Microfinance in Northeast Thailand

World Development Vol. 34, No. 9, pp. 1612–1638, 2006� 2006 Elsevier Ltd. All rights reserved

0305-750X/$ - see front matter

www.elsevier.com/locate/worlddevdoi:10.1016/j.worlddev.2006.01.006

Microfinance in Northeast Thailand:

Who Benefits and How Much?

BRETT E. COLEMAN *

Asian Development Bank, Manila, Philippines

Summary. — This paper evaluates the outreach and impact of two microfinance programs in Thai-land, controlling for endogenous self-selection and program placement. Results indicate that thewealthier villagers are significantly more likely to participate than the poor. Moreover, the wealth-iest often become program committee members and borrow substantially more than rank-and-filemembers. However, local information on creditworthiness is also used to select members. The pro-grams positively affect household welfare for committee members, but impact is insignificant forrank-and-file members. Policy recommendations include vigilance in targeting the poor, publiclydisseminating the program rules and purpose, and introducing and enforcing eligibility criteria.
� 2006 Elsevier Ltd. All rights reserved.
Key words — microfinance, microcredit, impact, poverty targeting, Asia, Thailand

* I would like to thank George Akerlof, Pranab Bard-

han, David Dole, Paul Gertler, Alain de Janvry, Elisa-

beth Sadoulet, Ken Train, Ploenpit Satsanguan, seminar

participants at the University of California at Berkeley,

and two anonymous reviewers for their helpful com-

ments; the staff of CRS/Thailand, especially Yupaporn

Boontid and Ruth Ellison, for their advice and support

throughout the surveys; the staff of RFA/Surin and

FIAM/Roi-Et for the able enumeration services of their

field staff and for other data on their village banks; the

staff of BAAC in Surin and Roi-Et for data on their

members; and, of course, the villagers surveyed who

generously gave of their time and their lives to me.

Special thanks are due to Thanorm Charoensiri for her

outstanding and dedicated research assistance. I grate-

fully acknowledge financial support from the Social

Science Research Council and the Fulbright Scholarship

Program. Final revision accepted: January 27, 2006.

1. INTRODUCTION

Historically, efforts to deliver formal creditand financial services to the rural poor in devel-oping countries have failed. Commercial banksgenerally do not serve the needs of the ruralpoor because of the perceived high risk andthe high transactions costs associated withsmall loans and savings deposits. To fill thevoid, many governments have tried to deliverformal credit to rural areas by setting up specialagricultural banks or directing commercialbanks to loan to rural borrowers. However,these programs have almost all failed becauseof the political difficulty for governments to en-force loan repayment, and because the rela-tively wealthy and powerful, rather than thepoor, received most of the loans (Adams, Gra-ham, & von Pischke, 1984; Adams & Vogel,1986; World Development Report, 1989).

The recent proliferation of innovative micro-finance programs, often based on group-lend-ing methods, has been inspired largely by thebelief that such programs reach the poor andhave a positive impact on various measures oftheir welfare, including economic measures(e.g., wealth and income), social measures(e.g., educational attainment and health status),and less tangible measures such as ‘‘empower-ment.’’ The popular press has waved the banner

161

of microfinance as perhaps the most importantrecent tool to reduce poverty, 1 and the 1997Microcredit Summit called for the mobilizationof $20 billion over a 10-year period to supportmicrofinance (Microcredit Summit Report,1997). The United Nations proclaimed 2005as the ‘‘Year of Microcredit.’’ Much of thisfaith in microfinance is based on the highlyselective anecdotal evidence of individuals

2

MICROFINANCE IN NORTHEAST THAILAND 1613

who are reported to have pulled themselves andtheir families out of poverty with the benefit ofmicrocredit. On the other hand, prominent dis-senters to the popular view (Adams & von Pis-chke, 1992) have written that ‘‘debt is not aneffective tool for helping most poor people en-hance their economic condition—be they oper-ators of small farms or micro entrepreneurs, orpoor women.’’ They argue that access to creditis not a significant problem faced by small agri-cultural households and that factor and prod-uct prices, land tenure, technology, and riskare the factors limiting small farmer develop-ment. Yet, despite the proliferation of theseprograms and the outpouring of support bydonors, there has been little sound empiricalresearch that tests the hypotheses that theyare reaching and benefiting the poor. 2

To justify such a significant investment to re-duce poverty, compared to alternative invest-ments in other poverty alleviation programs,the proposition that microfinance reaches thepoor and positively affects their welfare shouldbe proven and not just assumed. This paper at-tempts to contribute to overcoming this short-coming in the literature by examining theresults of a survey of two Northeast Thailand‘‘village bank’’ programs that target the poor.The survey was designed and conducted in1995–96 with the express purpose of measuringoutreach and impact on the poor, while con-trolling for the endogeneity biases that haveplagued other studies.

