Does non-farm sector employment reduce rural poverty and vulnerablty? Evdence from Vetnam and Inda Katsush S. Ima * School of Socal Scences, Unversty of Manchester, UK & RIEB, Kobe Unversty, Japan Raghav Gaha Faculty of Management Studes, Unversty of Delh, Inda Ganesh Thapa Internatonal Fund for Agrcultural Development, Rome, Italy Ths Draft: 2 nd January 2015 Abstract The present study examnes whether rural non-farm employment has any poverty and/or vulnerablty-reducng effect n Vetnam and Inda. To take account of sample selecton bas assocated wth t, we have appled treatment-effects model. It s found that log per capta consumpton or log mean per capta expendture sgnfcantly ncreased as a result of access to the rural non-farm employment n both Vetnam and Inda - whch s consstent wth ts poverty reducng role of accessng - wth the aggregate effect larger n Vetnam than n Inda. Access to the rural non-farm employment sgnfcantly reduces vulnerablty too n both countres, mplyng that dversfcaton of household actvtes nto non-farm sector would reduce such rsks. When we dsaggregate non-farm sector employment by ts type, we fnd that poverty and vulnerablty reducng effects are much larger for sales, professonals, and clerks than for unsklled or manual employment n both countres. However, because even unsklled or manual non-farm employment sgnfcantly reduces poverty and vulnerablty n Inda and poverty n some years n Vetnam, ths has consderable polcy sgnfcance as the rural poor do not have easy access to sklled non-farm employment. Key Words: Poverty, Vulnerablty, Non-farm sector, Treatment Effects Model, Vetnam, Inda JEL Codes: C21, C31, I32, O15 Contact Address Katsush S. Ima (Dr) Economcs, School of Socal Scences, Unversty of Manchester, Arthur Lews Buldng, Oxford Road, Manchester M13 9PL, UK; Telephone: +44-(0)161-275-4827, Fax: +44-(0)161-275-4812 Emal: Katsush.Ima@manchester.ac.uk. Acknowledgements Ths project was funded by IFAD. The authors are grateful to (then) Drector of Asa and the Pacfc Dvson, Thomas Elhaut, for hs gudance and support. The authors acknowledge valuable comments and advce from Samuel Annm, Md. Azam, Md. Faruq Hassan, Woojng Kang, Armando Barrentos, Takahro Sato and two anonymous referees. They thank Woojn Kang for preparng VHLSS Data sets at the early stage of the project. The second author would lke to acknowledge the support and gudance of Davd Bloom at Harvard School of Publc Health durng the completon of ths study. The vews expressed are, however, those of the authors and do not necessarly represent those of the organsatons to whch they are afflated. 1
Does non-farm sector employment reduce rural poverty and vulnerablty? Evdence from Vetnam and Inda 1. Introducton Across the developng world, t s well recognzed that rural economes are not purely agrcultural and farm households earn an ncreasng share of ther ncome from non-farm actvtes. Tradtonally, rural non-farm economy (RNFE hereafter) was consdered to be a low-productvty sector dmnshng over tme where agrcultural households smply supplemented ther ncome. But, snce the late 1990s, ts role n economc growth and poverty reducton began to be ncreasngly recognsed gven the growng share of RNFE across developng countres (e.g. Reardon et al., 1998, 2000, Lanjouw and Lanjouw, 2001, van de Walle and Cratty, 2004, Haggblade, et al., 2010, Hmanshu et al., 2013) The share of ncome from RNFE n total rural ncome vares - from 34% n Afrca, to 47% n Latn Amerca and 51% n Asa (Thapa et al., 2013), but t s recognsed that RNFE s becomng ncreasngly mportant n terms of ts share and growth as well as potental roles n poverty reducton n Asa, partcularly n emergng countres, such as Chna and Inda. Although most of the low and mddle-ncome Asan countres tradtonally reled on agrculture, they have undergone structural changes n recent years, due to ndustralsaton and globalsaton as well as commercalsaton of agrculture. Wthn Asa, the share of ncome from RNFE vares from over 70% for the Phlppnes and Sr Lanka to below 40% for Chna, Inda and Nepal (Thapa et al., 2013). Wth constrants on farm expanson and contnung growth of rural populaton, greater attenton s thus beng gven to non-farm actvtes. Polcy nterest n RNFE arses not just because of ts sgnfcance n generatng ncomes, but also because of ts ncreasng mportance n creatng employment, especally for rural women and the poor. 2
Among Asan countres, the present study focuses on Vetnam and Inda, both of whch experenced mpressve economc growth but varyng poverty reducton n recent years. These two countres are charactersed by hgh average GDP per capta growth rate n 1990-2010 (Vetnam 5.8%; Inda 4.9%) and a decreasng share of agrcultural value added n GDP n the same perod (Vetnam 39% to 20%; Inda 29% to 16%) (World Bank, 2014). Poverty ndces have declned durng ths perod, but there s a varaton n the speed of poverty reducton. Whle Vetnam experenced a faster poverty reducton n terms of headcount rato based on US$1.25 (64% n 1993 to 21% n 2006, further down to 13% n 2008), the speed of poverty reducton has been relatvely slow n Inda (45% n 1994, 37% n 2005, 32% n 2009) (Hmanshu and Sen, 2014). As shown by Ima et al. (2012, 2014) and Gaha et al. (2012, 2014), the speed of mprovement n nutrtonal ndcators has been slow n Inda n recent years despte the country's economc growth. There s a need for nvestgatng the reasons for dverse progress n ncome and non-ncome poverty focusng on household's lvelhood strateges, ncludng the choce of farm and non-farm employment. The present study ams to provde nsghts nto varyng pace of poverty reducton and vulnerablty n these two countres. The man hypothess we examne s whether access to rural non-farm employment reduces poverty and vulnerablty - defned as a probablty of fallng nto poverty n the next perod - n rural areas n Vetnam and Inda. We focus only on rural areas because rural economy s dstnct from urban economy n ts structure and rural poverty s stll predomnant n these countres. We wll use Vetnam Household Lvng Standards Survey (VHLSS) n 2002, 2004 and 2006 for Vetnam and Natonal Sample Survey (NSS) Data n 1993-4 and 2004-5 for Inda. Gven the sample selecton bas assocated wth access to RNFE or non-farm sector employment and the data structure where only large 3
