAABE and ULRIKE GROTE *

Leibniz University Hannover, Germany

Summary. — This paper investigates the interaction of migration, vulnerability to poverty, and welfare of rural households in threeprovinces in Central Vietnam. It addresses three questions. (1) To what extent do shocks motivate rural household members to moveto urban areas?, (2) Are migrants in the new urban settings better off in terms of working conditions and quality of life?, and (3) Whatis the effect of migration on rural household’s welfare and vulnerability to poverty? The analysis uses panel data of 2200 households fromrural Vietnam covering the period 2007–2010, and a tracking survey of 299 migrants from 2010. The empirical evidence from a probitmodel shows that migration, especially migration for employment, is a livelihood support strategy for households exposed to agriculturaland economic shocks. Migration for education is more likely observed among households with higher human capital and being finan-cially better off. Nevertheless, the probability of migration decreases with the employment opportunity in the village. Migrants perceivethemselves to be better off at the place of destination, but income losses from shocks of their rural households may reduce their employ-ment quality. The results from difference-in-difference specifications with propensity score matching techniques suggest that migrationhas positive income growth effects, and that these effects are more pronounced in provinces with fewer job opportunities. These effectshelp not only migrant households moving out of poverty, but it also improves the poverty situation in rural areas.� 2013 Elsevier Ltd. All rights reserved.

Key words — migration, vulnerability to poverty, difference-in-difference, propensity score matching, Vietnam

* The research was financed by the German Research Foundation (DFG)

under the umbrella of the research project ‘Impact of shocks on vulner-

ability to poverty: Consequences for development of emerging Southeast

Asian economies’ (DFG FOR 756). The authors would like to thank

Mulubrhan Amare, Lena Hohfeld, Dean Jolliffe, and Hermann Waibel as

well as two anonymous reviewers for their valuable comments.

1. INTRODUCTION

Rural households in developing countries face several typesof unpredictable events threatening their livelihoods. Amongothers, these include economic shocks like price fluctuations,and natural disasters like droughts or floods. Due to the ab-sence of basic social safety nets and sufficient and comprehen-sive insurance schemes, rural people, especially the poor, oftenhave to cope with the effects of shocks and associated risks ontheir own. Specifically, they diversify their livelihoods, save forprecautionary purposes, or join mutual support networks(Dercon, 2002).

Migration is one livelihood strategy that households in eco-logically vulnerable communities pursue to diversify their in-come sources and to overcome the adverse welfare effects ofsocial, economic, and institutional constraints in their placesof origin (Ezra, 2001; Tongruksawattana, Schmidt, & Waibel,2010). Migration increases household income and smoothesincome fluctuations mainly through remittances (Stark &Bloom, 1985).

Like in other developing countries, rural households in Viet-nam, whose livelihoods deeply depend on agriculture, facesubstantial income variability because of climate change andprice fluctuations in the context of rapid liberalization and re-form processes. Moreover, the gap between government sup-port and the loss through damage have increased over time(see Appendix Figure 1). Therefore, rural residents smooththeir consumption through savings, mutual support, or privatetransfers including remittances (Newman & Wainwright, 2011;Phung & Waibel, 2009).

Over the past decade, Vietnam has experienced an exponen-tial increase in the movement of people both within and acrossits borders. By meeting the demand for labor created by indus-trial development and foreign direct investments following the“Doi Moi” reforms, migration plays an important role in Viet-nam’s economy, and contributes to poverty reduction (Cu,2005, chap. 5; Dang, Tacoli, & Hoang, 2003). Migration has

79

become a strategy for households in rural areas of Vietnamto reduce income fluctuation. However, a substantial shareof individuals and households who migrated in search for bet-ter income opportunities could not improve their living condi-tions. These problems arise from the lack of knowledge andexperience when living in modern cities. Additionally, theinadequate implementation of labor laws (Le, Tran, &Nguyen, 2011), or the limited access to affordable health careservices, among others (United Nations Population Fund(UNFPA), 2010) made the migrants become vulnerable intheir destinations. Moreover, the 2008 global economic crisisaggravated the vulnerability of migrants. Some migrantsstopped sending remittances or returned to their householdsat the place of origin (Oxfam & VASS, 2010). These challengesare likely to affect the motivation of migration and the welfareof rural households.

The main objective of this paper is to assess the success ofmigration as a livelihood support and risk coping strategy ofrural households who sent some of their household membersto urban areas during the period 2008–2010. However, notonly original households in the rural areas are analyzed, butalso migrants from the urban areas. Therefore, an employ-ment quality index (EQI) is developed that deploys informa-tion on a variety of indicator variables of working and livingconditions to quantify the success of individual migrants inthe city. The effectiveness of migration as livelihood supportstrategy of rural households is determined by comparing thechanging welfare outcomes of migrant and non-migrant

80 WORLD DEVELOPMENT

households by means of the propensity score matchingprocedure.

Past studies on migration in Vietnam are based on House-hold Living Standard Surveys which are not appropriate tostudy migration as only officially registered migrants are in-cluded in the sample. This current paper builds on (1) a paneldata set of about 2200 rural households from three provincesin Vietnam and (2) a tracking migrant survey of 299 migrants.Given this unique data set, the results provide a new perspec-tive on migration in Vietnam.

The paper is structured as follows. Section 2 reviews themain theories of migration and related empirical evidence.Section 3 describes the data set and methodology applied toassess the success of migration in rural and urban Vietnam.Section 4 presents and discusses the empirical results of the pa-nel data analysis and the migrant survey analysis. Finally, Sec-tion 5 concludes.

2. LITERATURE REVIEW

The economic literature on migration assumes that individ-uals or households rationally consider various locations andchoose the place that maximizes the expected gains frommigration. The expected gains from migration depend on alarge number of factors such as personal characteristics andexperiences, social networks, wealth, or reduced vulnerabilityto poverty. Different theories and models have advanced to ac-count for their importance.

In his “laws of migration”, Ravenstein (1885, 1889) linkedmigration patterns to conditions of labor force surpluses anddeficits, with people moving from surplus labor areas to deficitlabor areas in order to improve their living conditions. He alsodeveloped the idea of the “pull” and “push” factors in order toexplain the forces driving migration. Pull factors are social,economic, political, or environmental incentives at the placeof destination, such as job opportunities, better education,and living conditions. Push factors are incentives at the placeof origin that force people to out-migrate. Specific factors in-clude insufficient job and employment opportunities, insecu-rity regarding political, social, or economic conditions, orthe loss of wealth (Lee, 1966). Other classical migration mod-els exist from Sjaastad (1962) and have been further developedby Harris and Todaro (1970) and Mincer (1978).

Stark and Bloom (1985) developed a fundamentally differenttheory of migration called the New Economics of LaborMigration (NELM). According to this approach, migrationdecisions are joint family decisions, although this does not im-ply that household members move jointly. In fact, householdsdecide on the migration of few household members so as tomaximize and smooth household income and ensure sustain-able livelihoods through the spatial/local diversification ofhousehold resources such as labor. Migration is thus a strategyfor managing and minimizing risk to household income andsurvival.

One of the important contributions of NELM is the link be-tween migration decision and risks. The costs of migration(including those associated with risks and opportunities) areshared among household members, thus creating a co-insur-ance system between migrant and non-migrant householdmembers. The co-insurance system involves (1) family supportto the migrant in the case of need (risk) in the destination areaand (2) migrant support to the family via remittances to facil-itate risk coping at the place of origin. In addition to risk andwage differentials, models also link migration to social capital,the existence of functioning social networks among migrants,

non-migrants and return migrants, and migration institutions(Massey, 1990).

Migration studies also depart from the NELM approach toidentify the factors behind migration and the well-being of mi-grants. For instance, Agesa and Kim (2001) used data fromKenya to identify the determinants of migration decisions.Their results show that migration is relatively more likelyamong workers facing a positive urban to rural earnings differ-ence, suggesting that skilled workers self-select to migrate tourban areas. Giesbert (2007) reports evidence from WesternKenya according to which the propensity to migrate dependson education and migrant networks, but not on householdwealth. Ezra (2001) finds that individuals belonging to eco-nomically poor households in ecologically vulnerable commu-nities have a higher propensity to out-migrate than those fromless vulnerable regions in Ethiopia.