The NGO programs studied in this paper tar-geted ‘‘the poorest of the poor’’ according toproject documents and donor policy. The abilityof any program to achieve this goal depends onthe institutional context in which it is imple-mented, and the main premise on which micro-finance programs are based is that the poor arecredit constrained and have limited access to for-mal sector credit. In Thailand, however, theBank for Agriculture and Agricultural Coopera-tives (Bank for Agriculture & AgriculturalCooperatives, 1997) claims to serve over 80%of rural households. Hence, it is possible thatthe rural poor in Thailand are not credit con-strained. However, the BAAC’s outreach in theNortheast, the country’s poorest region, is smal-ler than the rest of the country. In the fourteenvillages surveyed for this study, 63% of villagehouseholds were BAAC members. Moreover,as is often the case in government-led credit pro-grams, the BAAC’s clientele is largely male; only29.5% of BAAC members surveyed by theauthor were women. Hence, only 18.6% of sur-

veyed households included women who had ac-cess to BAAC loans. On the other hand, 25.8%of surveyed households included women whowere in debt to moneylenders. At the time ofthe surveys, BAAC’s annual interest rate variedfrom 3% to 12%, whereas moneylenders chargedbetween 60% and 120% per year, and the NGOprograms evaluated in this paper charged 24%per year. Hence, there is evidence that womenin Northeast Thailand may be credit constrainedand may benefit from access to lower-cost insti-tutional credit.

Two main problems plague attempts toevaluate the impact of microfinance programs. 3

The first is self-selection of participants. Toillustrate this source of bias, consider a sampleof households drawn only from villages witha village bank: some households will haveselected to be village bank members, whileothers will have selected not to be members.It is likely that there are significant differencesbetween self-selected village bank membersand nonmembers. To the extent that such dif-ferences can be observed and measured (e.g.,age, education, and wealth endowment), theycan be statistically controlled for when estimat-ing village bank impact. However, to the extentthat such differences cannot be observed (e.g.,entrepreneurship, risk preferences, trustworthi-ness, attitudes regarding the role of women inthe household, and attitudes toward belongingto a program targeting the poor), direct com-parison of village bank members and nonmem-bers will yield biased estimates of village bankimpact. This bias arises because the same unob-servable characteristics that lead some womento become village bank members will also affectimpact measures such as income, accumulationof assets, and spending on education and healthcare. For example, women who are more entre-preneurial (a characteristic that is virtuallyimpossible to measure) would be expected tohave a tendency to self-select into the program,but such women would also be expected tohave higher welfare measures such as incomeand expenditures even without the program.Uncontrolled comparisons between membersand nonmembers, therefore, might incorrectlyattribute such higher incomes to the villagebank program. Alternatively, the relativelypoor might self-select into the program if beingpoor is a publicly known selection criterion,and the relatively wealthy might not join toavoid any stigma related to being poor. If theprogram has positive impact on participantsbut this impact is not strong enough, then an

1614 WORLD DEVELOPMENT

uncontrolled comparison between poor mem-bers and rich nonmembers might lead to theerroneous conclusion that the program wasimpoverishing participants.

The second problem affecting attempts tomeasure impact is endogenous program place-ment and is similar to the self-selection prob-lem. To understand this problem, it is usefulto consider a commonly used sample, which in-cludes households from villages with a villagebank and households from villages without avillage bank (e.g., MkNelly & Watetip, 1993;Wydick, 1999a, 1999b). Prior to program place-ment, some villages may be perceived as moreentrepreneurial or better organized and gov-erned, or may have more dynamic leaders, withsuch leadership spilling over to affect others’behavior in the village. These unmeasurablecharacteristics may cause villagers in these vil-lages to have higher incomes, spend more oneducation and health, and generally have high-er measures of welfare than households in othervillages, even without the program. If the spon-soring NGO uses such unmeasurable character-istics to select a village for program placement,then a comparison of households in the pro-gram village and households in a nonprogramvillage may overestimate impact. Similarly, ifthe sponsoring NGO deliberately selects poorervillages because of its mission to reduce pov-erty, a comparison of households in the pro-gram village and households in a nonprogramvillage may underestimate impact. Coleman(1999), using the same data set examined in thispaper, demonstrates the extent to which esti-mates uncorrected for member self-selectionand endogenous program placement signifi-cantly overestimate average program impact.

Assessing the targeting of a village bank, forexample, determining if, and to what extent, theprogram is reaching poor households as in-tended, also encounters difficulties. 4 Foremostamong these is that, typically, empirical studiesfocus on program impact, and such studies nec-essarily require that households be surveyedafter the program has operated for some time.Because measures of poverty, such as wealthand income, have been influenced by the pro-gram, it is typically impossible to determine ifa participating household was poor when it firstjoined the program. An ex post survey findingthat a participating household is relativelywealthy could indicate either unsuccessful tar-geting (resulting in the poor being excludedand the rich co-opting program benefits) or suc-cessful program impact.

This paper extends and refines the methodol-ogy used in Coleman (1999). First, it exploitsthe unique characteristics of the survey sampleto evaluate targeting, to determine if, prior tojoining the program, participating householdsare relatively poor or not. Second, it extendsthe impact estimates of the earlier study to mea-sure differential impact on different classesof participants, specifically on the relativelywealthy village bank managers and rank-and-file members who tend to be poorer. Resultsindicate that self-selected program participantsare significantly wealthier than nonparticipantseven prior to program intervention, and thewealthiest villagers are almost twice as likelyto participate in the program as the poorer vil-lagers. Moreover, some of the wealthiest villag-ers obtain a disproportionate share of programloan volume by virtue of holding influentialpositions as village bank committee members.Specifically, they do this by using multiplenames to borrow more than the program’s ceil-ing per client. Positive impact is seen largely inthis wealthier group. Impact on rank-and-filemembers is significantly smaller than impacton the wealthy, and is largely insignificant.Hence, similar to previous attempts to deliverlow-cost credit to the poor, most of the benefitin the village banks studied is going to thewealthiest villagers.

