Impact of Lending to Women on Household Vulnerability

7/31/2019 Impact of Lending to Women on Household Vulnerability

1/23

The Impact of Lending to Women on

Household Vulnerability and Womens

Empowerment: Evidence from India

SUPRIYA GARIKIPATI *

University of Liverpool, UK

Summary. Impact evaluation studies routinely find that lending to women benefits their house-holds. However, a number of them also find that this may not empower the women concerned. This

seemingly paradoxical conclusion is confirmed by our study with respect to a lending program inrural India. We investigate this result by examining a combination of loan-use data and borrower-testimonies. We find that loans procured by women are often diverted into enhancing householdsassets and incomes. This combined with womans lack of co-ownership of familys productive as-sets, we conclude, results in her disempowerment. If empowering women is a crucial objective, thenthe patriarchal hold on productive assets must be challenged. 2008 Elsevier Ltd. All rights reserved.

Key words group lending, household vulnerability, womens empowerment, India

1. INTRODUCTION

Microcredit programs, an increasingly com-mon intervention against poverty, generallytarget poor rural women. The basic argumentbehind lending to women is that they are goodcredit risks, are less likely to misuse the loan,and are more likely to share the benefits withothers in their household, especially their chil-dren. In addition to the economic benefits, itis argued that womens increasing role in thehousehold economy will lead to their empower-

ment. During the past few decades, microcredithas enjoyed tremendous growth and womencontinue to be the major beneficiaries. DuringDecember 1997December 2005, the numberof people receiving microcredit increased from13.5 million to 113.3 million with 84% of thembeing women (Daley-Harris, 2006). It is antici-pated that such programs will contribute to theachievement of the Millennium DevelopmentGoals which, among other things, aim to pro-mote gender equality and empower women(see Kabeer, 2005).

Despite methodological variations, evalua-tion studies fairly widely accept that lendingto women does improve household incomes

and is also linked with other associated benefitslike increased livelihood diversification, morelabor market activity, more education and bet-ter health (see, for instance, Hulme & Mosley,1996, Vols.1 & 2; Khandker, 1998; Morduch& Haley, 2002; Mosley & Rock, 2004; Todd,1996; Zaman, 2004). However, there is littleconsensus regarding the empowerment poten-tial of such schemes and studies make diametri-cally opposite claims. Some find thatmicrocredit has helped women increase their in-come earning capabilities, leading to greater

confidence and ability to overcome cultural

* The author gratefully acknowledges the financial

support received from Department for International

Development (Award No. R7617) and Newton Trust

(Award No. INT 2.05[d]). She thanks Sara Horrell,

Brendan McCabe, Paul Mosley, Stephan Pfaffenzeller,

G.N. Rao, David Sapsford, and three anonymous

referees for useful comments on earlier versions of this

paper. She is deeply indebted to her field research team:

Achari, Chandrasekhar, Lakshmamma, Narsimhulu,

Lakshmi, Padma, Ravi, Rathish, and Sridevi. Please

send comments to [email protected] Final revisionaccepted: November 28, 2007.

World Development Vol. 36, No. 12, pp. 26202642, 2008 2008 Elsevier Ltd. All rights reserved

0305-750X/$ - see front matter

www.elsevier.com/locate/worlddevdoi:10.1016/j.worlddev.2007.11.008

2620


2/23

asymmetries (see, for instance, Hashemi, Schu-ler, & Riley, 1996; Kabeer, 2001; Pitt & Khand-ker, 1998; Pitt, Khandker, & Cartwright, 2006;Rahman, 1986). Others find that loans made towomen are usually controlled by their hus-

bands, leading to womens dependence on themfor loan installments and at times in domesticdissension and violence (see, for instance, Goe-tz & Gupta, 1996; Leach & Sitaram, 2002; Rah-man, 1999). Ignoring the conceptual andmethodological differences among these stud-ies, the suggestion here is that although lendingto women benefits their households, its benefi-cial impact on women themselves is somewhatuncertain. 1 By focusing on a case study, thispaper seeks to unravel some of the reasons be-hind this paradoxical conclusion.

In this study, we separately examine the im-pact of microcredit on beneficiary householdsand on the women concerned. We use datafrom two villages in Andhra Pradesh (AP), In-dia, that participate in the Self Help Group(SHG) program which lends mainly to ruralwomen. We find the same paradoxical resultsthat haunt the microcredit literature: that whilelending to women has helped households acrossincome groups to diversify livelihoods and re-duce their vulnerability to shocks, it has failedto empower the women concerned. We refer

to this result as the impact-paradox andinvestigate the reasons behind it by examininga combination of loan-related data from a sam-ple survey and borrower-testimonies. Our find-ings suggest that womans loan may easily getdiverted into enhancing household assets andincomes but given her lack of co-ownership offamilys productive assets, access to creditmay not result in her empowerment. In such asituation, the household may benefit, but thewoman herself is likely to see further deepeningof the resource division between her and her

husband.The remainder of this paper is organized as

follows. Section 2 briefly discusses Indias ruralfinancial system and Section 3 discusses itsmicrocredit program. Section 4 describes thequestionnaires used in our fieldworks and theresulting data sets used in the empirical partsof this paper. The empirical analysis is carriedout in two parts: in the first part we analyzethe impact of microcredit on household vulner-ability and female empowerment and in the sec-ond part, we examine loan-use and repayment

data to understand the paradoxical results ob-tained. Section 5 presents the empirical modelsused to examine the impact of microcredit on

beneficiary households and the women con-cerned. It also provides the descriptive statisticsof the variables and discusses the outcomes ofthe first part of the empirical analysis. Simi-larly, Section 6 presents the models used to

investigate the impact-paradox. In additionto providing the descriptive statistics of thesample and the results of the second part ofthe empirical work, it also summarizes the tes-timonial evidence collected from the loanee wo-men. Section 7 concludes.

2. A BRIEF OVERVIEW OF INDIASRURAL FINANCIAL SYSTEM

India has a long history of rural credit insti-

tutions. The rural cooperatives were initiatedin 1904 to be a major source of rural finance.These were unable to cope with the steep in-crease in rural credit requirements caused bythe advent of green-revolution in the 1960s. Pri-vately owned commercial banks also playedonly a very nominal role in rural finance, bothin matters of outreach and share. This ostensi-bly led to the nationalization of 14 major com-mercial banks in 1969 which were thencompelled to open rural branches. This markedthe beginning of the state intervention which

became a constant feature in Indias ruralfinancial system.

Intervention was justified mainly on groundsof market-failure, which was also the reason formaking credit an integral component of thestates numerous poverty-alleviation schemes.Handing out credit was largely preferred overother politically sensitive measures like landredistribution and implementation of tenancylaws. State intervention in the banking sector,mainly driven by short-term political gains, re-sulted in policies for bank branching, directed

credit, frequent loan waivers, subsidies, andthe refinancing of loss-making institutions.Although these policies resulted in expansionof commercial banks into rural areas and sig-nificant lending to rural population they alsocontributed to erosion in borrower disciplineand a weakened financial sector (Meyer & Nag-arajan, 2000).

During the 1970s, two major initiatives withsignificant bearing on the rural financial systemwere launched. First, the Regional Rural Banks(RRBs) were established in 1975 as a subsidiary

of the public-sector commercial banks to ser-vice the rural poor so far excluded from formalcredit. This resulted in widening of the

THE IMPACT OF LENDING TO WOMEN 2621
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


3/23

geographical spread and functional reach ofbanks in rural areas and vastly improved accessof rural poor to formal credit (Chavan &Ramakumar, 2002). The average populationcovered by a bank branch declined from

65,000 in 1969 to 12,800 in 2003 (Basu & Sri-vastava, 2005). 2

Second, the Integrated Rural DevelopmentProgram was launched in 1978 [this was re-placed by the Swarnajayanti Gram SwarozagarYojana (SGSY) in 1999]. This was a creditbased poverty-alleviation program imple-mented through the commercial banks targetedat households with income below the povertyline. The program is estimated to have reachedabout 51 million people since its inception butcame under sever criticism mainly on account

of large proportion of non-performing loans(Narasimham Committee, 1991). Loans to thepriority sectors (agriculture and cottage indus-try) were frequently waived, especially duringthe times of elections and this did not help mat-ters since subsequent borrowers expected loanwaivers and did not repay even where theycould (Mahajan & Ramola, 1996).

In order to rationalize the provision of ruralfinancial services, the National Bank for Agri-culture and Rural Development (NABARD)was formed in 1982. This is an apex refinancing

institution for cooperatives, RRBs, and ruralbanks and is mandated to coordinate and buildtheir institutional capacities (Meyer & Nagara-

jan, 2000). Although its creation provided therural financial system with a clear institutionalstructure, it did little to mitigate the inherentweakness that had crept into the system. Theloan recovery rate measured as a percentageof loans collected to total amount due was5060% throughout the 1980s to mid-1990s(NABARD, 1999). By early 1990s, it becameapparent that refinancing of a large number

of loss-making units within the extensive staterural banking apparatus could not continueand that monitoring and enforcing repay-ments could not be sustained in a centralizedsetting.