cross-sectonal data are avalable and the panel data are not avalable 1, we wll apply treatment effects model, a varant of Heckman two-step sample selecton model (Heckman, 1979). We also examne whether the effect of reducng poverty and vulnerablty s dfferent among dfferent types of non-farm sector employment, namely, unsklled manual work, producton, sales, and professonals/ clerk. Whle the farm or agrcultural sector has played a central role n these countres, the share of non-farm actvtes has ncreased sgnfcantly n recent years. However, detaled emprcal studes estmatng the drect and/or ndrect effects of rural non-farm ncome or employment on poverty reman lmted and the present study seeks to fll ths gap. Our emprcal analyss of the role of rural non-farm employment n reducng household poverty can be placed n a large lterature of growth and development theores. Our basc assumpton s that overall economc growth s benefcal for the poor and ther escape from poverty can further promote overall economc growth (e.g. Dollar and Kraay, 2002). Ths basc assumpton can be theoretcally justfed as follows. 2 Frst, the poor - typcally the unsklled labour can beneft from the ncrease n wage rate, whch s enabled by physcal captal accumulaton by the rch durng the growth process. The poor can then nvest n assets and educaton, whch further renforces development (Galor and Moav, 2004)). Furthermore, an ncrease n the amount of sklled labour ncreases the wage rate of unsklled labour, strengthenng the postve correlaton between ablty and ntergeneratonal moblty (Maoz and Moav, 1999). Non-farm employment - partcularly sklled - can thus have a substantal poverty-reducng effect. The second justfcaton can be made n terms of the connecton between the dvson of labour and the economc growth, whch orgnated n Adam Smth s (1776) emphass on the role of the dvson of labour n the ncrease n labour 1 It s possble to construct a small panel based on the ntersectons of dfferent rounds of household cross-sectonal data of VHLSS n Vetnam, but attrton bas s serous as only a small subset of the total households were revsted. 2 We are grateful to one of the revewers for ths valuable suggeston. 4
productvty or ts further extensons. For nstance, Becker and Murphy (1992) examned the process of specalzaton and the dvson of labour and showed that a more extensve dvson of labour rases productvty. Km (1989) n a smlar ven analysed the mpact of human captal nvestment decsons on the depth and breadth of sklls and showed that workers nvest more n skll depth than skll breadth as the sze of the labour market ncreases. Our emprcal results on the role of sklled employment n RNFE are n lne wth these studes. The rest of the paper s organsed as follows. The next secton revews extant studes of the effects of non-farm sector on poverty n Vetnam and Inda. Secton 3 brefly summarses the data sets we wll use. Sectons 4 and 5 dscuss the specfcaton of econometrc models and results, respectvely. Concludng observatons are offered n the fnal secton. 5
2. Lterature Revew Whle the farm or agrcultural sector has played a central role n Vetnam and Inda, the share of non-farm has ncreased sgnfcantly n recent years. However, formal emprcal studes to estmate the drect and/or ndrect effects of ncome or employment n non-farm sector employment on poverty are stll few. On the drect effects, van de Walle and Cratty (2004) usng Vetnam Lvng Standard Survey (VLSS) data n 1993 and 1998 found sgnfcant effects of non-farm employment n reducng poverty. Whle van de Walle and Cratty (2004) clam that they consder the endogenety of non-farm sector n reducng poverty, they smply estmated the share of hours worked n non-farm sector n total (or the probablty of partcpatng n non-farm sector) and poverty separately and compared the sgns and statstcal sgnfcance of coeffcent estmates of explanatory varables wthout takng account of smultanety. Thus ther results are only suggestve of dfferent covarates of non-farm employment and poverty. More recently, usng both long-term survey data n Palanpur, a vllage n western Uttar Pradesh, and the NSS data, Hmanshu et al. (2013) have shown that the dversfcaton nto RNFE not only ncreased household ncome but also reduced poverty. They have also provded the evdence from Palanpur whch suggests that the ncome nequalty has ncreased wth ths dversfcaton process. However, no dscussons or formal analyses have been carred out by Hmanshu et al. (2013) of the endognety assocated wth household access to RNFE. RNFE would be potentally mportant for breakng the poverty traps caused by, for nstance, lack of educaton or nutrton. For example, people who are educated at secondary school or hgher levels are lkely to have a hgher probablty of fndng a job n rural non-farm sector (e.g. n tradng, manufacturng offce works) and ther chldren tend to be more educated, whch creates a 'vrtuous' crcle (e.g. Knght et al., 2009, 10). However, those 6
who are not educated tend to be trapped n a 'vcous' crcle. Lkewse, undernourshed people tend to be trapped n poverty as low nutrtonal levels mply low effcency and hgh probablty of beng unemployed as predcted by the effcency-wage hypothess (e.g. Blss and Stern, 1978, Dasgupta and Ray, 1986, 87). The poverty-nutrton hypotheses have been recently examned by Jha et al. (2009) and Ima et al. (2012) n the context of rural Inda. Reardon et al. (2000) also emphasse the barrers faced by poor households that prevent them from nvestng n non-farm assets, suggestng the exstence of the poverty trap. That s, t s not an automatc process for poor agrcultural households to enter the non-farm sector. Unlke agrcultural jobs, rural non-farm employment tends to be less physcally ntensve and requres lower calores, as the actvty ntensty determnes the nutrtonal status n rural Inda (Ima et al., 2012). Snce RNFE tends to better promote food securty to the poor than farm employment (Owsu et al., 2011), the former has the potental to break the poverty trap. Whle buldng upon van de Walle and Cratty (2004) and Hmanshu et al. (2013), our study takes account of the endogenety ssues based on natonal data n Vetnam and Inda. In our estmatons, we have also estmated wage equatons for male and female workers to consder the effect of male or female wage rates on non-farm labour market partcpaton. The novelty of our study relatve to the exstng emprcal lterature ncludes () addressng the endogenety ssue formally usng natonal data n Inda and Vetnam; () consderng the effects of male and female wage rates on the non-farm labour market partcpaton; and () estmatng the effects of non-farm labour market partcpaton on vulnerablty of households after takng nto account ts endogenety. 3. Data Vetnamese Data 7