The remainder of the paper is organized asfollows. Sections 2–4 describe the design ofthe programs studied, the survey design anddata, and the survey area. Section 5 presents re-sults relating to participation, including mem-ber selection and borrowing, while Section 6presents results on program impact. Section 7concludes and discusses policy implications.

2. THE NGO PROGRAMS STUDIED

The two microfinance programs studied arerun by Thai NGOs: the Rural Friends Associa-tion (RFA), located in the northeast provinceof Surin, and the Foundation for IntegratedAgricultural Management (FIAM), located inthe adjacent province of Roi-Et. RFA andFIAM have been promoting microfinance since1988 and have received financial and technicalassistance from the American NGO CatholicRelief Services (CRS). Both Thai NGOs followthe ‘‘village bank’’ group-lending methodologypioneered by the Foundation for InternationalCommunity Assistance (FINCA), 5 in whichborrowers form their own peer groups of 20–


60 members. Smaller solidarity groups are gen-erally not used, although some of the sampledvillage banks independently formed suchgroups. Both NGOs lend only to women. TheNGO grants a loan simultaneously to eachmember, but group members co-guarantee eachother’s loans. If the group does not meet itscollective responsibility to repay all of itsmembers’ loans, then all group members aredenied future credit.

The first loan is the same amount for allmembers (1,500 baht). 6 For each subsequentloan cycle, the member is entitled to borrowan amount equal to her previous loan plusher accumulated savings in the village bank,up to a fixed maximum (7,500 baht). Moreover,the group also gives loans to its members (andsometimes to nonmembers) from its members’savings. Loans from the NGO lender are‘‘external account’’ loans, and loans from mem-bers’ savings are ‘‘internal account’’ loans.External account loans must be repaid in sixmonths. 7 External account loans are madefor five years, after which time the funds usedfor external account lending are used to financenew village banks, and the internal accountbuilt up is supposed to continue to finance thevillage bank members’ needs. 8 The interest ratecharged on external account loans is 2% permonth, with both principal and interest due atterm. 9 The interest rate on internal accountloans is determined by each village bank, butis generally also set at 2% per month, althoughsome village banks set it at 3% per month. Atthe time of the surveys, RFA had a 97% on-time repayment rate on its external accountloans, and FIAM had a 100% on-time repay-ment rate. Savings are not mandatory, but areencouraged in two ways. First, external accountborrowing rights grow with increasing savings;and second, the profits from internal accountlending are allocated among members accord-ing to the amount of their savings.

Group formation and membership was gen-erally determined as follows. The initial contactbetween NGO and village could be initiatedeither by the NGO or by the village, but almostalways involved contacts between the NGOand ‘‘leading members’’ of the village. As iscustomary in Thailand, the NGO fieldworker’sfirst visit to the village would include a visit tothe village chief to introduce the purpose of thevisit. Frequently, the task of assembling andorganizing the village women into a villagebank group would be delegated to the villageleadership, either in the form of the village chief

or of a female leader designated by him. Therole of the fieldworker was, therefore, fre-quently reduced to explaining the programto the group thus formed, and to administer-ing the delivery of loans and collection ofrepayments. In three of the villages surveyed,however, the other extreme was observed—namely, that the fieldworkers, who were resi-dent in the villages, handpicked the women tojoin the program.

Each village bank was managed by a com-mittee comprising a president, vice president,treasurer, and other officers. These committeemembers make most of the day-to-day deci-sions, including (importantly) decisions onmembership and borrowing eligibility and theallocation of loans. In principle, the committeemembership was elected annually, but in prac-tice the leading women designated to organizethe village bank became committee membersand remained in their positions from year toyear. 10

3. SURVEY DESIGN AND DATA

I conducted a unique survey of 444 house-holds in 14 villages in Northeast Thailand in1995–96. Eight of the villages were supportedby RFA, and the other six were supported byFIAM. Of the 14 villages surveyed, six hadnever benefited from village bank support,and did not receive any village bank loansduring the survey period. These ‘‘control’’villages were identified as follows. Based ontheir expansion plans, RFA pre-identified fourvillages and FIAM pre-identified two villagesthat they would begin supporting in 1996. InFebruary and March 1995, RFA and FIAMfield staff organized the villagers into the newvillage banks, allowing them to self-select,according to standard procedures, the only dif-ference being that the villagers were told thatloans would not begin for approximately oneyear.

A random sample of eight ‘‘treatment’’ vil-lages (four from RFA and four from FIAM)was also chosen from a list of their villagebanks (32 for RFA and 26 for FIAM). I in-cluded one village that was just due to beginreceiving loans because it could also serve asa control village for certain purposes 11 and be-cause I wanted to observe the initial develop-ment of a new village bank.

A stratified random sample of householdswas then selected in all 14 villages, with the


stratification done on ‘‘original participant’’status to ensure a sufficient number of programparticipants in the sample and to ensure no biasfrom program attrition. Specifically, the partic-ipants were chosen from the list of householdswho joined the program at its outset. It was afortunate characteristic of the programs (froma statistical point of view) that no new membershad been allowed to join after the initial mem-bers joined. This made the participants in treat-ment villages directly comparable with those incontrol villages. It is noteworthy that these ori-ginal participants in treatment villages includedseveral households who had subsequentlydropped out of the program for various rea-sons. Hence, to the extent that ‘‘drop-outs’’had been positively or negatively affected bythe program, this is fully accounted for inthis study. 12 A total of 505 households wereselected. Of these, 455 were located and agreedto be interviewed during the first survey,and 445 (including 294 program participantsand 151 nonparticipants) finished the surveys.