3. MICROCREDIT IN INDIA

In response to the imminent crisis facing therural financial system and inspired by the glo-bal success of the microcredit movement, the

SHG-bank linkage program was initiated in1992 (Karmakar, 1999). 3 A significant shareof the SHG scheme was later tied into the

SGSY. The program uses the existing extensivestate banking apparatus to provide credit to therural poor while at the same time uses innova-tions like group-lending and peer-monitoringto cultivate the much needed borrower disci-

pline. In this respect, it endeavors to build onthe good aspects of the rural financial sectorwhile finding a solution to the malady of non-performing loans.

An SHG typically consists of around 1015women from poor communities. While thereare some urban groups, the main emphasis ison rural SHGs. Women take advantage of theirsocial networks to come together as an SHG.Group formation is generally facilitated byNGOs or government agencies that arrangemeetings and give information (72% of the

existing SHGs are formed this way). In somecases, the credit institutions may directly facili-tate group formation (20%) and in yet othersthe NGOs may act as both facilitators andfinancial intermediaries (8%) (NABARD,2004). The scheme primarily focuses on creditand there is little explicit attempt to encouragegroup building. Even where NGOs get in-volved, their role is limited to that of facilita-tors rather than capacity builders. The ruleson eligibility are vague but because the savingsand loan amounts are very small, there is little

incentive for the very wealthy to participate inthe program. The group begins its credit activ-ity with members own savings of 1 rupee perday per member, which are collectively usedas a revolving fund to provide loans to individ-ual members. After six months of regular sav-ing, the SHG is eligible to enhance itsrevolving fund by obtaining loans (also grantsand interest-free loans) from NGOs, RRBs,and other financial institutions. These institu-tions are in turn 100% re-financed by NA-BARD. The existing institutional structure is

thus used to link individual SHGs to the ruralfinancial institutions and is popularly referredto as the SHG-bank linkage program.

Table 1 provides the growth rates of theSHG-bank linkage program over the last fewyears. It shows that by March 2007 there wereover 2.5 million SHGs, serving approximately40 million households. Moreover, the numberof SHGs linked to banks was growing at anannual rate of around 90%. This makes it thelargest and fastest growing microcredit pro-gram in the world. The repayment rates by

SHGs have consistently been over 95% whencompared to other rural modalities whichare in the range of 40% (for loans by rural

2622 WORLD DEVELOPMENT
http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


4/23

cooperatives) to 65% (for general loans tothe poor by commercial banks and RRBs)(NABARD, 2003). This gives further causefor euphoria despite the fact that the programremains regionally clustered [AP alone ac-counts for 40% of all SHGs (Bansal, 2003)]and serves a disproportionate number of rela-tively better-off households.

There are a growing number of studiesexploring the economic and social impact of In-dias microcredit schemes. Broadly, the eco-nomic impact is usually examined at the

household level and the social impact at the cli-ent level. Particularly illustrative in the formercategory are the spate of studies sponsored bythe NABARD that use data from a comprehen-sive impact evaluation exercise comprising 223SHGs sampled in 11 states from five differentregions. These studies broadly conclude thatthe SHG-bank linkage program has made a sig-nificant economic impact on its clients(Puhazhendi, 2000; Puhazhendi & Badatya,2002; Puhazhendi & Satyassi, 2000). For in-stance, members are found to have experienced

an increase of 17% in employment, 33% in netincome per household, 72% in assets, and200% in savings per capita post-group forma-tion (Puhazhendi & Satyassi, 2000). 4

Other studies that examine the economic im-pact of microcredit in India focus on NGO-ledinstitutions. A study of 20 microfinance institu-tions reported that on nearly all indictors ofcomparison, clients showed significant gainsover non-clients, with greater impact on poorerhouseholds (EDA, 2005). Comparison ofwealth ranks of non-clients with recent clients,possible in four of the 20 institutions, revealeda movement of client households into less poorwealth categories. However, the study also

finds that 30% of long-term clients remainedpoorsuggesting that the potential benefits ofmicrocredit are not evenly spread. Examiningthe impact of microcredit on the clients ofSHARE in AP, Todd (2001) found that thereis a noticeable shift in their employment pat-ternsfrom irregular, low-paid daily labor tofamily business, with livestock being the mostwidely acquired productive asset.

There is increasing evidence that suggests alinkage between microcredit and womensempowerment in India but the findings are

more mixed when compared to the economicimpact of microcredit. The findings of the NA-BARD sponsored studies mentioned earlieralso claim that SHG clients have experiencedsignificant externalities into personal and socialrelations. These studies carry out most of thesystematic quantitative analysis at the house-hold level only and there is little concerted ef-fort to collate information that might bepertinent for evaluating the programs impacton the women recipients. They neverthelessconclude that women were found to be more

assertive in confronting social evils and familysituations which may have resulted in a fall indomestic violence (Puhazhendi & Badatya,2002; Puhazhendi & Satyassi, 2000). A studyby Swain and Wallentin (2007) uses recall datato compare women from SHG groups withnon-SHG women from five different states ofIndia. They construct several ordinal variablesindicating womens empowerment and com-pare the changes experienced by the two groupsover time. The empowerment indicators includewomens primary activity, access to indepen-dent saving, her hypothetical response to possi-ble verbal, physical, and emotional abuse,awareness of rights, and whether she is politically

Table 1. Growth in volume of SHG-bank linkage program (19992007)

By 31st March Number of SHGs linked tobanks (% change over previous year)

Cumulative bank loans inmillion US$ (% change over previous year)

1999 32,995 (130.48) 13.57 (112.03)2000 114,775 (247.85) 44.53 (228.15)2001 263,825 (129.86) 105.26 (136.38)2002 461,478 (74.92) 215.20 (104.45)2003 717,306 (55.45) 455.00 (111.43)2004 1,079,091 (50.43) 867.00 (90.55)2005 1,618,456 (49.98) 1900 (119.15)2006 2,238,565 (38.31) 2850 (50)2007 2,894,505 (29.30) 4493 (57.63)

Sources: World Bank (2003), Sankaran (2005) and NABARD (2006, 2007).

http://-/?-http://-/?-http://-/?-


5/23

active. Their findings indicate that while bothgroups have become more empowered overtime, the change for the SHG members is dra-matic.

Other studies examine the impact of NGO-

led microfinance programs on womensempowerment. For instance, the study byEDA (2005) remarks on the supportive ap-proach of the microfinance institutions whichmay help in the capacity building of womenmembers via social networking. It, however,finds that cultural burden may restrain the po-tential for womens empowerment, rather moreemphatically in the north of the country ascompared to the south. Hunt and Kasynathan(2001) investigated three NGOs in Bangladeshand one in Bihar that use microcredit to em-

power women. They conclude that if credit pro-grams are to support empowerment, then theremust be a greater emphasis on strategies thattransform gender relations. Leach and Sitaram(2002) examine an NGO-led credit program forthe scheduled caste women working in Indiassilk-reeling industry. They highlight the nega-tive consequences of excluding male relativesfrom having a meaningful role and concludesthat antagonizing men can ultimately be detri-mental to female empowerment. A study byHolvoet (2005) investigates the importance of

borrowers gender and different lending tech-nologies for one dimension of empowerment:decision-making agency. She compares direct-bank lending to individual men and womenby IRDP with group-based NGO supportedschemes that lend to women. Her findings sug-gest that credit delivery to individual womenalone is insufficient to produce a substantial im-pact on decision-making patterns and that it ismost beneficial when channeled through wo-mens groups and combined with technicaland social awareness training.

Overall the impact evaluation literature thatis emerging from India once again reiteratesthe central paradox that we attempt to investi-gate in this paper; while the economic benefitsof microcredit at the household level are some-what predictable (if not guaranteed), the bene-fits for women clients are much moreambiguous and may depend on other factorsexogenous to lending.

4. THE DATA

During 200103 we carried out fieldwork intwo villages, Vepur and Gudimalakapura, of

the Mahabubnagar district in the southern stateof AP, India. Mahabubnagar is a compellingcase study because it has one of the oldest, big-gest and fastest growing SHG programs in thestate of AP (NABARD, 2003). Being

drought-prone, it is also one of the poorest dis-tricts of AP, with 45% of its rural householdsliving below the poverty line (Government ofAP, 1996). The state government has pursuedthe SHG program as part of its poverty-allevi-ation strategy with the primary objective ofhelping households to diversify incomes.

With regard to generalization of survey re-sults, there are at least two reasons why thismay be possible. First, the SHG-bank linkageprogram is Indias largest microcredit schemeand the organizational structure and the rules

surrounding eligibility are very similar acrossthe country. Our survey villages, moreover, fol-low the most common linkage modality where-by SHGs formation is facilitated by the NGOswithout involvement from the credit institu-tions. Second, our survey villages are in thestate of AP, which is widely acknowledged asthe undisputed leader of Indias microcreditmovement. The achievements in AP are putforward as exemplary and are considered worthreplicating elsewhere in the country (see NA-BARD, 2004). Hence, a careful impact assess-

ment is essential from the policy point of view.During 2001 and again in 2002, we conducted

detailed surveys among 291 married couplehouseholds from the two villages, of which117 participated in the SHG program (com-pleted at least one loan cycle) and the remain-ing 174 although eligible were not in theprogram. 5 In the surveys, we asked questionsabout the socio-economic characteristics ofthe household and details of its economic activ-ities. In addition, we included questions onmale and female asset holdings, time-use and

household decisions. We randomly interviewedeither the head of the household or his/herspouse such that equal number of men and wo-men were consulted. 6 On average, householdsconsisted of 6.20 members, ranging from 2 to21 members. The average landholding was2.50 acres, with the maximum holding of13.00 acres. Although on an average 63.02%of the household income was from agricultureand related wage work, there is a clear trend to-ward income diversity with 59.31% of thehouseholds receiving over 1/4th of their in-

comes from off-farm sources (mainly fromseasonal off-farm work and livestock). Theaverage monthly net per capita income was



6/23

206.45 Rs. which is considerably below themonthly per capita poverty threshold of262.90 Rs. for rural AP (Planning Commission,2001). 7 Around 60.32% of our survey house-holds fall below this threshold. Data from this

survey are used to analyze the SHG programsimpact on household vulnerability and femaleempowerment.