Several empirical studies investigated the impact of migra-tion on rural households’ welfare but with ambiguous results.Evidence from Thailand suggests that migration reduces in-come inequality mainly through changes in the distributionof productive assets (Garip, 2010). Another study from Thai-land reveals that poor rural households tend to produce poormigrants, which could be one of the reasons for the continuousexistence of a wide rural–urban gap in welfare (Amare,Hohfeld, & Waibel, 2012). Similarly, Azam and Gubert(2006) report findings from Mali and Senegal according towhich remittances cause rural households to reduce their workeffort, which reduces the effectiveness of migration as a pov-erty reduction instrument. Fuente (2010) uses household paneldata from Mexico for the period 1998–2000 to assess howlikely households with a high level of vulnerability to povertyreceive remittances. However, contrary to the expectation,households with a higher level of vulnerability to poverty havea lower probability to receive remittances.

The working and living conditions of migrants in the desti-nation places are also analyzed in several studies. Shah (2000)uses four indicators to assess the degree of success of migrantworkers in Kuwait. The indicator variables include (1) objec-tive measures such as the migrant’s salary and job permissionand (2) subjective measures regarding job satisfaction. Amongothers, migrant workers are asked to indicate whether the jobis the same or better than expected prior to migration. The re-sults show that human and social capital are the main factorscontributing to the success of migrants. Akay, Bargain, andZimmermann (2011) use general health questionnaires to iden-tify the factors that affect the subjective well-being of ruralworkers, migrants, and local urban workers in China. Theirstudy finds that the well-being of migrants positively dependson the length of the migration period, the quality of workingconditions, and the existence of community ties. Amare et al.(2012) calculate employment quality indices for migrants inThailand. They confirm that human capital is a major factor.Along with government support, human capital can improveliving and working conditions of migrants in the city.

In Vietnam, most empirical studies on migration are basedon data from three surveys of the General Statistical Officein Vietnam (GSO): (i) the population census which is con-ducted every 10 years, (ii) the Vietnam Living Standard Sur-vey (VLSS) from 1993–94 and 1997–98 which was replacedby the Vietnam Household Living Standard Survey (VHLSS)published biannually from 2002 onwards, and (iii) the migra-tion survey from 10 provinces conducted in 2004 in coopera-tion with the United Nations Population Fund (UNFPA)(GSO & UNFPA, 2005).

Based on the population census of 1999 and 2009, GSO(2011) analyzed the patterns and trends of internal migration,

RURAL–URBAN MIGRATION, HOUSEHOLD VULNERABILITY, AND WELFARE IN VIETNAM 81

defining a migrant as a person who has changed the livingplace in the last 5 years. The results show that rural–urbanmigration has become an important trend with an estimatedrate of 8.9% in 2009, directed especially to central cities orprovinces with high levels of industrialization. Migrantshave become younger and have better living conditions, butmigration also widens the socioeconomic gaps between ruraland urban areas. GSO and UNFPA (2005) analyzed thesocioeconomic conditions of migrants in the cities based onthe 2004 migration survey (see UNFPA, 2010). The results re-veal that the majority of migrants move because of economicreasons. This is also supported by Dang et al. (2003) andNiimi, Pham, and Reilly (2009), both arguing that rural out-migrants shifted to urban areas to benefit from increased eco-nomic opportunities. Although the dominant share of mi-grants feels better off and sends remittances to their families,migrants face many problems in the destination area. One ofthe biggest problems concerns Vietnam’s complex householdregistration system. The system controls and monitors thechanges of people’s residence in Vietnam by classifying theminto different residential categories, each being associated withcertain rights and obligations. The registration system may re-duce the benefits from migration by constraining the access ofinternal migrants to basic public services such as education orhealth services in the absence of registration (Niimi et al.,2009). Furthermore, even though incomes increase aftermigration, the average incomes of migrants are still lower thanthe incomes of the local residents in the destination areas. Inaddition, many migrants are employed on a temporary basiswithout formal labor contracts or social protection.

Using the VLSS 1993–94 and 1997–98, De Brauw andHarigaya (2007) find that migrant households’ expenditurelevels exceed those of non-migrant households by approxi-mately 5%. Nguyen, Tran, Nguyen, and Oostendorp (2008)use the VHLSS 2002–2004 panel data set to assess the effectof migration on household expenditures and inequality inrural areas. The study shows that migration is a highly selec-tive process, strongly influenced by household and villagecharacteristics. Furthermore, migration positively affectshousehold expenditures, while at the same time increasingthe degree of income inequality in rural areas. In a follow-up study with the VHLSS 2004–2006 panel data, Nguyen,Den Berg, and Lensink (2009) confirm the positive householdexpenditure effects of migration, but also report a slight de-cline in poverty and inequality. Pincus and Sender (2008)voice concern about studying migration based on the datasets of the VLSS and VHLSS, since only the officially regis-tered households being for at least 6 months in the surveylocation are covered. Migrants without a permanent resi-dence status are ignored in the sample. Based on their ownsmall survey of workers in VHLSS enumeration areas, theyconfirm that the number of unregistered workers is consider-able so that the official data sets underestimate the actualnumber of migrants. Further shortcomings of the official sur-veys are that it is not possible to link the migrants with theiroriginal households so that the impact of remittances on thewelfare of rural households cannot be assessed. Moreover,temporary migrants are not captured so that the results arelikely to underestimate any internal migration trends.

The present study is motivated by the existing ambiguous re-sults based on unreliable data sets. It seeks to provide newempirical evidence regarding the extent to which migration isa successful livelihood support strategy for rural householdsin Vietnam. It uses a unique data set linking rural householdswith their migrating members in urban cities.

The remainder of this paper addresses three research ques-tions. First, to what extent do shocks motivate rural house-hold members to move to urban areas? Second, are migrantsin the new urban settings better off in terms of working condi-tions and quality of life? Third, what is the effect of migrationon rural household’s welfare and vulnerability to poverty?

3. DATA AND METHODOLOGY

(a) Data

The study uses a panel data set that contains information ona random sample of 2200 households 1 from the three prov-inces Dak Lak, Thua Thien Hue, and Ha Tinh in Vietnam.Household data were collected in 2007, 2008, and 2010. Thequestionnaires for the household survey covered a broad setof questions regarding the socio-demographic and economicconditions of the sampled households. Among others, specificinterest was with the migration experience of the householdand the household members, with the composition of the in-come source portfolio, with borrowing and lending patterns,and especially, the exposure to demographic, social, economic,and agricultural shocks. In addition, village heads were inter-viewed in 2007 and 2010 in order to collect general informa-tion about their village such as village population,employment structure, infrastructure characteristics, and re-source use patterns.

For analyzing the motivations of migration and evaluatingthe impacts of migration on household welfare, a migrantand a migrant household need to be defined. In this paper, amigrant is a household member having lived outside of the vil-lage for at least 1 month in the year 2008 and/or 2010. A mi-grant household is a household that has at least one migrant inthat period. Since a household member may have migrated outalready in 2007, the household could have benefitted fromremittances having an effect on the “per capita income” vari-able. To avoid such endogeneity problems, households withmigrants in 2007 are dropped from the sample. 2 The remain-ing sub-data set consists of about 1711 households. Of these,443 are migrant households with 890 migrants, including bothrural–rural and rural–urban migrants. Since this paper focuseson rural–urban migration, the 158 rural–rural migrant house-holds are also dropped from the data set. The remaining 285rural–urban migrant households are used for the analysis.Almost 60% of those migrant households have at least onemember who migrated in search of a job, and 33% migratedfor educational reasons. The remaining 8% indicated otherreasons like “followed the family” or “went to help anotherhousehold”.

In addition to the household survey, a migration survey of299 migrants is the basis for this study. The migration surveyis a tracking survey in which the respondents are migranthousehold members of the surveyed rural households. The mi-grant status of the household members was determined in atwo-step procedure. First, information from the rural house-hold survey in 2008 was used to construct a list of migranthousehold members. Second, the migrant status of the house-hold members was confirmed during the rural household sur-vey in 2010. The respective information was then directlytransmitted to the migrant survey team, which implementedthe migrant survey at the same time in Ho Chi Minh Cityand its two surrounding and highly industrialized provincesDong Nai and Binh Duong which have the highest rates ofnet migration (UNFPA, 2010).