Each village was surveyed four times over thecourse of a year. The first survey was conductedin April 1995 and collected data on householddemographics, assets, and debts. The second,third, and fourth surveys were conducted inAugust 1995, October 1995, and February–March 1996, respectively. They collected dataon income, expenditures, and credit transac-tions during the dry season (February toMay), the rainy season (June to September),and the harvest season (October to January),respectively. In addition to the household sur-veys, village surveys were also conducted to col-lect data on village infrastructure, prices, andother characteristics. The household surveyswere administered by the staffs of RFA in Surinand FIAM in Roi-Et, under my supervision. 13

My research assistant and I conducted thevillage surveys as well as in-depth informalinterviews with numerous villagers.

4. SURVEY AREA

The provinces of Surin and Roi-Et are adja-cent to each other and are located in NortheastThailand, the country’s poorest region. Mostvillage households engage in small-scale agricul-ture: 90.4% of the adult men and 91.3% of theadult women in the households surveyed listedfarming as their primary or secondary occupa-tion. In Surin, rain-fed rice is the primary cropgrown, with planting in June and harvesting

from November to January. During the off-sea-son, a few households engage in small-scale irri-gated gardening, but most either engage innonagricultural income-generating activities orremain idle. The more common activities in-clude pig raising, itinerant wage labor (espe-cially construction) in the provincial capital orBangkok, and small business activities such aspetty trading, driving a motorcycle taxi, spin-ning and weaving silk and cotton, and operatingsmall food stands. Some of the wealthier house-holds buy and sell cattle and water buffalo.Agricultural and nonagricultural activities inRoi-Et are similar to those in Surin, with twodifferences. First, another important crop culti-vated during the main growing season is stickyrice. Second, because of different soil quality, to-bacco is commonly grown as a cash crop duringthe off season from November to April, leadingto less migrant labor compared to Surin.

Most people in Surin, which borders Cam-bodia, are culturally Khmer and primarilyKhmer-speaking: 98% of the households sur-veyed in Surin are ethnically Khmer. The peo-ple of Roi-Et, like those in most of NortheastThailand, are ethnically Lao and the languagespoken is a dialect of Lao: 98% of the house-holds surveyed in Roi-Et are ethnically Lao.In both provinces, however, Thai is understoodand spoken by all but the oldest villagers. 14

5. PARTICIPATION

(a) Selection of members

The raison d’etre of most microfinance pro-grams is to correct the market failure to delivercredit to the rural poor. Most microfinanceprograms state that their primary goal is toalleviate rural poverty by delivering credit andother financial services to poor households,especially to the women in those households.This is certainly the case for the programs stud-ied in this paper. For example, Catholic ReliefServices publishes ‘‘Eight Principles of VillageBanking,’’ the first of which is to loan to ‘‘thepoorest of the poor.’’ FINCA (1990), whose vil-lage bank model was followed in the programsstudied, lists several ‘‘criteria for membershipselection,’’ one of which is that membershipshould be ‘‘open to poorest-of-the-poor, butnot exclusively for poor.’’ However, FINCAgoes on to say, ‘‘Ultimately, membership isself-selecting and not the responsibility of spon-soring agency.’’


The poor are generally targeted by grouplending programs in one of two ways. In asmall number of programs, such as the Gram-een Bank in Bangladesh, households owningassets valued at more than a certain amountare in principle excluded from participating inthe program. 15 The majority of group lendingprograms, however, do not set strict eligibilitycriteria. Instead, they seek to target the poorand screen out the wealthy indirectly by settingsmall loan sizes (e.g., $50–$300) that wouldpresumably be of little interest to the wealthy,by requiring frequent group meetings (usuallyweekly or monthly) that would impose a coston the wealthy in excess of any benefit theymight derive from the small loans, and by thestigma of participating in a ‘‘poverty lending’’program. This second, indirect, screening methodis part of the ‘‘village bank’’ model of grouplending pioneered by FINCA and adopted bymost international NGOs such as CatholicRelief Services, CARE, Save the Children,and Freedom From Hunger. It is thereforeimportant to determine whether or not theself-selection process allows NGOs to reachtheir target group of ‘‘the poorest of the poor.’’

Most existing studies on program impacthave ignored the issue of program targeting.For example, the 32 studies reviewed in Sebstadand Chen (1996) (with the exception ofMkNelly & Watetip (1993) discussed in greaterdetail below), the 11 studies reviewed in Chen(1992), and the studies by Wydick (1999a,1999b), Pitt and Khandker (1998), and Khand-ker (2005) ignore the question of targeting.