During 2002, we also conducted a surveyamong 27 SHGs (which had completed at leastone loan cycle) from the same villages and ob-tained information from 397 group membersalso from married couple households (this in-cluded 106 of the 117 SHG member householdsinterviewed earlier). This survey was carriedout mainly with the objective to investigatethe paradoxical findings that emerged from

the fieldwork mentioned earlier. In this survey,we asked questions about the socioeconomiccharacteristics of the respondents and theirhouseholds, as well as details about the use,control, and repayment of their most recentloan. On average, groups were composed of14.70 members and had completed an averageof 3.78 loan cycles, ranging from a minimumof one to a maximum of six cycles. Loan termsvaried from 6 to 24 months and the averageloan amount received by a group was26138.20 Rs. and ranged from 18,000 Rs. to

91,500 Rs. Only occasionally did loan amountsvary from cycle to cycle. Loans were usually di-vided equally among group members and in

just two SHGs did members pool their loansfor investment in a group project. Individualloans were mainly used to meet householdsproductive and consumption requirementsand in some cases to finance self-managedenterprises. Repayment rate was reported tobe 100%. The average landholding amongmembers was 2.50 acres, with the maximumholding of 13.50 acres. The average monthly

net per capita income was 219.61 Rs. and52.10% of the respondents fall below the pov-erty threshold. Data from this survey are usedto investigate the findings on household vulner-ability and female empowerment.

Finally, during 200203, we also carried outseveral individual and focus group interviewswith borrowers who had also participated inthe above survey(s). These interviews were typ-ically unstructured and were designed to cap-ture the nuances behind several discernibleexperiences within borrower groups. Data from

these interviews are also used to further ourunderstanding of the findings on vulnerabilityand empowerment.

5. LENDING TO WOMEN, HOUSEHOLDVULNERABILITY, AND FEMALE

EMPOWERMENT

(a) The empirical models and description of the

variables

As mentioned in the introduction, studiesroutinely find that lending to women benefitstheir households but whether the women them-selves benefit is a much more debated issue. Inthis section, we separately investigate the im-pact of Indias microcredit program on recipi-ent households and women from twoparticipating villages. More specifically, weevaluate the programs impact on householdsvulnerability to crises and on womens empow-

erment by comparing the 117 participants with174 non-participants. We use five vulnerabil-ity and seven empowerment logit modelsto estimate the effects of independent variablesmeasuring program participation in reducinghousehold vulnerability and enhancing femaleempowerment, respectively. The dependentvariables in these models are measures ofvulnerability and empowerment. These havebeen constructed similarly to Hashemi et al.s(1996) empowerment indicators. We firstdescribe these measures and then the indepen-

dent variables used in the empirical models.Measures of vulnerability and empowerment

are highly contextual and indicators relevant toa certain society may be of little consequence toanother. With this in mind we developed a ser-ies of detailed questions relating to various as-pects of vulnerability and empowermentrelevant to the particular situation in the surveyvillages. In the end while some of the measuresused were specific to the survey villages, mosthad a much wider appeal. The responses tothese questions were collated to construct the

vulnerability and empowerment indicators.Each indicator consists of a number of compo-nents. To minimize subjectivity, as far as possi-ble, we have assigned equal weights to allcomponents. If the condition(s) set out in thecomponent were satisfied, one point (or two,if weights were used) was given to the house-hold or the woman, as appropriate. The finalscore was calculated by adding the points se-cured on all components within the variable.For each variable, a cut-off point was decidedand all observations with a score equal to the

cut-off point or better were classified as not-vulnerable or empowered as appropriateand coded as one while the remaining were

http://-/?-http://-/?-http://-/?-


7/23

coded as zero. Hence, the variables used in theanalysis were reduced to dichotomous variableswith a score of one or zero. In choosing the cut-off points for each variable, we attempted todistinguish between relatively less vulnerable

households and more empowered women thanmost others in similar situations, rather thanonly identifying those at the extremes. Thecut-off points for most variables were made ataround 30th to 35th percentile.

(i) Vulnerability indicatorsOne of the big problems that poor house-

holds encounter in the survey area is vulnerabil-ity to the risk of drought which dishevels theiralready limited coping strategies. Recurrent

exposure to drought is likely to impact on agri-cultural output and hence on the householdsability to feed and maintain the health of itsmembers during lean periods. The vulnerabilityindicators developed here focus on the house-holds ability to cope with drought in the shortrun and on its ability to diversify away fromagricultural incomes in the long run. Theseare described below.

Drought-related vulnerability (DROUGHT):One point was given if, during the lastdrought, the household met all its food

needs, one point if it met all its health needs,one point if no livestock or other assets weresold, and one point if none in the housemigrated (excludes routine seasonal migra-tion for off-farm work). An additional pointwas given for each category if respondentexpected the household to cope similarly ina future drought. One point was given ifincome enhancing plans were not postponedbecause of drought in the last three years. Ahousehold with a score of six or better wasclassified as not-vulnerable and coded as

one. Livelihood diversification (DIVERSE):One point was given if the householdreceived income from a non-agro business,one point if it received income from live-stock, and one point if it received incomefrom non-farm labor work. An additionalpoint was given in each category if approx-imately at least a quarter of its incomecame from this source. One point wasgiven if the household was expected tocope with its main earner out of work. A

household with a score of two or betterwas classified as diversified and codedas one.

Entrepreneurial behavior (ENTER-PRISE): One point each was given for leas-ing in extra land, one point for investing inirrigation, one point for investing in newfarm equipment, one point for investing in

draught animals, and one point for investingin a new business or upgrading an existingbusiness. Only investments in the last threeyears were considered. One point was alsogiven for regular use of hybrid seeds andone point for non-organic fertilizers. Ahousehold with a score of three or morewas classified as enterprising and codedas one.

Investment in and access to social capital(SOCIAL): One point was given if thehousehold provided childcare and livestock

care for neighbors (without explicit pay-ment), one point for receiving such support,one point if neighbors were helpful in find-ing waged work, one point if householdwas positively affected by an auxiliary pro-gram like forest conservation andwatershed. A household with a score oftwo or more was classified as having accessto social capital and coded as one.

Composite not-vulnerable (NOTVUL): Ahousehold was classified as not-vulnerableif it had a positive score on two or more of

the indicators described above.

(ii) Empowerment indicatorsIn rural India, female empowerment is still

largely an elusive concept and it is commonto find discourses that conflate womens wel-fare with households welfare. Hence, in-stead of working with an exogenously deriveddefinition of empowerment we attempted tounderstand its constituents for the context ofour survey villages. The chances of capturinga notion of empowerment using a structured

survey are at best limited and hence the indica-tors used here have been developed through along process of interaction with enumeratorsfrom the survey villages and reflect the realitiesof womens lives in rural AP. In general, wo-men here control few productive assets andhave little or no say in major household deci-sions. However, they are not expected to followthe norms of purdah and face few mobilityrestrictions within their village and local mar-kets. Most women also contribute substantiallyto family incomes. They work as wage laborers,

work on family farms, run small businesses,and some even undertake seasonal migration.Women in the survey villages were found to



8/23

be heavily involved in agricultural wage labor-ing when compared to men (81.9% of womenin our sample were farm laborers when com-pared to only 68.8% of men). Men, on the otherhand, mainly worked on own assets or as non-

farm wage laborer. This occupational differencebetween men and women is significant giventhat farm laboring is associated with undesir-able characteristics like hard menial labor,low pay, and negligible ability to negotiate overworking conditions and hence considered infe-rior to work on own assets or off-farm wagework (see Chowdhry, 1994; da Corta & Ven-kateshwarlu, 1999; Garikipati, 2008). Womenin our survey villages, it seems, were less ableto allocate their work time in a favorable waywhen compared to men. This is at least partly

the reason why contributing to family supporthas not helped women challenge the culturalnorms which, among other things, expect themto attend to all the household chores and carefor family members without much assistanceeither from their husbands or in-laws.

Given these conditions, in constructing theempowerment indicators we focus on four spe-cific aspects: her ownership and control overhousehold assets and incomes, her say in house-hold decisions, allocation of her work time, andher ability to share household chores. These

facets closely reflect the conceptual thinkingaround the notion of womens empowerment.They capture the fairly widely accepted viewthat empowerment comprises three essentialelements: preconditions, processes, and out-comes. The idea here is that empowerment re-quires preconditions or resources which canfacilitate the processes that expand womensagency or ability to make choices which in turndetermine outcomes that have direct implica-tions for their welfare (Kabeer, 1999; Malhotra& Schuler, 2005). The empowerment measures

operationalized from the survey data are de-scribed below.