82 WORLD DEVELOPMENT

Because the majority of migrants work in the informal sec-tor and frequently change their contact details, the migrantsurvey team succeeded in interviewing 299 out of 600 potentialmigrant respondents. Questions addressed the migration pro-cess, the shocks, risks, and socio-economic situation of mi-grants in the destination area, and the type and nature oflinks between the migrants and their rural households. Amongthe 299 interviewed migrants, there are 233 wage-employed, 15self-employed, and 51 other migrants with irregular jobs. Tomeasure the employment quality index to quantify the successof these migrants in the city, only wage-employed migrants areexplored.

(b) Methodology

This paper estimates three basic models, one for each re-search question. The first two models involve the estimationof probit and linear regression specifications. The third modelis estimated by means of the propensity score matching anddifference-in-difference procedure in order to control for unob-servables between migrant and non-migrant households.

(i) Model 1: The migration decision modelIn a first step, a non-linear probability model is estimated

that links the household migration status in 2010 and 2008to household and village characteristics in 2007, respectively.The model is defined as in Eqn. (1).

PrðDij;2010;2008 ¼ 1Þ ¼ FðXij;2007;Zj;2007;FEProvÞ: ð1ÞThe dependent variable captures the probability that house-hold i in village j is a migrant household in 2008 and/or2010. Here, the binary dummy Dij,2010,2008 equals one if house-hold i in village j had at least one migrant household memberin 2008 and/or 2010 and zero otherwise.

The probability of being a migrant household in 2008 and/or 2010 is a function of observable household characteristics Xin 2007 (i.e., Xij,2007). The vector specifically contains informa-tion on (1) socio-demographic household factors (HHCij,2007)such as the gender and education of the household head andhousehold members and the share of dependent householdmembers, (2) household wealth (WIij,2007) such as per capitaincome, household indebtedness, income sources, or landholdings, and (3) the household’s exposure to demographic,social, economic, or agricultural shocks (SEij,2007). Vector Xof household i in village j is accordingly defined as

Xij;2007 ¼ ½HHCij;2007;WIij;2007; SEij;2007�: ð2ÞIn addition to household characteristics, the probability ofbeing a migrant household is also linked to a set of observablevillage characteristics Z. Being equal to

Zj;2007 ¼ ½RCj;2007;Distj;2007�; ð3Þthe vector of village characteristics includes information onroad conditions (RCj,2007) and the distance to the districtheadquarters (Distj,2007) in 2007. Finally, province fixed effectsFEProv are included to capture the migration effects of prov-ince-specific unobservables.

Since households decide that their members migrate out forseveral reasons, the motivation for each purpose may bedifferent. Therefore, we estimate three alternative equationspresenting household migration decisions (i) for all types ofreasons; (ii) for employment, and (iii) for education.

(ii) Model 2: Measuring employment quality of migrantsOne of the main objectives of this paper is to identify the

factors affecting the livelihood of migrants in the urban

destination area. Closely related to Shah (2000) and Amareet al. (2012), we assess the livelihood situation of migrant mby using k indicator variables Ik of the migrant’s employmentand living conditions, which we then combine into a compositeindex—the employment quality index (EQI). The EQI com-bines information from a set of subjective indicator variables,which indicate whether migrants perceive their (1) income tobe stable, (2) working conditions to have improved since theirlast job, and (3) living conditions to have improved since theyhave left the rural area. The EQI also includes objective indi-cator variables, specifying whether migrants have (1) accumu-lated savings, (2) above average income levels, and (3) awritten employment contract.

Each indicator variable Ik equals one if the underlying con-dition holds in 2010 and zero otherwise. The EQI of migrant mis then defined as an unweighted average of the subjective andobjective dummy variables, defined as

EQIm;2010 ¼1

6

X6

k¼1

Imk;2010: ð4Þ

The EQI assumes values between zero and one, with employ-ment quality being better for larger values.

Given the specification of the EQI, we define a linear modelthat links the EQI to a set of (1) individual characteristics ofmigrant m (MCm,2010) and (2) characteristics of the house-hold i that migrant m belongs to (HHCim,2010). The character-istics of migrants include gender, age, education level, thelength of migration period, the type of job, job characteris-tics, and the way of getting the job. Household characteristicscover the loss from shocks that a household might have facedin 2010 like income loss and asset loss due to shocks (SEim,2010).Variables related to ethnicity and whether a householdbelongs to a political or social organization are also addedto the model. The EQI model is then given as in Eqn. (5).

EQIm;2010 ¼ FðMCm;2010;HHCim;2010; SEim;2010Þ: ð5Þ

(iii) Model 3: Evaluating the impact of migration on rural house-hold welfare

The third model seeks to quantify the effect of migration onrural household welfare. First, the average treatment effect onthe treated is identified (Heckman & Navarro-Lozano, 2004),where treatment refers to the household’s migration status andthe treatment effect arises by comparing outcome of house-holds with migrants against that of comparable householdswithout migrants. It specifically follows model (6).

ATT ¼ EðY1i �Y0iÞjD ¼ 1Þ ¼ EðY1ijD ¼ 1Þ � EðY0ijD ¼ 1Þ:ð6Þ

ATT abbreviates the average treatment effect on the treated,which measures the impact of migration on the outcome of mi-grant households. D is a binary dummy variable that equalsone if the household has at least one migrant and zero, other-wise. Y1i and Y0i denote the outcome of household i with andwithout migrants, respectively.

It is impossible to compute the outcome of the migranthousehold in case no one migrated (Y0i|D = 1), as thisvariable is unobserved. In order to solve this problem,this paper employs the method of Propensity ScoreMatching (PSM). Under the assumption of conditionalindependence, this method pairs the set of observablecharacteristics of migrant households to some group of“comparable” non-migrant households by creating propensityscores (Rosenbaum & Rubin, 1983). The outcome of migrant

RURAL–URBAN MIGRATION, HOUSEHOLD VULNERABILITY, AND WELFARE IN VIETNAM 83

households that they would have if they had not migrated isestimated from the outcome of the comparison groups ofnon-migrant households. Specifically, the outcome of themigrant household in case no one migrated (Y0i|D = 1) isapproximated by the observable value of the outcome of the“comparable” non-migrant household E(Y0i|D = 0).

In this paper, we use the migration decision model (1) toestimate the propensity score, and the Nearest-Neighborhoodand Kernel matching methods to estimate the outcome thatmigrant households would have had in case none of theirhousehold members had migrated.

However, the standard PSM method controls for selectionon observable variables, but cannot account for unobservedvariables and their simultaneous effect on the probability ofmigration and the outcome variable (Rosenbaum & Rubin,1983). We therefore use the method of difference-in-differenceswith PSM to eliminate the effect of unobserved (time-invari-ant) variables on the outcome variable (Smith & Todd,2005). This approach also helps to address the endogeneityproblem that usually precludes the identification of the out-come effects of migration.

Given this, the impact of migration on household outcomegrowth is rewritten as

ATT¼½Y12010�Y1

2007jX2007;D¼1��½Y02010�Y0

2007jX2007;D¼0�;ð7Þ

where ATT denotes the average treatment effect on the trea-ted. The subscripts 2007 and 2010 denote the baseline andthe year of migration, respectively. To assess the variabilityof matching estimators, we present bootstrap standard errorsbased on 1000 replications.

(iv) OutcomesIn this paper, the outcomes are household income growth,

vulnerability to poverty, and poverty and inequality indicesfor indicating the change in household welfare.

+Income growth: It is the observable change in income ofmigrant households from 2007 to 2010.

+Vulnerability to poverty (VTP): It is measured as theprobability that a household (or an individual), whether cur-rently poor or not, would find itself poor in the future. It mea-sures the household consumption with respect to theconsumption poverty line (Chaudhuri, 2003; Dercon, 2002;Klasen, Lechtenfeld, & Povel, 2013). The impact of migrationon the poverty indices of migrant households is calculated asfollows:

DVTP ¼ PðVTP12010jD ¼ 1Þ � PðVTP0

2010jD ¼ 1Þ; ð8Þwhere VTP1

2010 is the estimated vulnerability to poverty of mi-grant households in 2010, and VTP0

2010 the counterfactual vul-nerability to poverty which was estimated previously by PSMprocedure.

+Poverty and inequality indices: These include three Foster–Greer–Thorbecke (FGT) poverty indices and the Theil’s L andTheil’s T inequality indices (Haughton & Khandker, 2009).The impact of migration on the poverty indices of the migra-tion household are calculated as follows:

DP ¼ PðE12010jD ¼ 1Þ � PðE0

2010jD ¼ 1Þ; ð9Þwhere E1

2010 is the observed per capita expenditure of the mi-grant household, and E0

2010 is the counterfactual per capitaexpenditure which was estimated previously.