Exceptions to this neglect include MkNellyand Watetip’s (1993) study of village banks inNortheast Thailand (not far from where thecurrent study was conducted); Perry’s (1995)anthropological study of village banks in Sene-gal; Park and Ren’s (2001) study of micro-finance in China; and Amin, Rai, and Topa’s(2003) Bangladesh study. MkNelly and Wate-tip find that village bank members closelymatch a general cross-section of the village.They use an innovative survey of village chiefsregarding the wealth level of member and non-member households in the village. However,they conduct their study three years after theestablishment of the village banks. Hence, dif-ferences in wealth would have been influencedby the village bank itself. Moreover, politicallysavvy village chiefs might have deduced thatthe ‘‘correct’’ response is that village bankmembers are poorer than nonmembers. Theshortcomings of this method are underscored

by Perry’s (1995) study in Senegal. She foundthat, in response to open-ended questions, fo-cus groups of village bank members assertedthat the wealthy do not dominate village bankmembership. However, her more objectivemeasurements of wealth demonstrated thatthe wealthy in fact did have more access tomembership and to borrowing. The studies byPark and Ren (2001) and Amin et al. (2003)also evaluate targeting by comparing partici-pants and nonparticipants after program effectshave been realized, potentially biasing the re-sults.

The main problem plaguing attempts to eval-uate program targeting is that most empiricalstudies have focused on impact; therefore, datahave been collected after the programs hadoperated for some time. Differences betweenparticipants and nonparticipants are thereforecorrupted by the effects of the program.

My Thai data set, however, provides an idealsetting to study issues of self-selection and tar-geting because the sample includes seven vil-lages where villagers self-selected but had notreceived village bank loans at the time of thefirst survey. Differences in wealth of membersand nonmembers can be directly compared inthese villages. In addition, data for all house-holds surveyed were collected on land ownedfive years before the surveys, that is, landowned before any of the 14 villages had avillage bank. 16 Because land value makes up73.7% of the value of household assets in thesample, the value of land owned five years be-fore the surveys is an excellent proxy for priorhousehold wealth. 17

It should be kept in mind in what follows thata villager can become a member only if she self-selects and is selected by her peers. Hence, torefer to the selection process as ‘‘self-selection’’can be misleading; selection by peers may bethe more important process. Table 1 presentsweighted 18 t-tests on total wealth, land value,and the value of nonland assets (total and bygender of the owner) for households in theseven villages that had not received village banksupport at the time of the first survey. 19 Table 2presents weighted t-tests on land value (totaland by gender) for all 14 villages five yearsbefore the surveys.

Tables 1 and 2 clearly indicate that villagebank members, prior to any village bank sup-port, tend to be wealthier than nonmembers.Furthermore, the wealth difference comes pri-marily through the value of land owned bywomen in the household. 20

Table 1. Weighted t-tests on wealth of members and nonmembers in control villages

Variable (in baht) Village bankmembers (n = 140)

Nonmembers(n = 70)

p-Value ofdifference

Household wealth (assets less debt) 574,738 434,154 .157Female-owned wealth (assets less debt) 303,482 191,327 .055Male-owned wealth (assets less debt) 264,810 237,635 .740Value of household land 442,814 271,370 .028Value of female-owned land 254,089 128,560 .019Value of male-owned land 187,413 142,810 .455Value of household nonland assets 172,114 201,354 .476Value of female-owned nonland assets 61,604 76,252 .376Value of male-owned nonland assets 104,548 119,910 .670

Table 2. Weighted t-tests on land value of members and nonmembers in all villages five years before surveys

Variable (in baht) Village bankmembers (n = 294)

Nonmembers(n = 151)


Value of household land 480,745 249,604 .001Value of female-owned land 301,816 109,774 .001Value of male-owned land 176,624 137,851 .382


An alternative, and perhaps better, way toexamine targeting is to examine the probabilityof selection by different wealth groups. Examin-ing the selection process in this way with thecurrent data set is complicated by the stratifica-tion process used. Because stratification wasbased on participation status, the probabilityof selecting into the program has to be calcu-lated using the probability that a member (ornonmember) belongs to a certain wealth cate-gory, adjusted by the probability of being sam-pled. The results are presented in Table 3 whichis given below.

As this table shows, the probability of select-ing into the program is much higher for the twowealthiest groups of villagers; and the probabil-ity of the wealthiest group selecting into theprogram is nearly twice that of the four poorergroups.

Table 3. Probability of members of different

Wealth group(value of household landowned five years earlier, baht)

Total numberof households

in sample villages

T

0 (landless) 1430–99,999 257100,000–199,999 264200,000–399,999 351400,000–999,999 282Greater than or equal to 1,000,000 72

It is still possible, however, that wealth per se(and female land wealth in particular) is not asignificant determinant of member selection.For example, it is conceivable that the primarycriteria for membership are personal responsi-bility, trustworthiness, entrepreneurship, andan assortment of other unobservable character-istics that lead to the selection bias discussedearlier, and which are correlated with wealth.Therefore, as part of my field research, in everyvillage I interviewed two knowledgeable infor-mants (the village chief and village bank presi-dent) about the creditworthiness of eachhousehold surveyed. Specifically, I asked, ‘‘Ifthe adult women in this household borrowedmoney from any source (private bank, BAAC,friend, relative, moneylender, or any othersource), how sure would you be that theywould repay the amount due on time—very

wealth groups gaining access to program

otal participatinghouseholds

Totalnonparticipating

households

Percent ofwealth groupparticipating

in program (%)

65 78 45108 149 42111 153 42137 214 39153 129 5458 14 81


sure = 3, fairly sure = 2, or not verysure = 1?’’ By this question, I obtained a mea-sure of each household’s reputation for credit-worthiness, which would presumably be amajor criterion in determining which womenin the village would be allowed to join thevillage bank. It is intended to be a proxy formany of the unobservable characteristics dis-cussed above. A creditworthiness score wasthen calculated as the average of the chief’sand village bank president’s response for eachhousehold. 21