Ownership of household assets andincomes (ASSETS): One point each wasgiven if the woman owned the family homeone point if she owned any agricultural land,and one point if she owned any livestock(excludes poultry). Two points were givenif she contributed approximately at least aquarter of the household income (includesimputed income from work on family farm).A woman with a score of two or more was

considered empowered and coded as one. Control over minor finances (MINFIN):One point was given if she kept the money

from sale of livestock produce, one pointfrom sale of poultry, one point if she hadany regular personal spending money, andone point for having money for emergencyuse. A woman with a score of two or better

was coded as one. Control over major finances (MAJFIN):One point was given if she retains the moneyfrom the sale of crops, one point for moneyfrom sale of goats, one point for retainingher own wage earnings, one point for chil-drens wages, and two points for husbandswages. A woman with a score of two or bet-ter was coded as one.

Say in household decisions (DECI-SIONS) 8: One point was given if thewoman decided (individually or jointly with

others) about childrens education, onepoint for deciding on what crops to grow,one point for deciding to lease in/out agri-cultural land, one point for making a majorfinancial decision (open a bank account,apply for a loan, and so on). One additionalpoint was given for initiating the financialdecision. One point was given for decidingto sell crops and one point for deciding tobuy/sell large livestock and one point fordeciding to buy agricultural inputs. An addi-tional point was given in each category for

participating in the sale negotiations. Awoman with a score of three or better wasclassified as empowered.

Work time allocation (WORKTIME):One point was given if the woman managedor helped manage any business, one pointfor work on family farm, and one point fornon-farm wage work. One additional pointif any one of these was also her primarywork and one point if she did not want tochange the way she spent her work time. Awoman with a score of two or better was

coded as one. Division of domestic chores (CHORES):One point was given if the woman sharedthe tasks of fuel gathering and preparingwith others in the family (expect with daugh-ters), one point for water collection, onepoint for sweeping and cleaning, one pointfor cooking, one point for washing utensils,and one point for washing clothes. A womanwith a score of three or better was coded asone.

Composite empowerment (EMPOWER):

A woman was considered empoweredand coded as one if she had a positive scoreon three or more of the above indicators.

http://-/?-http://-/?-


9/23

(iii) Independent variablesThree sets of independent variables were in-

cluded in the regression analysis: those relatingto the credit program; control variables mea-suring household characteristics and those

measuring womens personal characteristics.The reason why we include control variablesis that personal characteristics may influencethe measures of vulnerability and empower-ment. The independent variables are describedbelow.

DURATION: Indicates the length ofmembership of the SHG in years. Non-members are coded as zero. 9

HHHSEX: Coded as one if the head ofthe household is female.

HHHAGE: The age of the head of the

household. HHHEDU: Is a categorical variable indi-cating the educational background of thehousehold head. It takes the values 0, 1,and 2 (where 0 = illiterate, 1 = secondaryschool or less, and 2 = high school ormore). 10

HOUSE: Coded as one if the outer wallof the house is made of concrete and thehouse has a durable roof (tiles or other syn-thetic materials) and zero otherwise. Thisvariable indicates the relative economic sta-

tus of the household. LABORSHARE: Household membersaged 13 or over as percentage of totalnumber in household divided by house-hold size. This number indicates thehouseholds demand for credit as well asgeneral pressure on resources. A low sharedenotes greater demand for credit andother resources.

OLOAN: Coded as one if the householdreceived credit from other sources in the lastthree years.

CASTE: Coded as one if the householdis from the Scheduled Caste or ScheduledTribe (low caste) and zero for all othercaste.

VILLAGE: Coded as one if household isfrom Vepur and zero if it is fromGudimalakapura.

WOMAGE: The age of the woman inyears.

WOMEDU: Is a categorical variableindicating the educational background ofthe woman similar to that of the household

head. MALECHILD: Coded as one if thewoman has one or more sons.

For DURATIONwe expect a positive sign ofthe coefficient in all the models: as length ofSHG membership increases, the probability ofinsulating the household against crises increasesand so does the probability of empowering wo-

men. In the vulnerability models, for HOUSEwe expect a positive sign of the coefficient: if thehousehold enjoys better economic status, theprobability of vulnerability to weather-relatedshocks decreases while the probabilities of diver-sifying livelihood and accessing social capital in-crease. In the empowerment model, forWOMEDUand MALECHILDweexpectaposi-tive sign of the coefficient: as womans educa-tional background improves and if she has malechildren, the probability of her enjoying betterstatusincreases. We have no explicit expectations

on the signs of the remaining variables.Note that given the possibility of selection

bias, DURATION is not used directly but isestimated using the instrumental variable (IV)technique. 11 Given the nature of the data andrules surrounding SHG formation we were ableto identify two instruments: (i) CLUSTER-SIZE: The approximate size of the respon-dents neighborhood cluster and (ii)MINORCASTE: A dummy variable coded asone if respondent belongs to a caste other thanthe dominant caste within the cluster (defined

as the one with the largest membership).Given that 15 members are required to form

an SHG, women from bigger neighborhoodclusters are more likely to form one. Womenmay prefer to group with others living closeby to minimize the transaction costs associatedwith screening and monitoring group members.Also, belonging to the dominant caste within acluster increases the probability of forming agroup and vice versa. This may be due to rea-sons of trust and cultural affinity. Neighbor-hood cluster maps were constructed using thevillage electoral lists. This information was thencombined with the precise household locationto identify the cluster for each household inthe sample. This information was also used toidentify the dominant caste in the cluster. In12.59% (N = 397) of the cases did SHG mem-bers not belong to the same cluster as themajority in their group and in 17.63% ofthe cases did they belong to a caste other thanthe dominant one in the cluster. 12

A two-stage estimation procedure was em-ployed because we have multiple instru-ments. 13 Given that DURATION is limitedto taking non-negative values, we select a

http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


10/23

first-stage estimation procedure such that apositivity condition could be imposed. We usea tobit model to estimate DURATION andpredict its observed values in the first-stage.As the control variables used in the vulnerabil-

ity models differ from those used in the empow-erment models, DURATION was estimatedseparately for both types of models. Table A1(Appendix A) reports the first-stage regres-sions. In the second-stage, the regressions ofinterest are estimated as usual, except thatDURATIONis replaced with its approximationDURATION(est) as estimated in the first-stage.Because we use estimated coefficients to predictDURATION we need to bootstrap the stan-dard errors. We compute standard errors usingup to 10,000 replications of the bootstrap for

each model.14

Two additional points pertinent to estimationprocedure are noteworthy. First, high correla-tion between variables of interest meant thatsome had to be dropped. For instance, therewas high correlation between DURATIONandVILLAGE (r = 0.148, p = 0.011) and betweenHHHAGE and HHHEDU (r = 0.204,p = 0.000). In each case, we use the likelihood-ratio test to decide on which of the correlatedvariables to delete. 15 Second, because of highcorrelation between HHHSEXand WOMAGE

(r = 0.119, p = 0.043) and HHHAGEand WO-MAGE(r = 0.499, p = 0.000); the variables rel-evant to the head of the household are used inthe vulnerability models only and those relevantto womens personal characteristics are used inthe empowerment models only. These choicesare consistent with the likelihood-ratio tests.

(b) Data description and empirical results

Table 2 provides the descriptive statistics of allthe variables used in the regression models for

both the SHG (N = 117) and the non-SHGhouseholds (N = 174). With respect to vulnera-bility indicators the first panel of the table showssignificant differences for DROUGHT, DI-VERSE and NOTVUL. For these variables,the t-statistic of comparing the mean of theSHG households versus other households differssignificantly. The SHG households are less vul-nerable to drought, are more diversified, andcan be considered somewhat less vulnerableoverall as compared to the averages of thesethree variables for other households. With re-

spect to the empowerment indicators, the secondpanel of the table shows significant differencesfor ASSETS, MINFIN, MAJFIN, WORK-

TIME, and EMPOWER. While the t-statisticfor ASSETS is positive, it is negative for theother four variables. Although the women inthe SHG households are more economically se-cure, they are less able to spend their work time

in a favorable way, they exert lesser control overminor and major household finances, and can beconsidered somewhat less empowered overall ascompared to the averages of these variables forthe other women. With respect to the controlvariables, the third panel of the table shows thatwhen comparing the characteristics of the SHGhouseholds with others, the only significant dif-ferences are with respect to the variable VIL-LAGE. This implies that the SHG householdsare more concentrated in the village Vepur ascompared to the averages of these variables for

the non-SHG ones. The final panel similarlyshows that none of the variables measuring wo-mans personal characteristics differ much whencomparing the SHG and non-SHG households.This may at least partly be because all eligiblewomen from the participating villages areencouraged to form SHGs.

Table 3 presents the results of the second-stage logit models that examine how exposureto the credit program impacts on householdvulnerability. Each column represents a sepa-rate model and the Z-statistics are given

between parentheses. Our results indicate thatthe length of SHG membership plays a role inreducing households vulnerability based onthese indicators. In particular, we find statisti-cally significant coefficient for DURA-TION(est1) in (3-1), (3-2), and (3-5). All threevariables have the expected sign. Of the controlvariables, we find statistically significant coeffi-cients for HHHAGE and ENTERPRISEin (3-2) and for CASTE in (3-3) and (3-5).