We wll use Vetnam Household Lvng Standards Surveys (VHLSS) 2002, 2004, and 2006. The VHLSSs were ntally mplemented n 2002 to collect detaled household and commune level data. These are mult-topc household surveys wth natonally representatve household samples. They commonly cover a wde range of ssues, ncludng household composton and characterstcs (e.g. educaton and health); detaled record on expendture for both food and non-food tems, health and educaton; employment and labour force partcpaton (e.g. duraton of employment and the precse categores of occupatons); ncome by sources (e.g. salary/wage, payment n cash and n knd, farm and non-farm producton); housng, ownershp of other assets and durable goods; and partcpaton of households n ant-poverty programs. Commune level surveys collect data on demography, economc condtons, agrcultural producton, and non-farm employment, local nfrastructure, publc servces such as educaton and health facltes. Occupatonal code of employment provded by VHLSS would enable us to classfy non-farm sector employment n several sub-components broadly defned (.e., manual/unsklled; producton; sales; professonals/clerk). Indan Data The NSS, set up by the Government of Inda n 1950, s a mult-subject ntegrated sample survey conducted all over Inda n the form of successve rounds relatng to varous aspects of socal, economc, demographc, ndustral and agrcultural statstcs. We use the data n the Household Consumer Expendture schedule, qunquennal surveys n the 50 th round, 1993 94, and n the 61 st round, 2004-05. 3 These form repeated cross-sectonal data sets, each of whch contans a large number of households across Inda. The consumpton schedule contans a varety of nformaton related to mean per capta expendture (MPCE), dsaggregated expendture over many tems together wth basc soco economc 3 We are not usng 55 th round n 1999-2000 as the consumpton data n 55 th round are not comparable wth those n 50 th or 61 st round because of the change n recall perods. The consumpton data are comparable between 50 th round and 61 st round. 8
characterstcs of the household (e.g., sex, age, relgon, caste, and land-holdng). To derve wages at the level of NSS agro-clmatc regon, we supplement the consumpton schedule by Employment and Unemployment schedule because the consumpton survey and the employment survey collect data on dfferent households and can be lnked only at the aggregate level (e.g. NSS regon level). 4 Non-farm sector employment can be classfed nto sub-categores by usng Natonal Classfcaton of Occupatons (NCO). 4. Methodologes (1) Treatment Effects Model To estmate the effect of non-farm sector employment on poverty and vulnerablty, we employ a verson of treatment effects model. The man dea of treatment effects model s to estmate poverty defned by household consumpton per capta for two dfferent regmes (de Janvry et al., 2005) - households partcpatng only n the farm labour market and those partcpatng n both farm and non-farm labour markets. It s a verson of the Heckman sample selecton model (Heckman, 1979), whch estmates the effect of an endogenous bnary treatment. Ths would enable us to take account of the sample selecton bas assocated wth access to non-farm sector. In the frst stage, access to non-farm sector s estmated by the probt model. 5 In the second, we estmate log of household consumpton or vulnerablty measure after controllng for the nverse Mlls rato whch reflects the degree of sample selecton bas. The mert of treatment effects model s that sample selecton bas s explctly estmated by usng the results of probt model. However, the weak aspects nclude: () strong 4 Defntons of the varables of VHLSS and NSS data are gven n Appendx Table. 5 More specfcally, we run the probt model at the household level for whether any household members have access to non-farm sector and then estmate the consumpton or vulnerablty equaton n the second stage. 9
assumptons are mposed on the dstrbutons of the error terms n the frst and second stages; () the coeffcent estmates may be senstve to choce of the explanatory varables and nstruments; and () vald nstruments are rarely found n non-expermental data and f the nstruments are nvald, the results wll depend on the dstrbutonal assumptons. The selecton mechansm by the probt model for accessng rural non-farm economy (RNFE) can be more explctly specfed as (e.g., Greene, 2003): D * X u (1) * * and D 1 f D X 0 D * 0 otherwse where PrD 1 X ( X ) Pr D 0 X 1 ( X ) * D s a latent varable. In our case, D takes the value 1 f an th household has at least one household member who has access to non-farm employment and 0 otherwse. X s a vector of ndvdual, household and regonal characterstcs and other determnants at commune or communty levels. denotes the standard normal cumulatve dstrbuton functon. Snce avalable varables are dfferent for Vetnam and Inda, we assume dfferent specfcatons (or the choce of explanatory varables) for ndvdual access to RNFE for X. Vetnam: * ˆ m ˆ f D D( W, W, M, E, H, L, R ) (2) m Wˆ : a household average of predcted wages of male members. Daly wage rate s estmated by ndvdual characterstcs, such as, age, ts square, dummy varables of educatonal categores, whether he s workng for the household s own farm (or non-farm) sector as a wage worker, whether the household belongs to ethnc majortes, sze of land and ts square, and regonal and locatonal dummy varables. 10
f Wˆ : a household average of predcted wages of female members. m Wˆ and f Wˆ serve as nstruments for the household s non-farm partcpaton equaton. 6 If non-farm jobs emerge as a result of the dvson of labour and the non-farm wages exceed the reservaton wage mpled by the agrcultural sector, people would work as nonfarm workers. If ths s the case, there should be a postve lnk between the average wages and non-farm labour market partcpaton. Alternatvely, we can assume that the labour productvty proxed by wage rate s an mportant determnant of partcpaton n non-farm sector employment. That s, only hgh productvty worker wth hgher agrcultural wages rate can partcpate n RNFE as an analogy of theory of workfare where only hgh productvty workers can partcpate n workfare scheme or hgher waged workers can afford exercsng the real opton of swtchng from the agrculture labour marker to workfare or the non-farm labour market gven the swtchng costs (Scandzzo et al., 2009). M : whether the household head s male. E : a set of dummy varables of educatonal attanment of the household head (whether he or she has no educaton; whether completed prmary educaton; whether completed lower secondary educaton; whether completed upper secondary educaton; whether completed techncal educaton; whether completed hgher educaton). H : household composton/ characterstcs (household sze; the share of female members; dependency burden (the share of household members below 15 years or above 65 years; whether a household belongs to ethnc majorty) of the th household). L : sze of land (n hectare) owned by the household and ts square for the th household. 6 We estmate the wage equatons for male and female workers separately gven the segmentaton of labour markets by gender n developng countres. If we use the household averages of ndvdual varables (e.g. educaton or age) as well as household-level varables n estmatng the household probablty of accessng the non-farm sector employment n one equaton, some of the ndvdual varables (e.g. educaton) wll be dropped automatcally due to multcollnearty. Hence, to utlse the data structure of our survey data, we estmate the wage equatons frst usng the ndvdual data and then estmatng the rural non-farm employment equaton by usng predcted wages for Vetnam. 11