We also measure the impact of migration on poverty andinequality for the total rural population capturing also synergy

effects of migration on the poverty and inequality in the vil-lages:

DI ¼ IðZ12010Þ � IðZ0

2010Þ; ð10Þwhere IðZ1

2010Þ is calculated directly for the entire data sample,

and IðZ02010Þ is measured through the estimated counterfactual

expenditure per capita of migrant households and observedper capita expenditure of non-migrant households.

4. RESULTS AND DISCUSSION

This section presents the econometric results of the study.Section (a) presents the evidence on the factors driving migra-tion. Section (b) discusses the employment quality index andidentifies the factors that influence employment quality. Sec-tion (c) presents the evidence on the effect of migration on rur-al household welfare.

(a) Explaining the household’s migration decision

This section discusses the factors influencing the household’smigration decision by the probit model (1). Table 1 shows theresults. Summary statistics of the model’s variables are pre-sented in Appendix Table 6.

In general, the coefficient estimates show the expected signs,and the significance properties are robust to alternative modelspecifications. 3 The evidence suggests that migration is thepreferred livelihood support strategy of households havingexperienced agricultural and economic shocks. This is espe-cially true for migration for employment, implying that anagricultural or economic shock forces a rural household tosend out its member(s) to find a job in urban areas to compen-sate for the loss in household’s income. Agricultural shocks in-clude floodings, droughts, crop pests or livestock diseases,whereas economic shocks relate to job loss, collapse of busi-ness, strong increase of input prices, or strong decrease of out-put prices. Demographic and social shocks like e.g., illness ordeath of a household member, theft or conflict with neighborsin the village are in contrast not likely to incite a migratory re-sponse within a household. Interestingly, these shock variablesare not statistically significant in the case of migration for edu-cation (Appendix Table 6).

Considering the socio-demographic household characteris-tics, the propensity of migration in general, and migrationfor employment in specific, significantly increases with theage of the household head. Thus, the older the householdhead, the more likely his or her member(s) migrate out to findemployment. This variable becomes statistically insignificantin case of migration for education. Moreover, the propensityof migration significantly decreases with the relative numberof household dependents. This finding suggests that the pro-pensity to migrate is higher in households that are character-ized by a larger share of productive laborers.

Consistent with the predictions of migration theory, migra-tion depends on the level of human and social capital. It ap-pears that the probability of migration increases with theshare of household members with completed secondary educa-tion. It is more pronounced in case of migration for educationwhen both, the share of household members with completedsecondary education and the share with professional trainingincrease. These variables become statistically insignificant incase of migration for employment. This means that migrationfor employment does not necessarily require a higher educa-tion level. Looking at the descriptives of the sample, we findthat indeed 44% of all migrants lack secondary school

Table 1. Determinants of household migration (probit regression)

Variables All migration decision Job migrationdecision

Education migration decision

(1) (2) (3)

Household characteristics in 2007

HH experienced demographic shocks (1-yes, 0-no) 0.108 0.143 �0.155(0.084) (0.100) (0.143)

HH experienced social shocks (1-yes, 0-no) 0.096 �0.178 0.252(0.180) (0.317) (0.342)

HH experienced agriculture shocks (1-yes, 0-no) 0.146* 0.220** 0.249(0.087) (0.108) (0.158)

HH experienced economic shocks (1-yes, 0-no) 0.273* 0.392** 0.322(0.159) (0.190) (0.216)

Female headed HH (1-yes, 0-no) �0.146 �0.105 �0.101(0.122) (0.143) (0.205)

Age of HH head (years) 0.016*** 0.019*** 0.003(0.003) (0.004) (0.007)

Share of HH members w/completed secondary school 0.178*** 0.054 0.390***

(0.044) (0.054) (0.064)

Share of HH members w/completed high school or professional education 0.041 0.010 0.191***

(0.039) (0.046) (0.062)

HH members belong to political or social organization (1-yes, 0-no) 0.187* 0.164 0.281(0.109) (0.139) (0.199)

Dependency ratio� �1.553*** �1.966*** �2.190***

(0.216) (0.263) (0.430)

Log of monthly HH per capita income (PPP $ in 2005) 0.042 0.002 0.120*

(0.046) (0.048) (0.072)

HH engaged in off-farm activities (1-yes, 0-no) �0.073 �0.050 �0.262*

(0.087) (0.098) (0.147)

Log of land per capita (hectare) �0.091*** �0.133*** 0.031(0.034) (0.041) (0.060)

HH is indebted (1-yes, 0-no) 0.032 0.243** 0.047(0.098) (0.124) (0.186)

Village characteristics in 2007

Village road condition (1-good condition, 0-bad condition) �0.143 �0.151 �0.112(0.112) (0.128) (0.181)

Log distance from village to district headquarters (km) �0.209*** �0.219*** �0.102(0.053) (0.058) (0.079)

Ha Tinh province (1-yes, 0-no) 0.535*** 0.517*** 0.271(0.142) (0.183) (0.200)

Thua Thien Hue province (1-yes, 0-no) 0.253 0.407** 0.062(0.155) (0.181) (0.228)

Constant �1.836*** �2.225*** �2.520***

(0.310) (0.329) (0.538)

Number of observations 1432 1326 1245Wald chi2(18) 231,09 177.32 144.28Prob > chi2 0.00 0.00 0.00Pseudo R2 0.16 0.17 0.27Log pseudolikelihood �582.86 �387.04 �184.47

Standard errors are clustered at the village level.Source: Own calculations based on the DFG Rural Village Survey 2007 and the DFG Rural Household Survey 2007, 2008 and 2010.* Denotes statistical significance at 10%.** Denotes statistical significance at 5%.*** Denotes statistical significance at 1%.� Dependents are household members below the age of 10 years and above the age of 65 years. The dependency ratio is the number of dependents relativeto the total number of household members.

84 WORLD DEVELOPMENT

RURAL–URBAN MIGRATION, HOUSEHOLD VULNERABILITY, AND WELFARE IN VIETNAM 85

qualifications. 4 Households with membership in political orsocial organizations also display a larger propensity to migratein general. However, the results turn statistically insignificantin case of migration for employment and education as well,indicating that social capital is not really important for themigration decision.

As regards the measures of household wealth, householdswith larger per capita income levels in 2007 are more likelyto be engaged in migration for education. They may considermigration as an investment strategy. For migration foremployment, the income variable becomes statistically insig-nificant. When looking at the level of income of the house-holds in the data set, it can be seen that households withmembers migrating for education have a relatively higher levelof per capita income as compared to households with mem-bers migrating for finding a job. Furthermore, a quarter ofall rural households with migrants for employment are poorin 2007. These observations support the idea that if house-holds support migration out of desperation, then this only ap-plies to households with migrants who look for employment.

Surprisingly, the variable “Household engaged in off-farmactivities” is only statistically significant in case of migrationfor education. Therefore, the more the households are engagedin off-farm activities, the less likely they allow their membersto migrate out for education. In general, households withmore land per capita are less likely to be migrant householdsbecause they need more laborers for agricultural production.This is especially true for the case of migration for employ-ment but not for the case of migration for education.

Considering indebtedness, migration for employment ismore likely observed among households with open financialdebt positions. Especially rural households thus send a mi-grant out expecting that remittances will facilitate the repay-ment of their outstanding debts. In this case, migration foremployment is more likely a desperation strategy than aninvestment strategy. 5

Overall, the descriptives show that three quarters of all mi-grant households have only one migrant, while 18% have twomigrants, and 7% have three or more migrants. About half ofall migrants send remittances, no matter how many migrants arural household has. But interestingly, of those migrants whosend remittances, a quarter is better off, while three quartersare not better off. Again, this evidence confirms that migrationis likely to be a desperation strategy.

Gender does not have any significant effect on the migrationdecision. This is true even when separating into migration foremployment and migration for schooling. In general, we findthat about 60% of all migrants are male and 40% female. Halfof the migrants for education are male and around 60% of themigrants searching for jobs are male. The average age of a mi-grant is 24 years with female migrants being on average2 years younger than male migrants. With respect to educa-tion, the male migrants have on average one additional yearof schooling, but overall, 44% of all migrants lack secondaryschool qualifications. This is true for male and female mi-grants alike. 15% of the female migrants are from poor house-holds, whereas this share amounts to 11% for the malemigrants.