With respect to exposure to the SHG pro-gram, the results suggest that as length of par-

ticipation in the credit program increases, theprobability of the household coping withdrought and diversifying income increases. Thisalso increases the probability of its overall pre-paredness for crises as measured by the com-posite vulnerability score.

Table4 provides the results of the second-stagelogit models that examine the affect of the creditprogram on female empowerment. The resultsindicate that participation in the credit programdelimits womens status. In particular, the tableshows that the coefficient for DURATION(est2)

is statistically significant in (4-3), (4-5), (4-6), and(4-7). It has the wrong sign in all these models. 16

Of the control variables, we find statistically sig-

http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


11/23

nificantcoefficients for HOUSEin(4-1)and(4-7)and for WOMEDU(2) in (4-4). We also find that

the coefficient for HOUSEis almost statisticallysignificant in (4-2) and (4-6).

With respect to exposure to the SHG pro-gram the results suggest that as the length ofprogram membership increases, the probabilityof her control over major household finances isreduced. Increase in the length of membershipalso reduces the probability of womans worktime being allocated in a favorable way andthat of her sharing domestic chores with others.It also reduces the probability of her overallempowerment as indicated by the compositescore. With respect to the control variables,the results suggest that if the family enjoys bet-ter economic status, the probabilities of wo-

mens ownership of its assets and minorfinances increase, as do the probabilities of

her sharing domestic chores with others andof her overall empowerment. Taken together,these results suggest that households economicstatus rather than the length of SHG member-ship helps enhance womens relative power.

Overall our estimates indicate that whilelending to women helps their households diver-sify and strengthen their coping strategies, itmay have a perverse impact on their own rela-tive status. These results compare with theoverall suggestion that emerges from the evalu-ation literature discussed in the introduction.We refer to these results as the impact-para-dox, and investigate the possible reasons be-hind them in the next section.

Table 2. Descriptive statistics of the variables used in the vulnerability and empowerment models

SHG households (n = 117) Control households (n = 174)

Mean Standard deviation Mean Standard deviation t-Statistica

Dependent variables

(1) Vulnerability indicatorsDROUGHT 0.40 0.49 0.26 0.44 2.54**

DIVERSE 0.40 0.49 0.24 0.43 2.98***

ENTERPRISE 0.37 0.48 0.34 0.48 0.50SOCIAL 0.42 0.50 0.34 0.48 1.27NOTVUL 0.57 0.50 0.39 0.49 3.09***

(2) Empowerment indicatorsASSETS 0.45 0.50 0.34 0.48 1.95*

MINFIN 0.27 0.45 0.37 0.49 1.81*

MAJFIN 0.31 0.46 0.41 0.49 1.86*

DECISIONS 0.42 0.50 0.36 0.48 0.97WORKTIME 0.33 0.47 0.47 0.50 2.44**

HHCHORES 0.31 0.46 0.33 0.47

0.36EMPOWER 0.22 0.42 0.31 0.46 1.69*

Household characteristicsHHHSEX 0.04 0.20 0.09 0.28 1.53HHHAGE 44.51 9.82 45.81 11.90 1.02HOUSE 0.16 0.37 0.15 0.36 0.29LABORSHARE 16.22 10.45 18.02 10.12 1.47OLOAN 0.27 0.44 0.25 0.44 0.23CASTE 0.33 0.47 0.26 0.44 1.25VILLAGE 0.62 0.49 0.43 0.50 3.29***

Womans personal characteristicsWOMAGE 35.74 11.53 33.94 11.13 1.33WOMEDU 0.70 0.58 0.68 0.55 0.34MALECHILD 0.86 0.35 0.87 0.33 0.46

a t-Statistic refers to comparing mean values of variables for SHG and control group households.* Significant at the 10% level.** Significant at the 5% level.*** Significant at the 1% level.



12/23

Before that, we test for the robustness of ourresults in three ways. First, we check whetherthe results are robust to the estimation proce-

dure used. We re-estimate each of the modelsusing ivprobitthis is a two-stage estimationprocedure that fits models with dichotomousdependent variables where one (or more) ofthe regressors is endogenously determined. Inthe first-stage the endogenous regressor isinstrumented using ordinary least squares andin the second-stage a probit model is used toestimate the main regression (hence the nameivprobit). Although ivprobit is a readymade two-step procedure, the drawback ofusing it for this study is that we cannot usefullypredict the observed DURATION in the first-stage due to the positivity condition on thedependent variable. As mentioned earlier, it is

for this reason that we use a tobit model to esti-mate DURATION. The ivprobit method, how-ever, is useful to check our results for

robustness to estimation procedure. Using thismethod suggests that our results for the impactof duration of SHG membership on householdvulnerability and female empowerment are ro-bust. There is, however, one important excep-tion. The ivprobit results suggest that thecoefficient for DURATION(est2) is statisticallysignificant in (4-2) while the results using tobitin the first-stage as reported earlier do not con-firm this relationship. In both cases, however,the variable retains its negative sign. This sug-gests that while our specific results may be sen-sitive to the estimation procedure, the inferredqualitative characteristics are mutually consis-tent.

Table 3. Logit estimation of determinants of household vulnerability: Second-stage (N = 291)

Dependent variables: Vulnerability indicators

3-1 3-2 3-3 3-4 3-5DROUGHT DIVERSE b ENTERPRISE SOCIAL NOTVULc

Program-relatedvariableDURATION(est1) 0.195 (3.89)***a 0.194 (3.75)*** 0.005 (0.10) 0.010 (0.20) 0.141 (2.80)***

HouseholdcharacteristicsHHHSEX 0.377 (0.67) 0.027 (0.04) 0.563 (0.91) 0.543 (0.92) 0.646 (1.15)HHHAGE 0.015 (1.28) 0.039 (2.73)*** 0.003 (0.27) 0.003 (0.22) 0.012 (0.93)HOUSE 0.342 (0.80) 0.456 (1.17) 0.361 (1.01) 0.309 (0.83) 0.112 (0.31)LABORSHARE 0.002 (0.12) 0.013 (0.93) 0.001 (0.07) 0.015 (1.17) 0.011 (0.89)OLOAN 0.219 (0.72) 0.276 (0.76) 0.101 (0.34) 0.024 (0.08) 0.068 (0.24)CASTE 0.040 (0.13) 0.386 (1.14) 0.717 (2.34)** 0.216 (0.75) 0.604 (2.14)**

Other variablesENTERPRISE 0.544 (1.87)*

CONSTANT 1.907 (2.93)*** 0.164 (0.23) 0.299 (0.48) 0.733 (1.16) 0.173 (0.27)Observations withdependent = 1

92 88 102 109 135

Number ofbootstrapreplications

9991 9980 9936 9974 9992

Log likelihood 172.293 162.327 184.173 190.449 192.646a Z-statistics are given between parentheses.b DURATION(est1) was estimated separately for (3-2) since it includes ENTERPRISE as one of the explanatoryvariables. ENTERPRISE is used as an explanatory variable in (3-2) because whether a household is enterprising or

not can have a bearing on the extent of its livelihood diversification and not including a variable that measures thiscan lead to an overestimation of the impact of the credit program.c The results were robust even when any one of the vulnerability indicators was disregarded in the computation ofNOTVUL. This could at least partly be attributed to significant correlation between DROUGHTand ENTERPRISE(r = 0.120, p = 0.041) and DIVERSE and ENTERPRISE (r = 0.128, p = 0.029).* Significant at the 10% level.** Significant at the 5% level.*** Significant at the 1% level.



13/23

Table 4. Logit estimation of determinants of female empowerment: Second-stage (N = 29Dependent variables: Empowerment indicators

4-1 4-2 4-3 4-4 4-5 ASSETS MINFIN MAJFIN DECISIONS WORKTIME

Program-related variableDURATION(est2) 0.027 (0.54)a 0.071 (1.33) 0.093 (1.73)* 0.009 (0.19) 0.103 (1.91)

Household characteristicsHOUSE 0.631 (1.71)* 0.591 (1.61) 0.426 (1.17) 0.165 (0.43) 0.128 (0.34) LABORSHARE 0.004 (0.28) 0.007 (0.48) 0.014 (1.06) 0.008 (0.58) 0.007 (0.55) OLOAN 0.248 (0.84) 0.188 (0.59) 0.098 (0.33) 0.004 (0.01) 0.215 (0.73)CASTE 0.116 (0.40) 0.057 (0.19) 0.160 (0.54) 0.195 (0.67) 0.406 (1.42)

Womans personal characteristicsWOMAGE 0.005 (0.47) 0.008 (0.67) 0.018 (1.45) 0.003 (0.27) 0.007 (0.52)WOMEDU(1) 0.138 (0.51) 0.058 (0.21) 0.155 (0.55) 0.075 (0.27) 0.287 (1.06)WOMEDU(2) 0.336 (0.47) 0.298 (0.43) 0.481 (0.72) 1.834 (2.59)** 0.369 (0.48)MALECHILD 0.229 (0.59) 0.036 (0.09) 0.075 (0.18) 0.015 (0.04) 0.002 (0.00)CONSTANT 0.641 (1.03) 0.265 (0.40) 1.143 (1.67)* 0.289 (0.45) 0.315 (0.48)Observations with dependent = 1 112 97 108 112 119 Number of bootstrap replications 9981 9806 9985 9516 9827 Log likelihood 190.679 181.987 186.588 188.303 191.213

a Z-statistics are given between parentheses.***Significant at the 1% level.**

Significant at the 5% level.* Significant at the 10% level.