R : a set of regonal dummy varables (whether a household s located n red rver delta regon; northeast regon; northwest regon; north central coast regon; south central coast regon; central hghlands regon; north east south regon; Mekong rver delta regon; central coast regon; low mountans; and hgh mountans). Inda: Because of data lmtatons, a dfferent set of explanatory varables s chosen as determnants of accessng rural non-farm employment. * D D( W, E, H, L, B, R ) (3) W : wage rate estmated usng employment data and aggregated for NSS regon. 7 8 Explanatory varables n the wage rate equatons nclude age and ts square, a number of dummy varables on lteracy and educatonal attanments, land, Scheduled Trbe (ST), Scheduled Caste (SC), non-agrcultural or agrcultural self- employment, relgon. State fxed effects are consdered by nsertng state dummy varables. E : a set of varables on the hghest level of educatonal attanment of household members (e.g. whether completed prmary school, secondary school, or hgher educaton). H : a set of varables ndcatng household composton, such as whether a household s headed by a female member, number of adult male or female members, dependency burden: the share of household members under 15 years old or over 60 years old. L : owned land as a measure of household wealth. 7 Unlke VHLSS data for Vetnam, matchng of ndvdual wage data and household data n NSS for Inda s only possble at the level of NSS agro-clmatc regon, because, as we have already noted, dfferent sets of households or ndvduals are surveyed for the Employment and Unemployment Schedule whch contans the ndvdual wage data, and for the Expendture Schedule coverng household consumpton or other household varables. 8 The results for wage equatons for Vetnam and Inda wll be provded on request. 12
B : socal backwardness of the household n terms of () whether a household belongs to Scheduled Castes (SCs) and () whether t belongs to Scheduled Trbes (STs). R : a vector of state dummy varables. The lnear outcome regresson model n the second stage s specfed below to examne the determnants of poverty - as proxed by household consumpton (log of MPCE for the Indan data and log of per capta real household consumpton for the Vetnamese data) or vulnerablty derved by Chaudhur s (2003) method whch captures the probablty of a household fallng nto poverty n the next perod. It s noted here that non-farm labour market partcpaton s estmated n the frst stage of the treatment effects model, whle poverty s estmated (proxed by log per capta household consumpton or household vulnerablty) n the second stage. We use log household consumpton and vulnerablty as a measure of poverty because treatment effects model requres that the dependent varable n the second stage s contnuous and the standard bnary measure of poverty (0 or 1) cannot be used. Moreover, as suggested by prevous lterature, households n Inda and Vetnam tend to be vulnerable to shocks (e.g. Ima et al, 2011; Gaha and Ima, 2009). We denote household poverty - ether log per capta household consumpton or vulnerablty asy. Y Z D (4) u, ~ bvarate normal0,0,1,,. where s the average net effect (ANE) of access to rural non-farm sector employment. In case log per capta household consumpton s estmated, the postve estmate of mples that accessng RNFE ncreases consumpton and thus decreases poverty unless ncome dstrbuton changes. In the case of vulnerablty, the negatve estmate of mples that access to rural non-farm sector employment decreases vulnerablty. Here Z s a vector of determnants ofy. For Vetnam ths s estmated by: Z Z M, E, H, L, R ) (5) ( 13
and for Inda Z Z E, H, L, B, R ) (6) ( That s, we nclude all the varables used for the non-farm sector partcpaton equaton ((2) or (3)) except the nstruments, predcted wages. Usng a formula for the jont densty of bvarate normally dstrbuted varables, the expected poverty for those wth access to rural non-farm sector employment s wrtten as: E Y D 1 Z E D 1 X Z X where s the standard normal densty functon. The rato of and s called the nverse Mlls rato. Expected poverty (or undernutrton or vulnerablty) for non-partcpants s: E Y D 0 Z E D 0 X Z 1 X The expected effect of poverty reducton assocated wth RNFE s computed as (Greene, 2003, 787-789): (7) (8) X D 1 E Y D 0 E Y X 1 X (9) If s postve (negatve), the coeffcent estmate of usng OLS s based upward (downward) and the sample selecton term wll correct ths. Snce s postve, the sgn and sgnfcance of the estmate of (usually denoted as ) wll show whether there exsts any selecton bas. To estmate the parameters of ths model, the lkelhood functon gven by Maddala (1983, p.122) s used where the bvarate normal functon s reduced to the unvarate functon and the correlaton coeffcent. The predcted values of (7) and (8) are 14
derved and compared by the standard t test to examne whether the average treatment effect or poverty reducng effect s sgnfcant. The results of treatment effects model wll have to be nterpreted wth cauton because the results are senstve to the specfcaton of the model or the selecton of explanatory varables and/or the nstrument. Also mportant are the dstrbutonal assumptons of the model. Despte these lmtatons, the model s one of the few avalable methods to control for sample selecton bas and capable of yeldng nsghts nto whether access to rural non-farm sector employment leads to poverty reducton. (2) Vulnerablty Measure It would be deal to use panel data to derve household s vulnerablty measures, but, n ts absence, we can derve a measure of Vulnerablty as Expected Poverty (VEP), an ex ante measure, based on Chaudhur (2003) and Chaudhur, Jalan and Suryahad (2002) who appled t to a large cross-secton of households n Indonesa 9 and defned vulnerablty as the probablty that a household wll fall nto poverty n the future after controllng for the observable household characterstcs. Accordngly, t takes the value from 0 to 1, and the hgher the value of vulnerablty measure, the hgher s the probablty of a household fallng nto poverty n the next perod. Ima et al. (2011) derved and analysed Chaudhur s vulnerablty measure usng the VHLSS data for Vetnam, and Ima (2011) dd so usng the Indan NSS data. We wll use these cross-sectonal vulnerablty measures subject to the caveat of estmatng vulnerablty from a sngle cross-secton that cannot capture the effect of aggregate shocks affectng all the households n the sample area. The detals of dervaton of Chaudhur s vulnerablty measure s found n Hoddnott and Qusumbng (2003b). Ima et al. (2011) and Ima (2011) provde a full set of results of vulnerablty for Vetnam and Inda. 9 See a summary by Hoddnott and Qusumbng (2003a, b) of methodologcal ssues n measurng vulnerablty. 15