The results of weighted 22 logit regressions onmember selection using this and other house-hold characteristics are presented in Table 7in Appendix A. Estimates in the left-hand por-tion of the table use data only from the controlvillages and include the value of land and non-land assets owned when the first survey began.Similar estimates, using the entire sample butslightly different regressors (namely, landowned five years before our survey, that is,the initial wealth measure exogenous for allhouseholds) are presented in the right-handportion of the table. 23 Both regressions yieldsimilar results, showing that creditworthinessis a significant determinant of member selection(in control villages: coef = .709 and p = .046;in all villages: coef = .384 and p = .097), so lo-cal information apparently is being used in theselection process. At the same time, however,even controlling for this use of local informa-tion, the value of land owned by women is stillhighly significant (in control villages: coef =2.73 · 10�6 and p = .025; in all villages: coef =2.44 · 10�6 and p = .001).

This evidence that the village banks are fail-ing to reach the poor is also supported by infor-mal interviews with villagers. For example,when I asked villagers what kind of personjoined the village bank, many nonmembers intwo treatment villages identified the villagebank as a ‘‘group for the rich’’ which theywould ‘‘not be qualified to join’’ although theywould be interested in joining such a group ‘‘forpoor people.’’ In five other villages (two controland three treatment), nonmembers made simi-lar remarks that suggested that the programwas open primarily to the richer householdsin the village. None of the villagers interviewedidentified the village bank as a programthat targeted the poor. 24

As mentioned in Section 2 above, in most vil-lages the NGO field worker’s first contact waswith the village chief, and after this contact,the NGO field worker usually was uninvolved

in member selection, leaving it up to the villagechief to organize the villagers. The village chiefwould normally then contact ‘‘leading women’’in the village to assist in the organization of thevillage bank. The village chiefs and leading wo-men were generally among the wealthiest andmost influential residents of the village (andthe women were often relatives of the villagechief), and the other women they asked to joinwere also richer than average.

Frequently, the village chief’s wife was thevillage bank president or held another influ-ential committee position, and other wealthyleading women in the village also usually be-came committee members. Their influence con-tinued beyond the selection of members to thedetermination of the amounts of money thatthey and other members borrowed. We nowturn to this process.

(b) Borrowing by members

One of the most fascinating phenomena dis-covered during the course of the surveys is theextent to which village bank members use othernames, in addition to their own, to borrowfrom the village bank. For example, sometimeswhen a member does not want to borrow, shewill let another member use her name to bor-row from the village bank. Of more impor-tance, however, is the use of multiple namesto borrow. For instance, some members alsoenroll as ‘‘members’’ a daughter (who may ormay not live at home) or a sister or other rela-tive who lives in Bangkok. In other instances, amember will take over the account of a memberwho leaves the village bank. Virtually all of themembers who used multiple names to borrowfrom the village bank on a more-or-less perma-nent basis were also influential committee mem-bers (president, vice president, or treasurer ofthe village bank), who set village bank policy(especially regarding management and lendingof the internal account funds) and manage itsdaily operations. Committee members in sixof the eight treatment villages were engaged inthis practice. The use of multiple names has alarge impact on the amount that a membercan borrow, effectively multiplying the maxi-mum loan size by the number of names used.In the most extreme case, one village bank pres-ident (who was also the village chief’s businesspartner and a moneylender for several villagesin the area) used nine names to borrow. Alltold, households reported that 35% of loanvolume in the eight treatment villages was


borrowed by someone other than the person re-corded in the village bank records. Even thisestimate is probably low because some villagersreported that they had been instructed not totell me about the use of multiple names.

In principle, the most important determinantof a member’s cumulative borrowing is thelength of time she has been a member. Hence,borrowing by members is estimated in Tables8–10 in Appendix A, using as regressors varioushousehold characteristics, the number ofmonths of membership as a rank-and-file mem-ber, and the number of months as a committeemember. Table 8 presents results for externalaccount borrowing; Table 9 presents resultsfor internal account borrowing; and Table 10presents results for total borrowing (externalplus internal account). The left-hand portionof each table presents results when the depen-dent variable is borrowing according to officialrecords. The right-hand portion adjusts thedependent variable for use of multiple namesto borrow. F-tests of the hypothesis that‘‘months as a rank-and-file member’’ and‘‘months as a committee member’’ are equaldeterminants of borrowing are presented atthe end of each table.

The results make strikingly clear the extent towhich committee members borrow more thanrank-and-file members and the extent to whichthis difference is magnified when adjusting forthe use of multiple names. In the adjusted (right

Table 4. Weighted t-tests on wealth of committee

Variable (in baht) Committee(n =

Household wealth (assets less debt) 859,7Female-owned wealth (assets less debt) 546,4Male-owned wealth (assets less debt) 312,9Value of household land 722,2Value of female-owned land 495,4Value of male-owned land 226,8Value of household nonland assets 224,6Value of female-owned nonland assets 59,68Value of male-owned nonland assets 164,6

Table 5. Weighted t-tests on land value of committee and ran

Variable (in baht) Committee membe(n = 40)

Value of household land 729,190Value of female-owned land 481,140Value of male-owned land 248,050

column) borrowing equations, committeemembers borrow more than twice as much asrank-and-file members: real external accountborrowing increases by 4,953 baht per monthfor a committee member, and by 2,359 bahtfor a rank-and-file member; internal accountborrowing increases by 8,704 baht per monthfor a committee member, and by 4,236 bahtfor a rank-and-file member; combined borrow-ing increases by 5,960 baht per month for arank-and-file member, and by 12,982 baht permonth for a committee member. 25 For bothexternal and internal account borrowing, thep-value on the difference between the coeffi-cients is .000. Overall, committee members,who make up 15.6% of the treatment villagebank members (and 3% of all households intreatment villages), borrowed 27.0% of totalloan volume.