14/23

Second, we test for the robustness of the var-iable DURATION(est1) and (est2) using back-ward stepwise regression which begins with afull model (reported) and eliminates variablesin an iterative process. The fit of the model is

tested after the elimination of each variable toensure that the model still adequately fits thedata. When no more variables can be elimi-nated from the model, the analysis is complete.We use the likelihood-ratio test to decide ondeletion of variables. 17 Stepwise regressionshows that the values of our coefficients forDURATION(est1) and (est2) remain relativelystable through the deletion process suggestingthat our conclusions regarding the implicationsof participation in the credit program are ro-bust.

Finally, we check whether our findings holdat the level of the components that were usedto construct the vulnerability and the empower-ment indicators. We repeat the two-stage pro-cedure with the individual components usedto construct the dependent variables for modelsin which the coefficient for DURATIO-NON(est) was found to be statistically signifi-cant. Broadly, this gives us the same storyline,that is, models in which coefficient for DURA-TION(est) was found to be statistically signifi-cant retained the sign of the main models.

6. LOAN-USE AND ISSUESSURROUNDING REPAYMENT

(a) The empirical models and description of thevariables

As mentioned before, in this section we inves-tigate the paradoxical results obtained earlierby closely examining loan-related data collectedfrom 397 SHG members. Women in our sample

use their loans in broadly four different ways:as working capital in family farm or enterprise(FAMFARM), to purchase or improve familyland (LAND), toward household maintenance(CONSUME), and in enterprises that theymanage or help manage (OWNBUSINESS).According to our data, 79.35% of the loansprocured by women were diverted into house-hold activities. Loans were primarily used infarms or businesses controlled by their hus-bands (57.18%). Loans were also used to buyor improve land (10.08%in all cases land

was bought in husbands name except in onecase where gold was purchased) and to meethouseholds consumption needs (12.09%). This

suggests that the demand for credit within thehousehold, both for productive and for con-sumption purposes, is high and that householdsare able to divert womens loans into suchactivities. 18 Also, loans procured by women

are mainly used to enhance or create assets con-trolled primarily by their husbands, indicatingthat lending to women may actually amplifythe existing resource divide between men andwomen. A mere 20.66% of the loans were usedin enterprises that women manage or help man-age. Some of these were managed jointly by theSHG (37%) but the majority was controlled byindividual women (63%). 19

We use a multinomial logit model withOWNBUSINESS as the reference category toestimate the effects of independent variables in

determining loan-use. In addition to the inde-pendent variables described earlier, we usetwo further program-related variables.

PEERP: Indicates the percentage ofwomen in the respondents group who usetheir loans for an enterprise they manageor help manage. The higher this percentagethe greater is the (passive) peer-pressureshe faces for investing similarly.

CONTROL: One point was given if therespondent decided (individually or withothers) to join the SHG, one point for

deciding on loan-use, one point for decidingon marketing aspects of the loan, one pointfor maintaining/helping with loan accounts,one point if she controlled income fromloan enterprise. A woman with a score ofthree or more was considered to be in sig-nificant control of her loan and was codedas one.

Given that OWNBUSINESSis the referencecategory for PEERP and CONTROL we ex-pect a negative sign in (6-1), (6-2), and (6-3):women who encounter peer-pressure and are

in control of their loans are less likely to useit in family enterprise or for purchasing landand consumption when compared to the prob-ability of using it in self-managed business. ForHOUSE, we expect a negative sign of the coef-ficient in (6-3): as the households economic sta-tus improves, women are less likely to use loansfor consumption. We have no expectations onthe signs of the remaining variables. 20

The estimation procedure is similar to thatoutlined earlier. Table A1 (Appendix A) re-ports the first-stage regression. Once again we

use the bootstrap procedure to correct for thestandard errors. As before, the data are resam-pled 10,000 times.

http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


15/23

(b) Data description and empirical results

Table 5 provides the descriptive statistics ofall the variables used in the empirical analysisfor the 397 SHG women. With respect to pro-

gram-related variables, the table shows thaton average nearly 20% of a groups memberswill use their loans for self-managed enterprise.This pattern, however, varies widely acrossgroups. In some groups, not a single womanuses her loan in own enterprise while in somegroups nearly 3/4th of the members do so. 21

We also find that just over a third of the womenretain significant control over their loans. 22

Table 6 provides the results of the empiricalinvestigation that examines the determinantsof loan-use. Each column in the table presents

the results for a separate loan-use category.As before, the Z-statistics are given betweenparentheses. Our results indicate that peer-pres-sure and control over loan play a crucial role indetermining loan-use. In particular, we find sta-tistically significant coefficients for PEERP in(6-1), (6-2), and (6-3) and for CONTROL in(6-1) and (6-3). We also find that the coefficientfor CONTROL is almost statistically significantin (6-2). All the variables have the expectedsign. Of the control variables we find statisti-cally significant coefficients for HOUSE in (6-

1) and for WOMAGE in (6-3).With respect to peer-pressure, the results

show that as the number of women from therespondents group who use their loans forself-run enterprise increases, the less is she

likely to use her loan in family enterprise, forpurchasing land or toward household con-sumption. 23 With respect to control over loan,our results suggest that women who controltheir loans are less likely to use it in family-

run enterprises, for buying land or for familymaintenance. The resistance that women putup against using loans in family enterprise orfor land purchases reflects their real lack ofco-ownership of households productive assetsand the associated fear that this may bring withrespect to loan repayment. 24 With respect tothe control variables, the results suggest thatif the family enjoys better economic status, thenthe loan is less likely to be used in family enter-prise as compared to using it for womans ownenterprise. Also, younger woman are more

likely to see their loans being diverted into fam-ily consumption. If younger women are alsolikely to be newer clients, then this suggests thatsome capacity building may be necessary beforethey are introduced to credit. In such cases, aBRAC type intervention that exclusively tar-gets destitute women in rural Bangladesh maybe much more beneficial. BRACs Income-Generating Vulnerable Group DevelopmentProgram (IGVGD) targets poor women toreceive a monthly food ration over a two-yearperiod (for details, see Ahmed et al., 2007).

The participants receive complementary train-ing in income-generating activities, awareness-raising training on social, legal, health, andnutrition issues; and basic literacy andnumeracy education through NGO partners.

Table 5. Descriptive statistics of the variables used in the loan-use modela (N = 397)

Minimum Maximum Mean Standard deviation

Credit program-related variableDURATION(est3) 0 years 10.33 years 5.08 2.70

PEERP 0% 73.77% 19.24 20.70CONTROL 0 1 0.32 0.47

Household characteristicsHHHSEX 0 1 0.58 0.23HOUSE 0 1 0.17 0.38LABORSHARE 2.12% 50.00% 12.97 7.18OLOAN 0 1 0.10 0.31CASTE 0 1 0.28 0.45VILLAGE 0 1 0.55 0.50

Womans personal characteristicsWOMAGE 15 years 73 years 31.10 9.88

WOMEDU 0 2 0.89 0.58MALECHILD 0 1 0.87 0.34

a If loan was used for more than one purpose (9.32%), the primary use was recorded.

http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-http://-/?-


16/23

In addition, the women are given access to per-sonal savings and microcredit services upongraduation from the program. The idea is tobuild their economic capacity so that whenloans are provided clients can engage in in-come-generating activities.

From the above results, we can commentconclusively on the link between control overloan and loan-use. Women with a firm controlover their loans (and those from better-offhouseholds) are more likely to invest in busi-

nesses they manage, while majority of the oth-ers see their loans used in family enterprise,for land purchases, or for consumption pur-poses. This may not only leave them in a weak-er position with respect to repayments and

jeopardize their access to credit in the future,but also have adverse implications for theircontrol over households productive assetsand hence their overall empowerment. In thenext section, we examine the repayment datato explore some of these linkages.

We test the robustness of the variables forwhich we find significant coefficients as before.The values of the coefficients with respect tothe credit program and households economic

status remain relatively stable through the step-wise deletion suggesting that our conclusionsregarding the determinants of loan-use are ro-bust.

(c) Loan-use and repayment experiences

We examine repayment data in this sectionto understand the implications that loan-usemay have for repayment. Table 7 reports thesource of repayment by loan-use. As men-

tioned earlier, repayment rates in our sampleare 100%, but as seen here this may camouflagethe various problems women encounter inrepaying loans. In particular, the table showsthat where loans were used for purposes otherthan self-managed enterprises women mainlyrelied on their own earnings from wage labor-ing to repay loans. Significantly, majority ofthe women who use loans for own enterpriseuse the earnings from their business to repayloans. Using unstructured interview techniques,we gathered testimonial evidence from the loa-nee women to investigate these experiences fur-ther. We explore each specific experience insequence.