4. Econometrc Results Ths secton summarses the results of treatment effects model whch s appled to estmate the effects of accessng rural non-farm sector employment. Vulnerablty estmates based on VHLSS and NSS data are reported n Ima et al. (2011) and Ima (2011) and we hghlght only the results of treatment effects model. Table 1 gves the results of treatment effects model appled to VHLSS data n 2002, 2004 and 2006. For each year, two dfferent proxes for poverty have been tred as a dependent varable - log of per capta consumpton and vulnerablty. The frst panel reports the results of the frst stage probt model for whether a household member partcpates n the non-farm sector labour market and the second panel gves the results for OLS whereby log per capta consumpton or vulnerablty s estmated. The frst panel of Table 1 suggests that predcted wage rates as well as household characterstcs (e.g. educatonal attanment, household composton) affect the probablty of a household member partcpatng n the non-farm sector. In 2002 and 2006, both predcted male wage rate and female wage rate postvely and sgnfcantly ncreased the probablty of the household havng a member partcpatng n the non-farm sector employment, whle n 2004 only male wage rate was postve and sgnfcant. A household owng more land tends to have a smaller possblty of partcpatng n the non-farm labour market wth an ndcaton of non-lnearty, suggested by a postve and sgnfcant coeffcent estmate of the sze of land squared (wth ts coeffcent estmate n absolute term larger than the level of sze of land). Ths ndcates that, other thngs beng equal, the households wth some amount of land tend to concentrate on only farmng, whle the landless or those wth a large area of land tend to partcpate n non-farm employment, ether because ther earnngs from farmng are small or uncertan, or they can nvest n some sklls due to the large ncome from farmng. Other 16
varables show more or less expected results (e.g. hgher educatonal attanment tends to ncrease the probablty of partcpatng n non-farm employment; belongng to ethnc majorty ncreases the probablty; a younger household head s more lkely to partcpate n non-farm employment; locaton affects the probablty). or n equaton (7) s statstcally sgnfcant (except n case of consumpton n 2004 and n 2006), mplyng that there exsts sample selecton bas that should be corrected for n dervng the average treatment effects. Use of treatment effects model s justfed n most cases. The second panel of Table 1 shows the results of determnants of per capta household consumpton and household vulnerablty for 2002, 2004 and 2006. For example, sze of household sgnfcantly decreases consumpton n all the years and sgnfcantly decreases vulnerablty n 2002, 2004 and 2006. An older household head tends to have hgher per capta consumpton and lower vulnerablty wth non-lnear effects. Hgher dependency burden s assocated wth lower per capta consumpton and hgher vulnerablty. Educaton and locaton are mportant determnants of both consumpton and vulnerablty. ˆ, an estmate of n equatons (7) and (9) shows the Average Net Effects (ANE) and t s postve and sgnfcant except n the case of consumpton for 2004. However, ANE should not be treated as a treatment effect f the sample selecton term,, s statstcally sgnfcant. At the bottom of the table, we report the Average Treatment Effect (ATE), the dfference of the expected outcome for partcpants n non-farm employment and for non-partcpants after controllng for sample selecton (as n equaton (9), the sum of ANE and the sample selecton term). In order to evaluate the effect of access to non-farm employment on poverty after takng account of sample selecton, we need to base our dscusson on ATE, rather than ANE (Ima, 2011). In 2002, per capta consumpton was sgnfcantly hgher by 19.2% for partcpants n the non-farm labour market than for non-partcpants after takng account of sample selecton, 17
whch s consstent wth the poverty reducng role of rural non-farm employment. In the same year, vulnerablty as a probablty of fallng nto poverty s reduced by 14.9% as a result of partcpatng n non-farm employment. In 2004, per capta consumpton s sgnfcantly hgher by 12.9% for non-farm labour market partcpants than for non-partcpants after controllng for sample selecton, whle the vulnerablty s lower for non-farm labour market partcpants by 7.3%. In 2006, per capta consumpton s hgher by 13.1% and vulnerablty s lower by 5.9% f the household has access to non-farm employment. In sum, we confrm that rural non-farm employment substantally reduced consumpton poverty and vulnerablty throughout the perod 2002 to 2006. The results may suggest that RNFE opens up a new set of consumpton bundles whch others could not aval of. (Table 1 to be nserted) To see how non-farm sector employment n dfferent categores affects poverty and vulnerablty, we have repeated the same model by changng only the defnton of bnary-classfcaton of non-farm employment. Only the fnal results of ATE are summarsed n Table 2. Sub-categores are broadly defned by occupatonal code of ndvdual members partcpatng n the non-farm employment - Unsklled manual employment (mechancal and physcally demandng jobs e.g. unsklled constructon works), Producton (jobs classfed n manufacturng sector or assocated wth producton, e.g. employment n plant and machne operators/ assemblers or craftsman), Sales (jobs assocated wth sales and trade) and Professonals/ clerks (managers, professonals, techncans, clerks). It should be noted that these categores are broadly defned by the occupatonal categores wthn whch ranks or skll requrements are dverse. Hence, the results should be nterpreted wth cauton. Also, dfferent occupatonal codng systems are used for Vetnam and Inda and the results are not necessarly comparable. 18