Given the large and significant difference inborrowing by rank-and-file members and com-mittee members, combined with the goal of tar-geting the poor, the question naturally arises asto whether committee members are richer orpoorer than rank-and-file members prior to vil-lage bank intervention. Therefore, Tables 4 and5 present t-tests on the same wealth measuresused in Tables 1 and 2, but now test for differ-ences between committee members and rank-and-file members.

On virtually all measures, committee mem-bers are wealthier before village bank interven-

and rank-and-file members in control villages

members16)

Rank-and-file members(n = 124)


47 554,180 .14526 302,851 .19677 244,190 .57525 422,115 .14113 253,480 .18913 166,619 .60230 171,554 .2076 64,143 .86401 101,016 .061

k-and-file members in all villages five years before surveys

rs Rank-and-file members(n = 254)


409,354 .033264,144 .093141,392 .208


tion. When examining differences only in thecontrol villages (Table 3), however, only onedifference is significant beyond the .14 level—almost certainly because of the small numberof committee members (n = 16) from those vil-lages in the sample. When using land value fiveyears before our surveys, however, and expand-ing the sample to include treatment villages, thedifference in total land value is significant at the.03 level, and the difference in female-ownedland value is significant at the .09 level.

Hence, there is evidence that the wealthiesthouseholds in the villages play a dominant rolein the village banks, and use their positions toborrow significantly more than rank-and-filemembers. In some villages, this phenomenoncaused serious problems between committeemembers and rank-and-file members. Forexample, one of the oldest village banks in thesample originally had over 70 members, butat the time of our surveys, the number was40. Many of the original members resigned be-cause they were angry that the president (whowas also the village chief’s wife) and treasurerwere borrowing much more than rank-and-filemembers. 26 In another village, some rank-and-file members resented the fact that com-mittee members ‘‘rounded down’’ members’external account loan amounts (e.g., from2,230 baht to 2,000 baht), ostensibly becauseit was easier to calculate interest payments.The surplus, which could be large when allmembers’ loans were considered, was thenloaned out by the committee members to them-selves, the village chief, or others of theirchoice. A small number of members had re-signed as a result. In several villages, somemembers who had resigned were surprisedand angered to learn during our interviews thattheir names were still being used to borrowwithout their knowledge or permission.

6. IMPACT

To appreciate the bias potentially arisingfrom self-selection and endogenous programplacement, as discussed in Section 1, considerthe following empirical specification:

Bij ¼ X ijaB þ V jbB þ eij; ð1ÞY ij ¼ X ijaY þ V jbY þ BijdY þ lij; ð2Þ

where Bij is the amount borrowed from the vil-lage bank by household i in village j; Xij is avector of household characteristics; Vj is a vec-

tor of village characteristics; Yij is an outcomeon which we want to measure impact; aB, bB,aY, bY, and dY are parameters to be estimated;and eij and lij are errors representing unmea-sured household and village characteristics thatdetermine borrowing and outcomes, respec-tively. dY is the primary parameter of interestas it measures the impact of village bank crediton the outcomes Yij.

Econometric estimation of this equation sys-tem will yield biased parameter estimates if eij

and lij are correlated and this correlation isnot taken into account. Correlation betweeneij and lij can arise both from self-selection intothe village bank (and the subsequent decisionon how much to borrow) and endogenousprogram placement. For instance, if the moreentrepreneurial households join the villagebank, then unmeasured ‘‘entrepreneurship’’would influence both the decision to become amember (and, for members, how much to bor-row) and impact measures such as income andassets. In this case eij and lij would be corre-lated, and estimation of village bank impactwould be biased. Similarly, if program place-ment is endogenous, the eij and lij will be corre-lated across villages and, again, impactestimates will be biased.

Coleman (1999) demonstrates that the un-ique survey design of this study can beexploited to obtain unbiased (in the case ofuncensored dependent variables) or consistent(in the case of censored dependent variables)estimates of average program impact with thefollowing specification 27:

Y ij ¼ X ijaþ V jbþMijcþ VBMOSijdþ gij; ð3Þ

where Yij is an outcome of household i in vil-lage j on which we want to measure programimpact; Xij is a vector of household characteris-tics; Vj is a vector of village fixed effects (villagedummy variables); Mij is a membership dummyvariable equal to 1 if household ij is selectedinto the microfinance program, and 0 other-wise; and VBMOSij is the number of monthsvillage bank credit has been available to mem-bers. The membership dummy variable Mij isa proxy for the unobservable characteristicsthat lead a household to self-select into thevillage bank and that might affect outcomes.It equals one for both ‘‘treatment’’ members(who have received program support) and‘‘control’’ members (who have not yet receivedprogram support). The variable VBMOSij

measures the extent of program availability to


members who have self-selected. Unlike theamount borrowed, it is exogenous to the house-hold (positive in varying amounts for treatmentmembers, and zero for control members).Inclusion of nonparticipants in the sample,combined with the use of village fixed effects,controls for possible endogenous programplacement. In this specification, d is an unbi-ased or consistent measure of impact per monthof program availability.