Table 6. Multinomial logit estimation of determinants of loan-use (N = 397)

DEPENDENT VARIABLE: LOAN-USE

6-1 6-2 6-3FAMFARM LAND CONSUME

Program-related variableDURATION(est3) 0.002 (0.03) a 0.051 (0.56) 0.086 (0.91)PEERP 0.056 (6.20)*** 0.038 (2.78)*** 0.054 (3.51)***

CONTROL 1.562 (4.23)*** 1.997 (1.63) 1.516 (2.91)***

Household characteristicsHOUSE 0.761 (1.90)** 0.607 (0.63) 2.041 (0.17)LABORSHARE 0.033 (1.37) 0.001 (0.02) 0.020 (0.65)OLOAN 0.826 (1.56) 1.111 (0.10) 0.115 (0.13)CASTE 0.101 (0.24) 0.622 (1.15) 0.116 (0.20)VILLAGE 0.119 (0.35) 0.149 (0.32) 0.237 (0.46)

Womans personal characteristics

WOMAGE

0.027 (

1.58)

0.027 (

1.18)

0.049 (

1.84)*

WOMEDU(1) 0.631 (1.39) 0.059 (0.05) 0.266 (0.45)WOMEDU(2) 0.572 (0.95) 0.635 (0.13) 0.629 (0.22)MALECHILD 0.370 (0.73) 0.159 (0.13) 0.402 (0.32)CONSTANT 4.780 (4.77)*** 2.156 (1.11) 2.834 (1.66)*

Number of cases 227 40 48Number of bootstrap replications 10000Log likelihood 372.831

a Z-statistics are given between parentheses.* Significant at the 10% level.** Significant at the 5% level.*** Significant at the 1% level.



17/23

The experiences of women who have theirloans diverted into family enterprise or for landpurchases suggest that women have little influ-ence over households incomes and assets.G3W7, G3W11, V2W6, and V11W3 are wo-

men whose loans were used as working capitalin family farms. Before obtaining SHG loans,these women had worked on family farms orwithin their households, but now they findthemselves working as wage laborers mainlyto meet repayments. Moreover, in some in-stances, as in the cases of G3W7 and V11W3,they were explicitly asked to take up wagelaboring to repay loans by their husbands. Asdiscussed before, not only is waged work con-sidered socially inferior to work on own assets,but also women, compelled by the need to

make repayments, had to pledge their laborfor very low wages. 25 One of them, G3W11,expressed the desire to discontinue SHG mem-bership so that she could stop working as awage laborer. In addition, V11W3 finds thather husband, who used to discuss household fi-nances with her, is now secretive about incomefrom crop sale and remittances for fear that shemay ask him to make repayments. Our inter-views indicate that these women are resentfulabout having to withdraw their labor fromwork on own assets and work for wages in-

stead. These experiences at least partly explainwhy women who exert significant control overtheir loans would prefer not to use it in familyenterprise or for purchasing land.

Among the women whose loans were used tomeet the consumption needs of the householdG5W12, G7W2, and V4W9 had voluntarilyused their loans to avert a household crisis(G5W12husbands ill-health, G7W2 andV4W9food shortages). Both G5W12 andV4W9 exerted significant control over theirloans. All three women were involved in wagelaboring prior to joining the SHG but nowhad to divert their wages into repayments. Inaddition, as a result of peer-pressure (which in

case of G5W12 was hostile) G5W12 sold hercopper vessels and V4W9 sold her goat. Theirfamilies did not consent to these sales and bothwomen are suffering the consequences. For in-stance, G5W12 is not allowed to keep money

from sale of crop or her husbands wages, bothof which she controlled prior to the incident.She has even lost control over her own wages,which her husband now collects directly fromher landlord employer to stop her from usingwages to repay loan. This was also the experi-ence of several other women we interviewed,like V4W9, V10W2, and G3W5. Although notcommon, women were also actively punishedfor what was seen as acts of defiance. For in-stance, G7W2 and V4W9 experienced deliber-ate negligence from their families with respect

to their rice consumption during particularlylean periods. Testimonies suggest that, priorto the incidents, these women may have had agreater say over household decisions and in-comes and that this has now diminished lestthey try and divert resources away from thehousehold.

A number of women who used their loans forself-managed enterprises did perceive a positivechange in their statuses. Many of them like,G6W14, G7W7, V7W1, and members of G9spoke about how their ability to earn incomes

independent of their husbands and without re-course to wage laboring gave them a confidencein their own capability and worth and had alsochanged the general attitude of people aroundthem. They also valued the additional benefitsthat access to a group network provided intimes of family emergencies. However, evenamong them just 29.3% (N = 82) reported adefinite positive income after repayments. Thisis mainly the members of the group that in-vested jointly in a fertilizer shop. Pooling theirloans gave women the opportunity to investin a high investment and high returns businesswithout undue exposure to risk. The opportuni-ties available to women who managed business

Table 7. Source of repayment by loan-use (in percentage)a

Source of repayment Loan-use

SELFBUS FAMBUS LAND CONSUME

Self-managed business 85.4 0 0 0

Family enterprise 13.4 9.25 17.5 4.17Own wages 1.2 87.66 82.5 87.5Sale of asset 0 3.08 0 8.33Number of cases 82 227 40 48

a In case of multiple sources (8.32%), respondents were asked for the primary source.

http://-/?-http://-/?-http://-/?-


18/23

individually were highly restricted because ofsmall loan size, and severe competition amongthe women meant that very few made anyappreciable profit. 26

The loans procured by women may help their

households diversify and improve incomes, butwomens lack of authority over family assetsmeans that they are unable to divert incomefrom these sources toward repayments. In suchcases, they may lose control over the allocationof their work time and may even find their rel-ative powers in domestic relations depreciate.Where their loans are used to avert a family cri-sis, using own wages or selling assets for repay-ment can result in loss of authority overhousehold resources and in some instances evenresult in hostility toward the woman. These

experiences reveal some of the difficulties wo-men face in repaying loans which are mislead-ingly assumed away by the high repaymentrates. 27 Even when loans are used for self-man-aged businesses, women find it difficult to makeprofits mainly because of small loan size andlack of joint group investments.

Regarding the paradoxical results obtainedearlier, we can identify three broad points.First, the credit needs of poorer householdswithin our sample (for productive or consump-tion purposes) are high and families are by and

large able to divert loans procured by womeninto these activities. If the loan was not avail-able, many households would be worse off interms of income diversification while quite afew may have plunged into crises. Second, wo-mens lack of command over households pro-ductive assets means that she is unable todivert any income from such sources intorepayments and is having to rely on the limitedmeans available to herwage laboring and saleof smaller belongings. This has an adverse im-pact on both allocation of her work time and

her say over family resources. Finally, if loansgiven to women are continued to be divertedinto household needs without any change intheir asset positions, then this can over timewiden the existing resource divide betweenmen as owners and women as laborers andprove to be a disempowering experience forthe women concerned.

7. CONCLUSIONS

This paper sets out to investigate the para-doxical suggestion that emerges from studies

evaluating the impact of microcredit viz., whilelending to women benefits their households, itsbenefits for women themselves are not as cer-tain. Using detailed data sets from two villagesparticipating in the SHG program in India, it

examines the impact of the credit program oncore dimensions of household vulnerabilityand female empowerment. The same paradoxi-cal result that taunts the microcredit literaturesurfaces: we find that lending to women is likelyto strengthen the households ability to copewith vulnerability across income groups butthat the women themselves, especially the poor-est ones, are not likely to see consistentimprovements in their household status. Fur-ther, we investigate the mechanisms underlyingthis impact-paradox by examining loan-use

and repayment data and testimonies by womenborrowers. Our findings suggest that loans gi-ven to women are mainly diverted into produc-tive or consumption needs of their households.While this in general helps the householdsstrengthen their ability to cope with crises, itmay have adverse consequences for the womenconcerned. Womens lack of ownership of fam-ilys productive assets means that even whenher loans are used for productive purposes theyare unable to divert any of the incomes fromloan-sponsored activities into repayments.

Compelled to rely on their own devices, womenare forced to accept unfavorable use of theirwork time and may also find their control overfamily resources diminish. The findings of thisstudy also suggest that if womens lack of con-trol over family assets is not challenged, thenmicrocredit may fail to live up to its promisevis-a-vis empowerment.

The findings of this paper have a number ofpolicy implications. First, our results indicatethat lending to women undeniably benefits theirhouseholds diversify incomes and improves

their ability to cope with shocks. Hence, micro-credit can be a powerful vehicle for enhancingincomes and protecting households from therisk of crises. Second, our results suggest thatmicrocredit alone may not be the right inter-vention for new clients. A social security pro-gram like the IGVGD intervention offered byBRAC or an insurance backed credit schememay be a more beneficial in such cases. Suchfinancial services can also provide vital cush-ioning in time of economic shock due to naturaldisasters or ill-health. Third, our findings also

show that the benefit to women is greatestwhere loans are used for self-managed enter-prises and especially so if individual loans are



19/23

pooled into group projects. This suggests thatcredit must be accompanied by programsaimed at building the groups economic capa-bilities. It is likely that the lack of focus ongroup building activities within the SHG

scheme is one of the main reasons for its ineffec-tiveness in empowering women. Credit alone isunlikely to lead to womens emancipation interms of affecting her household position andallocation of her work time (also see EDA,2005; Hunt & Kasynathan, 2001). Our findingssupport the often made suggestion that wo-mens empowerment may be increased whencredit is offered as part of an integrated pack-age that includes other services like non-pro-

ductive loan facilities, insurance, enterprisedevelopment, and welfare-related activities(Berger, 1989; Holvoet, 2005; Johnson & Rog-aly, 1996; Mayoux, 2005). Finally and mostimportantly, our findings suggest that where

households demand for credit for productivepurposes is high, lending to women may notbenefit her personally. For this to happen, thepatriarchal hold on familys productive assetsneeds to be challenged. One of the ways inwhich this could be achieved is to make creditconditional on asset transfers in favor of thewomen concerned. Effective transfer is likelyto be achieved where assets are acquired usingwomans own loan money.