Finally, the household’s decision to migrate for education isnot influenced by any village characteristics like remotenessand road infrastructure development. However, householdsnear to the district headquarters have a higher propensity tosend members to migrate for employment. Moreover, migra-tion does depend on unobserved provincial effects. Specifi-cally, the evidence suggests that households from the HaTinh and Thua Thien Hue provinces are more likely to be

engaged in migration for employment in comparison to house-holds from the Dak Lak province. 6 Consistent with the argu-ment of UNFPA (2010), this may reflect cross-provincedifferences in economic development and cross-province dis-similarities in employment prospects and income opportuni-ties. Nevertheless, the results do not show these effects in thecase of migration for education.

(b) Assessing the well-being of migrants in the destination areas

Subjective and objective indicator variables are used to mea-sure the working and living conditions of migrants in urbanareas. These indicator variables are separately reported for mi-grants who are working in industry, production, and in theservice sector; and for migrants according to the number ofmigration years.

According to the subjective indicator variables, the majorityof migrants in industry and production as well as in the servicesector (1) perceive their income to be stable and (2) reportimprovements in working and living conditions. Maybereflecting the effect of exaggerated expectations, migrants withless than one year of migration experience are least likely toperceive the outcomes of migration positively. The share ofsatisfied migrants is highest among those with 3–5 yearsof migration experience. For migration periods in excess of5 years, migrants are again less likely to positively evaluateworking and living conditions.

Considering the objective indicator variables, migrants inthe service sector are more likely to have savings than migrantsin industry and production. This sectoral effect is consistentwith the observation regarding the higher average daily wagein the service sector (see Appendix Table 7).

In general, only roughly every second migrant has a writtenemployment contract. The probability of having a written con-tract increases with the years of migration experience and isslightly higher for migrants in industry and production. Awritten employment contract influences working and livingconditions as it provides migrants’ access to social protectionprograms (see also GSO & UNFPA, 2005; Oxfam & VASS,2010; UNFPA, 2010).

Approximately half of the surveyed migrants report incomelevels below the sample average. Pointing to the greater vul-nerability of recent migrants, below average income levelsare reported by three-quarter of all migrants with less thanone year of migration experience. The share of migrants withbelow average income decreases with the number of migrationyears, being lowest for the group of migrants with 7–10 yearsof migration experience (21%).

Finally, the indicators from Table 2 are used to construct acomposite employment quality index (EQI) for the sample ofwage-employed migrants, which account for 78% of all mi-grants in the destination areas. In order to account for possibledifferences in the relative importance of subjective and objec-tive indicator variables, we present three employment qualityindices according to Eqn. (4). These reflect the unweightedaverage of (1) the three objective migration assessment criteria(hereafter referred to as objective EQI), (2) the three subjectivemigration assessment criteria (hereafter referred to as subjec-tive EQI), and (3) the set of both objective and subjectiveassessment criteria (hereafter referred to as aggregate EQI).The use of subjective and objective indicator variables yieldsdifferent conclusions regarding the average share of migrantsthat is satisfied with the living and working conditions (seeAppendix Figure 2). Here, the distribution of the subjectiveEQI is skewed to the left pointing to a larger share of satisfiedmigrants. It thus appears that it is the environment in general

Table 2. Migrants’ working and living conditions by occupation and length of migration period (% of total)

Industry/production sector Public/private service sector Length of migration period (years)

<1 1–3 3–5 5–7 7–10 >10

Subjective assessments*

Income is stable 71.7 70.6 57.6 71.7 86.7 78.9 61.8 60.0Working conditions have improved 68.1 73.5 69.7 60.4 75.0 73.7 70.6 73.3Living conditions have improved 85.8 86.0 69.7 83.0 91.7 86.8 85.3 100.0

Average 75.2 76.7 65.7 71.7 84.5 79.8 72.6 77.8

Objective assessments

Migrant has accumulated savings 31.9 50.0 45.5 22.6 45.0 36.8 52.9 73.3Migrants with above average income� 52.2 51.5 24.2 37.7 60.0 68.4 79.4 73.3Migrants have written employment contract 55.8 47.8 48.5 45.3 51.7 68.4 55.9 66.7

Average 46.6 49.8 39.4 35.2 52.2 57.9 62.7 71.1

Note: Regardless of the choice of indicator variable, the number of observations equals 233.Source: Own calculations based on the DFG Migrant Survey 2010.* The subjective indicator variables reflect perceptions as reported by migrants.� Average income is computed across all migrants with employment in either the industry and production or service sector.

Table 3. Explaining the degree of migrants’ employment quality

Variables (1) (2) (3)

Year of schooling (years) 0.020*** 0.019*** 0.021***

(0.005) (0.005) (0.005)

Gender (1-female; 0-male) 0.160*** 0.158*** 0.141***

(0.039) (0.040) (0.041)

Debt (1-yes; 0-no) �0.008 �0.002 �0.006(0.046) (0.046) (0.047)

Age (years) 0.018*** 0.024***

(0.005) (0.004)

Length of migration period (years) 0.033* 0.071***

(0.018) (0.014)

Job in service sector (1-yes, 0-no) �0.030 �0.029 �0.020(0.041) (0.041) (0.042)

Pay to get a job (1-yes; 0-no) �0.040 �0.032 �0.062(0.055) (0.054) (0.055)

How to find a job (1-introducing from friends or relatives; 0-others) 0.062 0.061 0.061(0.042) (0.043) (0.043)

Household characteristics in 2010

Ethnic (Kinh & Hoa = 1, others = 0) 0.018 0.046 0.045(0.095) (0.098) (0.090)

HH members belong to political or social organization (1-yes, 0-no) �0.023 �0.026 �0.030(0.155) (0.156) (0.161)

Income loss from shocks (2005 $ PPP) �0.000* �0.000** �0.000(0.000) (0.000) (0.000)

Constant �0.282 �0.345 0.016(0.216) (0.218) (0.197)

Number of observations 228 228 228Prob > F 0.000 0.000 0.000Adjusted R2 0.211 0.202 0.171

Note: The estimates are derived from an OLS regression specification.Standard errors are adjusted for heteroscedasticity.Source: Own calculations based on the DFG Migrant Survey 2010 and DFG Rural Household Survey 2010.* Denotes the statistical significance at the 10% level.** Denotes the statistical significance at the 5% level.*** Denotes the statistical significance at the 1% level.

86 WORLD DEVELOPMENT

RURAL–URBAN MIGRATION, HOUSEHOLD VULNERABILITY, AND WELFARE IN VIETNAM 87

and factors influencing living conditions in particular thatinfluence the extent to which migrants perceive their stay in ur-ban areas to be a success. The aggregate index is also slightlyskewed to the left, suggesting that a relatively larger share ofmigrants is very satisfied with the living and working condi-tions in the urban destination area. The objective EQI, how-ever, is normally distributed and is thus most suitable toidentify the factors that affect the EQI for the sample ofwage-employed migrants (model 2). The descriptive variablesused in this model are presented in Appendix Table 8.

Table 3 summarizes the estimation results based on threealternative sets of independent variables. In the first alterna-tive set, all variables are included. Since the variables “lengthof the migration period” and “age” are correlated, one of thetwo variables is removed from the second but included in thethird alternative set, and vice versa. The results show that onthe one hand, migrants being female, better-educated, older,and with longer migration periods are more likely to reporta higher objective EQI. This means that they are more likelyto have accumulated savings, above average income, and awritten employment contract. On the other hand, indebtednessand the way they got the job do not statistically affect their

Table 4. Difference-in-difference estimates of the im

Outcome variables Treatment

General migration

Income growth (Kernel) 0.56Income growth (Nearest-Neighborhood) 0.55

Migration for employment

Income growth (Kernel) 0.56Income growth (Nearest-Neighborhood) 0.55

By province categoriesa (Kernel)

Ha Tinh province 0.88Thua Thien Hue province 0.40Dak Lak province 0.17

Standard errors (in parentheses) are bootstrapped using 1000 replications of tmatching technique based on propensity score matching.Source: Own calculations based on the DFG Rural Household Survey 2007, 2a These are calculated under all migration.* Denotes the statistical significance at the 10% level.