(Table 2 to be nserted) Gven the above caveats, ATE on consumpton and vulnerablty s dfferent across dfferent categores of non-farm employment. For example, Unsklled and Manual non-farm employment ncreased consumpton by 5.1% n 2002 and by 11.0% n 2004, but n the meantme ncreased vulnerablty (by 5.2% n 2004 and 5.8% n 2006). Ths mples that non-farm manual employment may ncrease household consumpton only at the cost of greater vulnerablty n Vetnam. Non-farm sector n Producton ncreased consumpton sgnfcantly over the years wth some varaton (by 15.7% n 2002, 3.2% n 2004 and 13.8% n 2006) and decreased vulnerablty by 15.6% n 2002 and by 2.1% n 2004. The poverty and vulnerablty reducng effects of non-farm employment on Sales and Professonals/ Clerks are more clearly observed. On average, access to employment n Sales ncreased per capta consumpton by 21.0% to 29.6% and reduced vulnerablty by 6.0% to 26.7%. The effect of non-farm employment n Professonals/ Clerks on per capta consumpton was also substantal over tme (rangng from 15.4% to 22.0%), whle vulnerablty was substantally reduced at the same tme. That s, wth some varaton, poverty and vulnerablty reducng effects of sklled non-farm employment are much stronger than those of unsklled or manual employment after controllng for sample selecton bases assocated wth partcpaton n such non-farm employment. Table 3 gves the results of treatment effects model for the Indan NSS data. As before, the frst panel presents the results of partcpaton equaton (probt model). Female headedness negatvely affected partcpaton n NSS61 (n 2004-2005). 10 Dependency burden s negatve and sgnfcant, that s, the household wth hgher dependency burden s less lkely to partcpate n the rural non-farm sector employment. A younger household head s more lkely to partcpate n non-farm employment but wth non-lnear effects. A 10 Because female headedness s measured wth error n NSS50/year, t was not used n the regresson. 19
household wth more educated members tends to partcpate n non-farm employment. If the household has more land, the probablty of partcpatng n non-farm employment s larger, whch s suggested by a postve and sgnfcant coeffcent for the dummy varable on whether a household owns land greater than 2.5 hectares (the frst, second and the last columns). Ths s because n Inda where farmers may face a greater degree of uncertanty, ownng a larger land s necessary for acqurng some sklls to be engaged n non-farm employment. But ths s only possble f the farms own a certan amount of land (.e. greater than 2.5 hectares) as we also observe a negatve and sgnfcant coeffcent estmate for a dummy on ownng a small area of land (between 0.1and 2.5 hectares) (the thrd and fourth columns, 2004-5). Belongng to the SCs and STs s also assocated wth lower probablty of partcpatng n non-farm employment. For NSS50,.e., n 1993-1994, hgher predcted wages sgnfcantly lead to hgher probablty of partcpatng n non-farm employment. The coeffcent estmate for regonal prce s postve, but not statstcally sgnfcant for NSS61. The coeffcent estmate of or s sgnfcant except the cases for vulnerablty n 1993-1994 and log MPCE n 2004-5. (Table 3 to be nserted) The second panel of Table 3 reports the regresson results of the second-stage equaton for log MPCE or vulnerablty. We report the regresson results only selectvely. For nstance, n contrast to Vetnam, somewhat surprsng s the fndng that dependency burden sgnfcantly ncreased log MPCE, but decreased vulnerablty. In 1993-1994, a household wth an older head was more vulnerable wth a strong non-lnear effect, whle age of the head had no sgnfcant effect on per capta consumpton. On the contrary, a household wth an older head consumed more wth a strong non-lnear effect n 2004-2005. In general, a household wth a more educated household consumed more and was less vulnerable. As expected, the larger the sze of the land a household owned, t consumed more and was less 20
vulnerable. Belongng to the SCs or STs was assocated wth a lower level of consumpton as well as a hgher level of vulnerablty. We have summarsed the results of ATE at the bottom of Table 3. It s confrmed that access to non-farm employment ncreased per capta consumpton on average by 10.2% n 1993-4 and 10.4% n 2004-5. That s, the consumpton ncreasng effect (or the effect of reducng consumpton poverty) contnued to be substantal. Vulnerablty was sgnfcantly reduced by partcpaton n non-farm employment - by 3.8% n 1993-4 and by 7.1% n 2004-2005 (n terms of the probablty of fallng nto poverty n the next perod). It can be concluded that n Inda partcpaton n RNFE s lkely to reduce household vulnerablty sgnfcantly. Table 4 reports a summary of the results for Inda where non-farm employment s dsaggregated by occupatonal categores. Professonals/ Clerks has the largest poverty and vulnerablty reducng effects n 1993-1994 and 2004-2005, followed by Producton and Sales whch have smlar magntudes of poverty and vulnerablty reducng effects. Unsklled/ Manual employment nvolves the smallest poverty and vulnerablty reducton effects among the four categores, but the role of these effects should not be neglected gven that the poor do not have easy access to sklled employment n non-farm sector. Access to unsklled/ manual employment n non-farm sector ncreased per capta consumpton by 6.0% (8.4%) n 1993-1994 (2004-2005) on average, whle t reduced the probablty of fallng nto poverty by 4.0% (7.6%) n 1993-1994 (2004-2005). (Table 4 to be nserted) 5. Concludng Observatons The present study has contrbuted to the exstng emprcal lterature, whch has already establshed that rural non-farm ncome s an mportant resource for rural agrcultural 21
households, n the followng three ways. Frst, ths s the frst study to show that rural non-farm sector employment has sgnfcantly reduced poverty and vulnerablty n Vetnam and Inda whle takng account of the endogenety problem. Second, and related to the frst pont, we have dentfed a larger poverty reducng effect of sklled employment n rural non-farm sector. Fnally, by focusng on both Vetnam and Inda, we have developed a comparatve perspectve n terms of the role of rural non-farm employment. More specfcally, the present study has examned whether partcpaton n the rural non-farm sector employment or nvolvement n actvty n rural non-farm economy (RNFE) has any poverty-reducng or vulnerablty-reducng effect n Vetnam and Inda drawng upon naton-wde cross-sectonal household data sets. To take account of sample selecton bas assocated wth RNFE, we appled treatment-effects model, a varant of Heckman sample selecton model. We fnd that partcpaton n non-farm sector employment sgnfcantly ncreased per capta consumpton or expendture as a proxy for poverty reducton - n 2002, 2004, and 2006 n rural Vetnam and n 1993-1994 and 2004-2005 n rural Inda. The results are consstent wth poverty and vulnerablty reducng roles of accessng RNFE. Ths s mportant as a sgnfcant number of households were found to be not only poor but also vulnerable to shocks n the future (e.g. weather shocks, llness of household members, macro-economc slowdown) n Vetnam as well as Inda (Gaha and Ima, 2009; Ima et al., 2011). Dversfcaton of household actvtes nto non-farm sector would reduce such rsks. Dsaggregaton of non-farm sector employment by occupatonal categores shows that access to more sklled employment s lkely to have larger poverty and vulnerablty reducng effects than unsklled or manual employment. Non-farm employment n Sales and Professonals/ Clerks categores has stronger effects n reducng poverty and vulnerablty n both Vetnam and Inda. Unsklled/ Manual employment sgnfcantly reduces poverty and vulnerablty n Inda over the years and access of the rural poor to unsklled or manual 22