NOTES

1. For a discussion on conceptual and methodologicalissues, see de Aghion and Morduch (2005), Kabeer(2001) and Morduch (1999).

2. The Rural Finance Access Survey, 2003 reported inBasu and Srivastava (2005), however, indicates thatpoorer households in rural India still have very littleaccess to formal finance. For instance, 70% of marginal/landless farmers do not have a bank account and 87%have no access to institutional credit.

3. The involvement of its vibrant NGO sector hasgreatly boasted Indias microcredit movement. Esti-mates suggest that by 2006 there were over 1,000 NGOengaged in mobilizing savings and providing creditservices to the poor (World Bank, 2006). By 1994, theselargely donor supported institutions also began toattract financial support from NABARD and otherstate development banks. Although the NGOs arecrucial to the microcredit sector, their outreach andvolume of loan is still relatively small. Among the mostprominent NGO-led microcredit institutions are those

managed by BASIX, CARE, MYRADA, SEWA, andSHARE. For a discussion, see World Bank (2006).

4. For a critic of the NABARD studies on methodo-logical grounds, see World Bank (2004).

5. A total of 302 households were surveyed but six defacto female and four de facto male headed householdsand one income outlier were excluded from the analysis.The interviews were carried out by a group of twointerviewers, one male and one female. The authorparticipated in over 1/3rd of these and also carried out

all the focus group interviews. For details on method-ology and survey protocol, see chapter 2 in Horrell,Johnson, and Mosley (2008).

6. A systematic analysis of male and female responsesdid not indicate a gender bias in responses, except in caseof variables pertaining to the head of the household(defined later in the paper).

7. The correctness of the official poverty figures isintensely debated (see Deaton & Dreze, 2002). Income isnet of costs but not of loan repayments.

8. Unlike Hashemi et al. (1996), we have focused on

neutral household decisions and excluded all those thatan SHG member is more likely to take when comparedto others (like the decision to buy a goat). Not doing socould lead to a bias in favor of the credit program.

9. The variable DURATION was preferred to thevariable SHG-MEMBER (coded as one if the householdhad an SHG member, r = 0.942, p = 0.000) since it notonly differentiated between members and non-membersbut also between early and late joiners.

10. For the independent categorical variables used inthis study, the contrast type is specified as Indicatorand the smallest category is identified as the referencecategory. The Indicator contrast type creates a set ofdummy variables that indicates the presence or absenceof category membership. Values for the reference cate-gory are set to zero such that no parameter estimates arecomputed for this category and those for the otherremaining categories represent deviations from the effectof being a member of the reference category.

11. Selection bias occurs if the credit program partic-ipants differ from the non-participants in unobservablecharacteristics. If these characteristics are related to thevulnerability and empowerment measures studied here,then the coefficient of DURATION will reflect these



20/23

effects and will be biased. More specifically it can causethe statistical effects of participation to be exaggerated(Morduch, 1999; Pitt, Khandker, McKernan, & Latif,1999). Pitt and Khandker (1998) and Pitt et al. (2006) arestudies that have, among others, used the instrumental

variable technique to correct for the endogeneity ofcredit program participation.

12. Note, however, that clusters are not very far fromeach other and spill-over effects are likely. The datawe currently have do not allow us to explore this issueany further.

13. We thank one of the referees for suggesting theestimation procedure discussed here, in particular theuse of the tobit model to estimate DURATION in thefirst-stage. The same referee also suggested the use of

ivprobit. We report on this later.

14. While the bootstrap procedure is well established,there is still no fixed rule concerning the number ofreplications one should use in computing bootstrapstandard errors. Since theory desires an infinite numberof replications, the decision often rests on practicalconsiderations (see Gould & Pitblado, 2005). In thisstudy, we carry out 10,000 replications each time on thebasis that estimates obtained remain robust to furtherreplications.

15. Significantly, this led us to drop the variableVILLAGE. Estimating the models separately for thetwo survey villages suggests that while the study resultsare generally valid at the village level, they are somewhatstronger for Vepur.

16. If DURATION is used as is the results aresomewhat mixed. Notably, its coefficient is statisticallysignificant in (4-1) and has the expected sign. Thissuggests that endogeneity may be a problem and justifiesthe use of the instruments.

17. In principle, a Wald test could also be used, but thelikelihood-ratio test is found to be more reliable forsmall sample sizes (Agresti, 2007; Menard & Menard,2001).

18. According to Mahajan and Ramola (1996), theaverage annual credit use by rural households in India isaround 14,549 Rs. Of this, 65% is for productivepurposes and 35% for consumption purposes.

19. Members of just two SHGs invested in jointprojects: a successful fertilizer shop and a rental com-pany that catered to special occasions like weddings andfunerals. Projects managed individually were usuallypetty businesses like livestock (65%), mobile shops(23%), tea stalls, grocery, and tailoring shops (13%).

20. We acknowledge that the way we have set up thismodel is potentially problematic, since (some of) theright-hand side variables may be endogenous and mayresult in biased estimates. In particular, as suggested byone of the referees, the variables CONTROL could be

endogenous to the outcomes being studied. To solve thispotential endogeneity problem, we should have usedvalid instruments. For example, we should have usedinformation on past behavior of members to endogenizeCONTROL. Unfortunately, our current data set doesnot have suitable instruments. We therefore suggest thatthe results be treated with caution.

21. Members of some SHGs display a greater tendencyto use loans for own enterprises than others. Thissuggests that within group dynamics might be animportant consideration. PEERP is designed to capture

one aspect of these dynamics. The limitations of ourdata set do not permit the inclusion of other relevantvariables like social ties and leadership.

22. This compares with Goetz and Guptas (1996)findings for BRAC, Bangladesh. If we classify all womenscoring two or more points as in control of their loans,then our data suggest that 67.51% of them control theirloans to some extent. This compares with findings fromstudies for Bangladeshs Grameen Bank (Goetz &Gupta, 1996; Hashemi et al., 1996; Rahman, 1986).

23. Hermes, Lensink, and Mehrteab (2005) suggestedthat the group leaders actions (and not of other groupmembers) matters for the performance of microcreditgroups. To test whether the impact of the group leadersloan-use is significantly different from that of the othergroup members, we separately analyze for a groupleader effect. Our results suggest that the group leadersloan-use matters but so does that of the other groupmembers.

24. There were instances where woman displayedremarkable courage to retain the ability to use theirloans on self-managed enterprise. Like in the case ofV1W8 who physically fought her husbands attempt totake her loan money to invest in his barbers shop. Insome instances, however, women gave up their loanswillingly. G2W8 gave her loan to her husband andfather-in-law because she considered herself well lookedafter by them and was not confident about using itherself.

25. Low rainfall and lack of off-farm work contributeto low female wages in the survey area. On an average,these were between 36.8% and 39.8% of the statutoryminimum wage (Garikipati, 2008).



21/23

26. During the survey, we faced the daily dilemma ofselecting from the half a dozen tea stalls that the SHGwomen operated. Mosley and Rock (2004) reportedsimilar evidence from Zimbabwe (CARE) and SouthAfrica (SEF), where women traders are forced to seek to

regulate the market, for instance, by agreeing to trade onalternate days.

27. This compares with findings from Mayoux (2005),who suggested that high repayment levels do notindicate womens control over the loans and may infact be a sign of social pressure to access resources forothers in the household. In further support of this thesis,

we find that very few women who reported negativeexperiences actually wanted to leave their SHGs.

REFERENCES

Agresti, A. (2007). An introduction to categorical dataanalysis. London: Wiley-Interscience.

Ahmed, A. U., Quisumbing, A. R., & Hoddinott J. F.(in collaboration with M. Nasreen, E. Bryan, and theData Analysis and Technical Assistance Limited).(2007). Relative efficacy of food and cash transfers inimproving food security and livelihood of the ultra-

poor in Bangladesh. Revised Draft Report submittedto the World Food Program, Bangladesh, September1, 2007. Washington DC: International Food PolicyResearch Institute.

Bansal, H. (2003). SHG-bank linkage program in India:An overview. Journal of Microfinance, 5(1), 2149.

Basu, P., & Srivastava, P. (2005). Scaling-up microfi-nance for Indias rural poor. World Bank PolicyResearch Working Paper 3646. Washington, DC:World Bank.

Berger, M. (1989). Giving credit to women: Thestrengths and limitations of credit as a tool foralleviating poverty. World Development, 17(7),10171032.

Chavan, P., & Ramakumar, R. (2002). Micro-credit andrural poverty: An analysis of empirical evidence.Economic and Political Weekly, 37(10), 955965.

Chowdhry, P. (1994). The veiled woman: Shifting genderequations in rural Haryana, 18801990. Delhi:Oxford University Press.

da Corta, L., & Venkateshwarlu, D. (1999). Unfreerelations and

Impact of Lending to Women on Household Vulnerability

Documents

Transcript of Impact of Lending to Women on Household Vulnerability