Table 5. Difference-in-difference estimates of the impact of migrat

Outcome variable Treatment

Migrant household

Vulnerability to poverty (VTP) 0.21Head count index (P0) 0.21Poverty gap index (P1) 0.05Poverty severity (P2) 0.02

Whole sample

Vulnerability to poverty (VTP) 0.29Head count index (P0) 0.32Poverty gap index (P1) 0.09Poverty severity (P2) 0.04Theil’s L 0.18Theil’s T 0.19

ns, not significant. Standard errors (in parentheses) are bootstrapped usingdifference-in-differences matching technique based on propensity score matchiSource: Own calculations based on the DFG Rural Household Survey 2007, 2* Denotes the statistical significance at the 10% level.

objective EQI. Having to pay for a job does not seem to guar-antee a higher objective EQI as indicated by the negative sign.

The gender effect reflects the fact that women have more sta-ble and predictable working relations. Indeed, around 60% ofthe female migrants have a job with a written contract, ascompared to 40% of the male migrants. The gender effectmay also reflect the different spending behavior. In fact, thedescriptive information confirms that female migrants gener-ally have higher savings than their male counterparts. The gen-der effect, however, does not reflect above average incomes, aswomen are paid lower salaries as compared to men.

In general, the types of jobs being conducted by female andmale migrants in the city are very diverse. 54% of all womenwork in industry/production, mainly in weaving, but also intextile and electronics factories that are more likely to providewritten contracts and stable employment. The remaining 46%women work in the service sector in jobs like as accountant inbanks (14%), tailor (7%), waiter, sales person, hair dresser, orcleaner/housemaid. In comparison, only 37% of all men workin industry/production, including the weaving sector (12%)and to a very small extent electronics and textiles factories.Otherwise, men are more likely to be employed in the service

pact of migration on household income growth

Control Difference in average outcome ATT

0.36 0.20*(0.09)0.28 0.27*(0.11)

0.37 0.19*(0.09)0.43 0.12(0.12)

0.54 0.34*(0.14)0.28 0.12(0.14)0.08 0.09(0.15)

he sample. Estimates are derived by means of the difference-in-differences

008 and 2010.

ion on vulnerability to poverty, poverty, and inequality in 2010

Control Difference in average outcome ATT

0.22 �0.01(0.02)0.26 �0.05*(0.03)0.07 �0.02*(0.01)0.03 �0.01*(0.00)

0.30 �0.01(0.00)0.33 �0.01*(0.01)0.1 �0.01*(0.00)0.04 �0.00*(0.00)0.17 ns0.17 ns

1000 replications of the sample. Estimates are derived by means of theng.008 and 2010.

88 WORLD DEVELOPMENT

sector as security guard, technician, electrician, plumber, or assales person.

As regards the set of household characteristics, the objectiveEQI is lower for employed migrants belonging to householdswith higher income losses due to shocks in 2010. These mi-grants are more likely to work in unstable working relations,without any contracts, and in lower paid jobs not allowingthem to accumulate any savings. The findings also reveal thatethnicity does not have any significant influence on the objec-tive EQI. This result is also not surprising as 96% of all inter-viewed migrants belong to the majority ethnic group Kinh orHoa (Vietnamese or Chinese).

(c) Effect of migration on rural household welfare

The evidence suggests that (1) migration is partly attribut-able to household-specific economic factors, and (2) migrantsdo not fare equally well in terms of living and working condi-tions in urban destination areas. Against this background, thissection presents the results on the impact of migration on ruralhousehold welfare from difference-in-difference estimationswith propensity score matching. The study also estimatedthe impact of migrants’ employment quality on rural house-hold welfare, but the result is not statistically significant.

Summarized in Table 4, the results show a large positive andsignificant effect of migration on rural households’ incomegrowth at least during the period 2007–2010. Dependent onthe matching method (Kernel or Nearest-Neighborhood),households’ income increased by 20–27%. Similarly, migrationfor employment also positively and significantly affects incomegrowth of original households (Kernel matching method).

Consistent with our expectations, the income growth effectof migration is particularly pronounced for households fromthe Ha Tinh province, while no significant effects exist forhouseholds from the Thua Thien Hue and Dak Lak provinces.This result points to the importance of migration as a sourceof income growth in structurally weak provinces with pooremployment and job opportunities (cf. UNFPA, 2010 andGSO, 2011). 7

Table 5 presents the impact of migration on vulnerability topoverty, poverty, and inequality in 2010. The vulnerability topoverty and poverty indices are measured depending on themigration status of a migrant household and for the wholedata sample, while Theil’s L and Theil’s T inequality indexesare only estimated for the whole data sample.

The results show that migration does not make rural house-holds less vulnerable, but it can help a household moving outof poverty and also reducing the depth and severity of pov-erty. Similarly, with respect to the whole population, migra-tion also has a positive impact on the poverty status (interms of the poverty head count index and poverty gap index)at the place of origin compared to the case no one migrated.The findings are thus consistent with previous studies ofNguyen et al. (2008, 2009) showing that migration can helprural households having higher income and lifting peopleout of poverty. However, while they show a slight increasein inequality at the place of origin, we find some synergyeffects meaning that also non-migrant households in thevillages indirectly benefit.

Finally, when comparing the poverty status of migranthouseholds in 2007 and 2010 in a poverty transition matrix(see Appendix Table 9), we find that migration shifted 57 mi-grant households out of poverty and pushed 17 householdsback into poverty. For 38 households, the poverty status didnot change in that period. Similarly, migration also made 38migrant households—from a total of 68 vulnerable households

in 2007—less vulnerable, while only 12 households becamemore vulnerable in 2010.

5. CONCLUSIONS

This paper investigated the interaction of shocks, the vulner-ability to poverty and welfare of rural households and rural–urban migrants in Vietnam. It provides responses to the threequestions: (1) To what extent do shocks motivate rural house-hold members to move to urban areas?, (2) Are migrants in thenew urban settings better off in terms of working conditionsand quality of life?, and (3) What is the effect of migrationon rural household’s welfare and vulnerability to poverty?The analysis is based on (1) a rural household panel data setfrom three provinces in Vietnam (Ha Tinh, Thua ThienHue, and Dak Lak) and (2) a tracking migrant survey fromthe Binh Duong and Dong Nai provinces as well as the HoChi Minh City. To avoid endogeneity problems, a sub-dataset of 1553 households was created by dropping all pre-2008migrant households and all rural–rural migrant households.

To explore the first question on the motivation of migration,three probit models are estimated, namely for: (i) the wholesample of migrant households, (ii) just for migration foremployment, and for (iii) migration for education. 92% ofall migrant households indicate employment opportunityand education to be the two main pull factors. At the sametime, the empirical evidence suggests that rural–urban migra-tion for employment is a livelihood support strategy for house-holds coping with agricultural and economic shocks likedroughts, floods or loss of job, or with financial debts. Migra-tion for education is more likely to be observed among house-holds with a large share of members having higher educationlevels and that are financially better off and do not face anyshocks. They consider migration rather as an investment strat-egy. Rural households who are engaged in off-farm activitiestry to involve their family members in these activities. Migra-tion, then, is less likely to happen, especially with respect tomigration for education. Similarly, the probability of migra-tion for employment decreases for households with large land-holdings, or being engaged in agricultural production since therural households seem to prefer using them as their own labor-ers as compared to hiring laborers. In general, the probabilityof migration decreases with the employment opportunity inthe village, as evidenced e.g., for the Dak Lak province withplenty of jobs available in the coffee sector, as compared tothe Ha Tinh and Thua Thien Hue provinces. This finding sug-gests that encouraging rural labor market development can re-duce rural–urban migration. Gender does not seem to haveany significant effect on the migration decision.

With respect to the second research question, the descriptiveand econometric results show that migration coincides withgeneral improvements in the living and working conditionsof wage-employed migrants in the urban destination area.Nevertheless, explicit training and wage standards might beuseful instruments for still improving migrant’s situation inthe urban areas. Migrants being female, better-educated, olderand, with longer migration periods are more likely to report ahigher objective employment quality index. However, house-holds’ income losses due to shocks may negatively affect a mi-grant’s situation in the city. Thus, savings schemes could be auseful instrument for smoothing income fluctuations e.g., fromshocks across all groups of migrants as well as across migranthouseholds in rural areas.