employment s lkely to be mportant gven that the poor do not have easy access to sklled employment n non-farm sector. On the contrary, the poverty reducng effect of unsklled/ manual non-farm employment s observed n 2002 and 2004, but not n 2006 n Vetnam but wth greater household vulnerablty n 2004 and 2006. Non-farm employment assocated wth Producton sgnfcantly reduced poverty and vulnerablty over tme n both Inda and Vetnam, except n 2006 when vulnerablty rose n Vetnam. That s, we observe more consstent poverty and vulnerablty effects of relatvely unsklled/ physcal demandng jobs n non-farm sector for Inda than for Vetnam. Ths fndng has consderable polcy sgnfcance as the rural poor do not have easy access to sklled non-farm employment. The ssue s whether there s enough ncentve for the non-farm or busness sector to create preferred job types n the rural areas. Gven the potental role of non-farm sector n reducng vulnerablty, central or local governments, or nternatonal donors or NGOs mght want to support ths process from both demand and supply sdes. For nstance, t wll be mportant for governments to relocate some busness actvtes from urban to rural areas, support the busness for processng and retalng agrcultural food products n rural areas, and/or develop the nfrastructure necessary for promotng non-farm economy. In the meantme, t wll be necessary for governments or donors to help poor farmers to take up sklls necessary for rural non-farm employment, e.g. through mcrofnance programmes or newly created tranng centre. Our results are consstent wth recent vews that non-farm sector plays a key role n helpng poor households escape poverty. Polcy nterventons desgned to help agrcultural households dversfy nto non-farm sector actvtes (e.g. skll tranng; mcrofnance) would potentally reduce not only poverty but also vulnerablty. Our results have ndcated that there are more smlartes than dfferences n the mpact of rural non-farm employment between Vetnam - an economy n transton - and 23
Inda - an emergng economy. Some clues may, however, emerge from a deeper understandng of dfferences n constrants to expanson of land, varaton n populaton pressure, and productvty, access to credt, decentralsed structures of governance, and weak rural nfrastructure, whch s left for future research. That Vetnam has adapted rapdly to a market-orented polcy regme may n fact be key to why smlartes n the mpact of rural non-farm employment are so much more strkng n these two countres. If ths observaton s combned wth our man result that rural non-farm employment reduces poverty and vulnerablty, t can be nferred that more market-orented polcy regme n Vetnam has enabled the country to reduce poverty more rapdly than Inda where the shft to market-orented polcy regme s slower and the average aggregate beneft from access to rural non-farm employment s smaller. Ths s another nterestng topc for future research. References Becker, G. S., & Murphy, K. M. (1992). The dvson of labor, coordnaton costs, and knowledge. Quarterly Journal of Economcs, 107, 1137-1160. Blss C., & Stern, N. (1978). Productvty, wages and nutrton: Part I: The theory. Journal of Development Economcs, 5(4), 331-362. Chaudhur, S. (2003). Assessng vulnerablty to poverty: concepts, emprcal methods and llustratve examples. mmeo, New York: Columba Unversty. Chaudhur, S., Jalan, J., & Suryahad, A. (2002). Assessng Household Vulnerablty to Poverty: A Methodology and Estmates for Indonesa. Columba Unversty Department of Economcs Dscusson Paper No. 0102-52, New York: Columba Unversty. Dasgupta, P., & Ray, D. (1986). Inequalty as a determnant of malnutrton and unemployment: Theory. Economc Journal, 96, 1011 1034. 24
Dasgupta, P., & Ray, D. (1987). Inequalty as a determnant of malnutrton and unemployment: Polcy. Economc Journal, 97, 177 188. de Janvry, A., Sadoulet, E., & Nong, Z. (2005). The role of non-farm ncomes n reducng rural poverty and nequalty n Chna. CUDARE Workng Paper Seres 1001, Unversty of Calforna at Berkeley. Dollar, D. & Kraay, A. (2002). Growth s good for the poor. Journal of Economc Growth, 7, 195-225. Gaha, R., & Ima, K. (2009). Measurng Vulnerablty and Poverty n Rural Inda. n W. Naudé, A. Santos-Paulno and M. McGllvray (Eds.), Vulnerablty n Developng Countres, Tokyo: Unted Natons Unversty Press. Gaha, R., Kacker, N., Ima, K., & Thapa, G. (2012). Demand for Nutrents n Inda: An Analyss Based on the 50th, 61st and 66th Rounds of the NSS. IFAD, Rome, mmeo. Gaha, R., Jha, R., & Kulkarn, V. S. (2014). Dets, Malnutrton and Dsease n Inda, Oxford, Oxford Unversty Press. Galor, O., & Moav, O. (2004). From physcal to human captal accumulaton: Inequalty and the process of development. Revew of Economc Studes, 71, 1001-1026. Greene, W. H. (2003). Econometrc Analyss 5 th edton, Upper Saddle Rver, NJ: Prentce-Hall. Haggblade, S., Hazell, P., & Reardon, T. (2010). The Rural Nonfarm Economy: Prospects for Growth and Poverty Reducton. World Development, 38(10), 1429 1441. Heckman, J. (1979). Sample selecton bas as a specfcaton error. Econometrca 47, 153-161. Hmanshu, Lanjouw, P., Murga, R., & Stern, N. (2013). Nonfarm dversfcaton, poverty, economc moblty, and ncome nequalty: a case study n vllage Inda. Agrcultural Economcs 44, 461-473. 25