As regards the third question on the effect of migration on rur-al household’s welfare, the results from difference-in-difference

RURAL–URBAN MIGRATION, HOUSEHOLD VULNERABILITY, AND WELFARE IN VIETNAM 89

specifications with propensity score matching techniques sug-gest that migrant households directly benefit from migration,especially migration for employment, through positiveincome growth effects. These effects help not only migranthouseholds moving out of poverty, but they also improvethe poverty situation in rural areas in general. Thus, also non-migrant households seem to indirectly benefit from remittances

of migrant households. The econometric results do not findany significant effects of migration in terms of improvingthe vulnerability to poverty and inequality. The descriptivefindings identify a quarter of the households to be vulnerablein 2007; half of them moved out of poverty in 2010. Furtherresearch should focus on the dynamics of vulnerability topoverty.

NOTES

1. See Hardeweg and Waibel (2009) for details on the data collectionprocedure.

2. Any remaining potential endogeneity problems are considered to befairly small due to the following reasons: (i) 50% of all households havemigrants of only up to three years, and only 20% of all householdshave members who migrated for a longer time period (>7 years). Thus,the share of households with migrants prior to 2007 that may havebenefitted from remittances is relatively small. (ii) It might take atransition period of a few years for a migrant to settle down in the newplace of destination so that remittances only arise after such atransition period. (iii) In our data set, only half of all migrants haveactually sent remittances.

3. Alternative specifications control for the correlation of differenttypes of shocks and of wealth variables. For instance, the per capitaincome correlates with the type of household activities, so thatmodel (1) was separately estimated for each of these wealthvariables.

4. With respect to the rural-rural migrants, the equivalent share amountsto 60%, showing that a higher percentage of migrants with belowsecondary school education move to rural areas as compared to rural-urban migrants.

5. With respect to the descriptives, we find from our household data setthat 75% of all rural-urban migrants are from rural households withfinancial debts. While three quarters of these migrate for finding a job, onequarter migrates in order to study. But even 45% of these latter migrantsfor education send remittances. Thus, some of the indebted householdsmay consider migration for education also as an investment strategy, eventhough they may face financial restrictions.

6. Being located in the Central Highland region of Vietnam, the Dak Lakprovince provides plenty of jobs in the coffee sector.

7. The number of out-migrants from the Ha Tinh province is signifi-cantly higher than that from the Thua Thien Hue and Dak Lak provinces(GSO, 2011).

REFERENCES

Agesa, R. U., & Kim, S. (2001). Rural to urban migration as a householddecision: Evidence from Kenya. Review of Development Economics,5(1), 60–75.

Akay, A., Bargain, O., & Zimmermann, K. F. (2011). Relative concerns ofrural-to-urban migrants in China. Journal of Economic Behavior andOrganization, 81, 421–441.

Amare, M., Hohfeld, L., & Waibel, H. (2012). Rural–urban migration andemployment quality: A case study from Thailand. Asian DevelopmentReview, 29(1), 58–80.

Azam, J. P., & Gubert, F. (2006). Migrants’ remittances and thehousehold in Africa: A review of evidence. Journal of AfricanEconomies, 15(2), 426–462.

Chaudhuri, S. (2003). Assessing vulnerability to poverty: Concepts,empirical methods and illustrative examples. Department of Economics,Columbia University, <http://info.worldbank.org/etools/docs/library/97185/Keny_0304/Ke_0304/vulnerability-assessment.pdf> AccessedMay 2011.

Cu, C. L. (2005). Rural to urban migration in Vietnam. In H. T. Ha, & S.Sakata (Eds.). Impact of socio-economic changes on the livelihoods ofpeople living in poverty in Vietnam (ASDP 71, pp. 115–143). Institute ofDeveloping Economies, Japan External Trade Organization.

Dang, N. A., Tacoli, C., & Hoang, X. T. (2003). Migration in Vietnam: Areview of information on current trends and patterns, and their policyimplications. Paper presented at the Regional Conference on Migration,Development and Pro-Poor Policy Choices in Asia, Dhaka, Bangladesh.

De Brauw, A., & Harigaya, T. (2007). Seasonal migration and improvingliving standards in Vietnam. American Journal of Agricultural Eco-nomics, 89(2), 430–447.

Dercon, S. (2002). Income risk, coping strategies and safety nets, Discussionpaper, 22. Helsinki: World Institute for Development EconomicsResearch, United Nations University (UNU-WIDER).

Ezra, M. (2001). Ecological degradation, rural poverty, and migration inEthiopia: A contextual analysis, Working Paper, 149. New York:Population Council, Policy Research Division.

Fuente, A. D. L. (2010). Remittances and vulnerability to poverty in ruralMexico. Journal of World Development, 38(6), 828–839.

Garip, F. (2010). The impact of migration and remittance on wealthaccumulation and distribution in rural Thailand, Working paper,0010. Weatherhead Center for International Affairs, HarvardUniversity.

Giesbert, L. (2007). Seeking opportunities: Migration as an incomediversification strategy of household in Kakamega district in Kenya,Working paper, 58. German Institute of Global and Area Studies,<http://www.giga-hamburg.de/dl/download.php?d=/content/publika-tionen/pdf/wp58_giesbert.pdf> Accessed August 2011.

GSO (2011). Migration and urbanisation in Viet Nam: Patterns, trends anddifferentials. Findings from the 2009 population and housing census.General Statistics Office, <http://vietnam.unfpa.org/webdav/site/viet-nam/shared/Census%20publications/7_Monograph-Migration-Urban-ization.pdf> Accessed April 2012.

GSO & UNFPA. (2005). The 2004 Vietnam migration survey: Internalmigration and related life course events. United Nations PopulationFund and General Statistics Office, <http://countryoffice.unfpa.org/vietnam/drive/InternalMigration_RelatedLifecourseEvents_GSO1206_e.pdf> Accessed October 2011.

Hardeweg, B., & Waibel, H. (2009). Collecting data to measure vulner-ability to poverty: An overview. Paper presented at the Workshop ofResearch Unit 756 of the Deutsche Forschungsgemeinschaft (DFG),Hue City, Vietnam.

Harris, J., & Todaro, M. (1970). Migration, unemployment and develop-ment: A two-sector analysis. American Economic Review, 60(1),126–142.

Haughton, J., & Khandker, S. R. (2009). Handbook on poverty andinequality. World Bank, <http://issuu.com/world.bank.publications/docs/9780821376133> Assessed April 2011.

Heckman, J., & Navarro-Lozano, S. (2004). Using matching, instrumentalvariables and control functions to estimate economic choice models.Review of Economics and Statistics, 86(1), 30–57.

Klasen, S., Lechtenfeld, T., & Povel, F. (2013). Using household surveysto capture vulnerability: Issues and challenges. In S. Klasen, & H.Waibel (Eds.), Vulnerability to poverty (pp. 80–114). New York:Palgrave Macmillan.

Figure 1. Natural disaster loss and gov’t budget for disaster prevention.

90 WORLD DEVELOPMENT

Le, B. D., Tran, G. L., & Nguyen, T. T. P. (2011). Social protection forrural–urban migrants in Vietnam: Current situation, challenges andopportunities, Research report, 08. Sussex, UK: Centre for SocialProtection, Institute of Development Studies.

Lee, E. S. (1966). A theory of migration. Demography, 47, 49–56.Massey, D. S. (1990). Social structure, household strategies, and cumu-

lative causation of migration. Population Index, 56(1), 3–26.Mincer, J. (1978). Family migration decisions. The Journal of Political

Economy, 86(5), 749–773.Ministry of Finance. (2010). The agriculture insurance pilot program in

Vietnam.Newman, C., & Wainwright, F. (2011). Income shocks and household risk-

coping strategies: Evidence from Rural Vietnam, Discussion paper no.358. Institute for International Integration Studies, <http://www.tcd.ie/iiis/documents/discussion/pdfs/iiisdp358.pdf> AssessedApril 2012.

Nguyen, V. C., Den Berg, M. V., & Lensink, R. (2009). The impact ofworking migration and non-working migration on household welfare,poverty and inequality: New evidence from Vietnam. Asia-PacificDevelopment Journal, 16(1), 59–92.

Nguyen, N. P., Tran, N. T. M. T., Nguyen, T. N., & Oostendorp, R.(2008). Determinants and impacts of migration in Vietnam, Workingpaper 01. Development and Policies Research Center (DEPOCEN),<http://depocenwp.org/upload/pubs/TranNgoMinhTam/Determinants%20and%20Impacts%20of%20Migration%20in%20Vietnam_DEPO-CENWP.pdf> Assessed October 2011.