Hmanshu, & Sen, K. (2014). Revstng the Great Indan Poverty Debate: Measurement, Patterns, and Determnants. BWPI Workng Paper Seres 203, Manchester, the Unversty of Manchester. Hoddnott, J., & Qusumbng, A. (2003a). Data Sources for Mcroeconometrc Rsk and Vulnerablty Assessments. Socal Protecton Dscusson Paper Seres No.0323, Washngton D.C.: The World Bank. Hoddnott, J., & Qusumbng, A. (2003b). Methods for Mcroeconometrc Rsk and Vulnerablty Assessments. Socal Protecton Dscusson Paper Seres No.0324, Washngton D.C.: The World Bank. Ima, K. (2011). Poverty, undernutrton and vulnerablty n rural Inda: Role of rural publc works and food for work programmes. Internatonal Revew of Appled Economcs, 25(6), 669 691. Ima, K., Gaha, R., & Kang, W. (2011). Poverty Dynamcs and Vulnerablty n Vetnam. Appled Economcs, 43(25), 3603-3618. Ima, K., Annm, S.K., Kulkarn, V.S., & Gaha, R. (2012). Nutrton, Actvty Intensty and Wage Lnkages: Evdence from Inda. RIEB DP2012-10, Kobe: RIEB, Kobe Unversty. Ima, K., Annm, S.K., Kulkarn, V.S., & Gaha, R. (2014). Women s Empowerment and Prevalence of Stunted and Underweght Chldren n Rural Inda. World Development, 62(10), 88-105 Jha, R., Gaha, R., & Sharma, A. (2009). Calore and Mcronutrent Deprvaton and Poverty Nutrton Traps n Rural Inda. World Development, 37(5), 982 991. Hmanshu, Lanjouw, P., Murga, R., Stern, N. (2013). Nonfarm dversfcaton, poverty, economc moblty, and ncome nequalty: a case study n vllage Inda. Agrcultural Economcs, 44, 461-473. 26
Km, S. (1989). Labor specalzaton and the extent of the market. Journal of Poltcal Economy, 98, 692-705. Knght, J., L, S., & Deng, Q. (2009). Educaton and the Poverty Trap n Rural Chna: Settng the Trap. Oxford Development Studes, 37 (4), 311-332. Knght, J., L, S., & Deng, Q. (2010). Educaton and the Poverty Trap n Rural Chna: Closng the Trap. Oxford Development Studes, 38 (1), 1-24. Lanjouw, J., & Lanjouw, P. (2001). The rural non-farm: sector: ssues and evdence from developng countres. Agrcultural Economcs, 26, 1-23. Maddala, G. S. (1983). Lmted-dependent and Qualtatve Varables n Econometrcs. Cambrdge: Cambrdge Unversty Press. Maoz, Y. D. & Moav, O. (1999). Intergeneratonal moblty and the process of development. Economc Journal, 109, 677-697. Owusu, V., Abdula, A., & Abdul-Rahman, S. (2011). Non-farm work and food securty among farm households n Northern Ghana. Food Polcy, 36, 108 118. Reardon T., Stamouls, K., Balsacan, A., Cruz ME, Berdegue, J., & Banks, B. (1998). Rural non-farm ncome n developng countres, n FAO (1998). The State of Food and Agrculture, Rome: FAO, 283-356. Reardon, T., Stamouls, K., Lanjouw, P., & Balsacan, A. (2000). Effects of Non-Farm Employment on Rural Income Inequalty n Developng Countres: An Investment Perspectve. Journal of Agrcultural Economcs, 51(2), 266-288. Scandzzo, P, Gaha, R., & Ima, K. (2009). Opton Values, Swtches and Wages - An Analyss of the Employment Guarantee Scheme n Inda. Revew of Development Economcs, 13(2), 248-263. Smth, A. (1776) An nequalty nto the nature and causes of the wealth of natons. Three volumes, 5th edton, London: A. Strahan and T. Cabell. 27
Thapa, G., Gaha, R., Kaur, S., Kacker, N., & Vashshtha, P. (2013). Agrculture-pathways to prosperty n Asa and the Pacfc. Dscusson papers Seres 17, Rome, the Asa and the Pacfc Dvson, IFAD. van de Walle, D., & Cratty, D. (2004). Is the emergng non-farm market economy the route out of poverty n Vetnam? The Economcs of Transton, 12(2), 237-274. World Bank (2014). World Development Indcators, Washngton DC, World Bank. 28
Table 1 The Results of Treatment Effects Model on the Effects of Indvdual Partcpaton n Rural Non-Farm Employment on Poverty and Vulnerablty for Vetnam 2002 2004 2006 1 st Stage: Dependent Varable Partcpaton n Non-farm sector employment Partcpaton n Non-farm sector employment Partcpaton n Non-farm sector employment Coef. Z value *1 Coef. Z value Coef. Z value Coef. Z value Coef. Z value Coef. Z value Explanatory Varables *2 Predcted Daly Male Wage Rate 0.205 (20.57)* 0.139 (15.78)** 0.017 (4.56)** 0.012 (4.78)** 0.007 (4.22)** 0.003 (3.49)** Predcted Daly Female Wage Rate 0.180 (13.74)* 0.076 (6.53)** -0.006 (-1.56) -0.004 (-1.57) 0.010 (3.94)** 0.006 (3.56)** Whether a head s male -0.170 (-6.12)* -0.128 (-4.05)** -0.064 (-0.82) -0.077 (-1.05) 0.190 (2.30)* 0.112 (1.45) Whether completed prmary school 0.051 (1.47) 0.004 (0.11) -0.323 (-1.08) -0.652 (-2.42)* 0.139 (0.44) 0.090 (0.29) Whether completed lower secondary school 0.260 (7.32)** 0.181 (4.98)** -0.083 (-0.28) -0.361 (-1.34) 0.290 (0.92) 0.284 (0.92) Whether completed upper secondary school 0.259 (6.51)** 0.296 (7.25)** 0.115 (0.38) -0.140 (-0.51) 0.424 (1.34) 0.425 (1.36) Whether completed techncal school 0.347 (7.04)** 0.478 (9.30)** 0.276 (0.91) 0.032 (0.12) 0.619 (1.94) 0.595 (1.91) Whether completed hgher school educaton -0.009 (-0.15) 0.277 (4.35)** 0.330 (1.04) 0.035 (0.12) 0.740 (2.20)* 0.673 (2.08)* Sze of household 0.033 (5.41)** 0.029 (4.56)** 0.031 (2.17)* 0.014 (1.08) 0.049 (3.30)** 0.048 (3.52)** Share of female members 0.023 (0.47) -0.066 (-1.32) -0.068 (-0.54) -0.075 (-0.65) -0.099 (-0.79) -0.114 (-0.99) Dependency Burden (share of household members under 15 or above 60) 0.171 (3.73)** -0.079 (-1.66) 0.020 (0.22) -0.052 (-0.61) 0.171 (1.54) -0.200 (-1.92) Sze of land (hectare) -24.483 (-22.71)** -16.296 (-14.29)** -20.501 (-7.63)** -13.885 (-6.18)** -10.523 (-4.29)** -7.270 (-3.34)** Sze of land squared 30.071 (16.90)** 42.264 (9.77)** 56.908 (5.56)** 42.433 (4.90)** 21.561 (2.59)* 17.278 (2.50)* Age of a household head -0.120 (-23.71)** -0.111 (-20.72)** -0.132 (-10.51)** -0.098 (-8.44)** -0.123 (-8.98)** -0.097 (-7.59)** Age squared 0.001 (25.47)** 0.001 (23.28)** 0.001 (10.88)** 0.001 (9.26)** 0.001 (9.04)** 0.001 (7.97)** Whether a household head s marred -0.122 (-3.81)** -0.100 (-2.86)** -0.181 (-2.01)* -0.032 (-0.38) -0.272 (-3.00)* -0.178 (-2.09)* Whether belongng to ethnc majortes 0.389 (10.53)** 0.383 (9.34)** 0.317 (3.55)** 0.807 (9.62)** 0.187 (2.24)* 0.554 (7.38)** Constant 0.049 (0.35) 0.355 (2.39)* 2.161 (4.68)** 1.082 (2.55) 1.136 (2.25) 0.387 (0.81) ˆ -0.217 (-21.12)** -0.207 (-57.62)** 0.041 (0.47) -0.157 (-45.61)** -0.056 (-0.80) -0.151 (-49.12)** ˆ (-106.47)* -0.473 (-23.89)** -0.795 (-95.06)** 0.103 (0.47) -0.865 * -0.142 (-0.81) -0.879 (-122.40)** 2 nd Stage: Dependent Varable log per capta consumpton Vulnerablty log per capta consumpton 29 Vulnerablty log per capta consumpton Vulnerablty Coef. Z value *1 Coef. Z value Coef. Z value Coef. Z value Coef. Z value Coef. Z value Whether a head s male -0.035 (-3.75)** 0.064 (9.95)** -0.044 (-1.85) 0.022 (1.99)** -0.005 (-0.22) -0.001 (-0.11) Whether completed prmary school 0.120 (10.93)** -0.085 (-12.99)** 0.112 (1.26) -0.076 (-1.90) 0.175 (1.94) -0.144 (-3.70)** Whether completed lower secondary school 0.222 (19.48)** -0.225 (-33.20)** 0.260 (2.97)** -0.192 (-4.77)** 0.270 (2.97)** -0.257 (-6.61)** Whether completed upper secondary school 0.397 (30.68)** -0.338 (-43.49)** 0.439 (4.97)** -0.272 (-6.71)** 0.442 (4.75)** -0.309 (-7.87)**