Niimi, Y., Pham, T. H., & Reilly, B. (2009). Determinants of remittances:Recent evidence using data on internal migrants in Vietnam. AsianEconomic Journal, 23(1), 19–39.

Oxfam & Vietnam Academic of Social Science (Oxfam and VASS) (2010).The Impacts of the global economic crisis on migration patterns inVietnam: Findings from rapid assessments in five provinces and cities,Research report. Oxfam GB Publisher, <http://policy-practice.oxfam.org.uk/publications/download?Id=364515anddl=http://oxfamilibrary.openrepository.com/oxfam/bitstream/10546/112523/1/rr-impacts-global-economic-crisis-migration-patterns-vietnam-230210-en.pdf> AssessedNovember 2011.

Phung, D. T., & Waibel, H. (2009). Diversification, risk management andrisk coping strategies: Evidence from rural households in threeProvinces in Vietnam: Paper no. 25. In: Proceedings of the Germandevelopment economics conference, Frankfurt, Germany.

Pincus, J., & Sender, J. (2008). Quantifying poverty in Vietnam: Whocounts?. Journal of Vietnamese Studies, 3(1), 114–150, <http://www.jstor.org/stable/10.1525/vs.2008.3.1.108> Assessed November,2012.

Ravenstein, E. G. (1885). The laws of migration. Journal of the StatisticalSociety of London, 48(2), 167–235.

Ravenstein, E. G. (1889). The laws of migration. Journal of the RoyalStatistical Society, 52(2), 241–310.

Rosenbaum, P., & Rubin, R. (1983). The central role of the propensityscore in observational studies for causal effect. Biometrika, 70(1),41–45.

Shah, N. M. (2000). Relative success of male workers in the host country,Kuwait: Does the channel of migration matter?. International Migra-tion Review, 34(1), 59–78.

Sjaastad, L. A. (1962). The costs and returns of human migration. TheJournal of Political Economy, 70(5), 80–93.

Smith, J. A., & Todd, P. E. (2005). Does matching overcome LaLonde’scritique of on experimental estimators?. Journal of Econometrics, 125,305–353.

Stark, O., & Bloom, D. E. (1985). The new economics of labor migration.The American Economic Review, 75(2), 173–178.

Tongruksawattana, S., Schmidt, E., & Waibel, H. (2010). Shocks andcoping actions of rural households: Empirical evidence from NortheastThailand. Paper prepared for the CPRC international conference 2010:Ten years of war against poverty hosted by the Brooks World PovertyInstitute, University of Manchester, UK.

United Nations Population Fund (UNFPA) (2010). Internal migration:Opportunities, challenges and social-economic development in Vietnam.The United Nations Population Fund in Viet Nam, <http://unfpa.org/webdav/site/vietnam/shared/Migration%20Main%20Paper_ENG_FINAL.pdf> Accessed August 2011.

APPENDIX

Table 6. Summary statistics of variables as included in the probit specification (1)

Variables Obs* Mean** Std. Dev. Min Max

Dependent variables in 2008 or 2010

Migrant HH (1-yes, 0-no) 1432 0.19 0.39 0 1Employment migrant HH (1-yes, 0-no) 1432 0.10 0.31 0 1Education migrant HH (1-yes, 0-no) 1432 0.04 0.21 0 1

Independent variables in 2007

HH experienced demographic shocks (1-yes, 0-no) 1432 0.42 0.49 0 1HH experienced social shocks (1-yes, 0-no) 1432 0.04 0.21 0 1HH experienced agriculture shocks (1-yes, 0-no) 1432 0.25 0.43 0 1HH experienced economic (1-yes, 0-no) 1432 0.07 0.26 0 1Female headed HH (1-yes, 0-no) 1432 0.17 0.37 0 1Age of HH head (years) 1432 47.23 14.62 0 91Dependency ratio 1432 0.31 0.28 0 1HH members w/completed secondary school 1432 0.65 0.92 0 6HH members w/completed professional education 1432 1.76 1.30 0 8HH members belong to political or social organization (1-yes, 0-no) 1432 0.68 0.47 0 1Log of monthly HH per capita income (PPP$ in 2005) 1432 4.07 1.02 1.49 6.80Household engaged in off-farm activities (1-yes, 0-no) 1432 0.50 0.50 0 1Log of land per capita (ha) 1432 �2.38 1.37 �8.01 2.02Household is indebted (1-yes, 0-no) 1432 0.69 0.46 0 1Village road condition (1-Good condition, 0-Bad condition) 1432 0.48 0.50 0 1Log distance from village to district headquarters (km) 1432 2.31 0.86 �1.61 4.32Ha Tinh province (1-yes, 0-no) 1432 0.31 0.46 0 1Thua Thien Hue province (1-yes, 0-no) 1432 0.35 0.48 0 1Dak Lak province (1-yes, 0-no) 1432 0.34 0.47 0 1

Source: Own calculations based on the DFG Migrant Survey 2010, DFG Rural Village Survey 2007, and the DFG Rural Household Survey 2007, 2008and 2010.* 121 observations had been dropped to control outliners.** For binary variables, the mean refers to the share of migrants for which the dummy is equal to 1.

Table 7. Daily income of migrants with wage–employment (percent of total)

Daily wage income (1000 VND) Industry/production sector (N = 106) Public/private service sector (N = 127) Total sample (N = 233)

<50 37.7 41.7 39.951–100 52.8 33.1 42.1101–150 7.5 15.7 12.0>150 1.9 9.4 6.0Stdev (1000 VND) 44.1 56.2 51.3Median (1000 VND) 66.7 66.7 66.7Average (1000 VND) 65.1 76.2 71.2

Source: Own calculations based on the DFG Migrant Survey 2010.

RURAL–URBAN MIGRATION, HOUSEHOLD VULNERABILITY, AND WELFARE IN VIETNAM 91

Objective EQI Subjective EQI0

2040

6080

100

Perc

enta

ge o

f mig

rant

s (%

)

-.5 0 .5 1 1.5EQI

010

2030

4050

Perc

enta

ge o

f mig

rant

s (%

)

-.5 0 .5 1 1.5EQI

Aggregate EQI

010

2030

Perc

ent

0 .2 .4 .6 .8 1EQI

Source: Own calculations based on the DFG Migrant Survey 2010.

Figure 2. Migrant distribution of employment quality indices.

Table 8. Summary statistics of variables as used in the EQI models (6)

Variables Obs* Mean** Std. dev. Min Max

Dependent variable

Objective EQI 228 0.50 0.33 0 1

Migrant characteristics

Years of schooling 228 10.72 3.82 1 19Female migrant (1-yes; 0-no) 228 0.53 0.50 0 1Debt (1-yes; 0-no) 228 0.31 0.46 0 1Age (years) 228 23.98 5.02 14 47Job in service sector (1-yes, 0-no) 228 0.54 0.50 0 1Length of migration period (Years) 228 3.11 1.44 1 6Pay to get a job (1-yes; 0-no) 228 0.82 0.39 0 1How to find a job (1-Introduced from friends or relatives; 0-Others) 228 0.33 0.47 0 1

Household characteristics

Ethnicity (1-Kinh, Hoa; 0-Others) 228 0.96 0.18 0 1HH members belong to political or social organization (1-yes, 0-no) 228 0.97 0.17 0 1Income loss from shocks (2005 $ PPP) 228 207.11 543.79 0 4510.38

Source: Own calculations based on the DFG Migrant Survey 2010 and DFG Rural Household Survey 2010.* Fifteen cases had been dropped from the analysis.** For binary variables, the mean refers to the share of migrants for which the dummy is equal to 1.

92 WORLD DEVELOPMENT

Table 9. Changes in vulnerability to poverty and poverty status in period 2007–2010

2010

Non-poor Poor Total

2007 Non-poor 156 17 173Poor 57 38 95Total 213 55 268

2010

Non-vulnerability Vulnerability Total

2007 Non-vulnerability 188 12 200Vulnerability 38 30 68Total 226 42 268

Source: Own calculations based on the DFG Rural Household Survey 2007, 2008 and 2010.

RURAL–URBAN MIGRATION, HOUSEHOLD VULNERABILITY, AND WELFARE IN VIETNAM 93

ScienceDirectAvailable online at www.sciencedirect.com