Coleman (1999) found that, when controllingfor endogenous member selection and programplacement, average program impact was notsignificant. A more refined impact estimation,however, will account for the apparentdifference in access to loans by rank-and-filemembers and committee members, who maytherefore experience different levels of impact.Because of this difference, the empirical specifi-cation of Eqn. (3) will be modified as follows.First, the dummy variable Mij will be replacedby two dummy variables: a rank-and-file mem-ber dummy variable to capture unobservabledifferences between rank-and-file members andnonmembers, and a committee member dummyvariable to capture similar unobservable differ-ences between rank-and-file and committeemembers. 28 Second, the regressor VBMOSij

will be replaced by two regressors: months ofrank-and-file membership and months of com-mittee membership, which will allow for differ-ential impact to be measured on rank-and-filemembers and committee members. Hence, theempirical specification to be estimated in thispaper is as follows:

Y ij ¼ X ijaþ V jbþMRijcR þMCijcC

þ RMOSijdR þ CMOSijdC þ lij; ð4Þ

where MRij is a dummy variable equal to oneif the household has a rank-and-file memberand zero otherwise; MCij is a dummy variableequal to one if the household has a villagebank committee member and zero otherwise;RMOSij is the number of months of rank-and-file membership; CMOSij is the numberof months of committee membership; and theother variables are defined as before. Again,MRij and MCij equal one for treatment andcontrol participants, and equal zero for non-participants in all villages. RMOSij and CMO-Sij measure the different access to programloans by rank-and-file and committee mem-bers, respectively, and are exogenous to thehousehold (equal to zero for nonparticipantsand treatment participants, but positive and

varying for treatment participants). Inclusionof nonparticipants in the sample, combinedwith the use of village fixed effects, again con-trols for possible endogenous program place-ment. In this specification, dR measures theimpact of an additional month’s programavailability to a rank-and-file member; and dC

measures the impact of an additional month’sprogram availability to a committee member.F-tests can be conducted to determine ifdR = dC, for example, if the impact of the vil-lage bank on rank-and-file members and com-mittee members is equal.

Table 11 in Appendix A presents the com-plete output of a typical regression (for house-hold nonland assets); and Table 12 presentsthe coefficients and associated p-values for theregressors of interest (months of rank-and-filemembership and months of committee mem-bership) in all the regressions. 29

(a) Impact on rank-and-file members

The results presented in Table 12 for rank-and-file members are consistent with the averageimpacts found in Coleman (1999)—that is, theimpact estimates are largely insignificantly dif-ferent from zero. In fact, the only measures thatare significantly different from zero indicate anegative impact. It is noteworthy that men’soverall self-employment expenses (coef =�9,986; p = .077) are negative and statisticallysignificant, and men’s overall self-employmentsales (coef = �11,023; p = .106) are negativeand nearly significant at the 10% level. Esti-mated impact on men’s agricultural expensesand business sales and expenses, and labor timeare all negative and statistically significant, andestimated impact on all other measures ofmen’s economic activity is also negative thoughnot statistically significant. Assuming thatmen’s leisure is a normal good, these resultsmay represent (expected) income effects of pro-viding low-cost credit to women: essentially,women’s increased income generated by accessto program credit could provide men with eco-nomic rents taken in the form of increasedleisure. Unfortunately, however, these rentsappear to be perceived rather than realized, asthe measured impact on women’s outcomes inrank-and-file households is not significant.

Another important result is the lack of im-pact on savings for rank-and-file members. Infact, measured impact is negative, though in-significantly different from zero. One possibleexplanation for this result is that, especially


after the third year, members have reducedincentive to save. Recall that external accountloan size grows according to a member’s sav-ings, up to a maximum of 7,500 baht. In prac-tice, however, although external account loansare made for five years, the NGOs start reduc-ing loan size after the third year, so as to grad-ually wean members from external accountloans. Hence, members have little incentive(in term of access to external account loans)to continue to save after a certain point. 30

Moreover, because member savings make upthe internal account fund, and because thisfund is often monopolized by the committeemembers, rank-and-file members may be fur-ther inclined to reduce their village bank sav-ings. 31 This lack of impact on mobilizingsavings from rank-and-file members could havemajor consequences because the long-term sus-tainability of village banks depends on the sus-tainability of their internal account funds.

(b) Impact on committee members

Estimated impact on committee members,however, is significant and positive on a rangeof dependent variables. Impact results are pre-sented below under the headings of physical as-sets; savings, debt, and lending; production,sales, expenses, and labor time; and health careand education.

(i) Physical assetsThe village banks appear to have had a large

and positive impact on the value of committeemembers’ household wealth (coef = 3,010; p =.043), and this impact is seen primarily on wo-men’s wealth (coef = 1,705; p = .076). Brokendown further into landed wealth and nonlandassets, there is a positive and significant impacton total household nonland assets (coef =3,187; p = .018). Interestingly, this effect is mostpronounced for men’s nonland assets (coef =2,122; p = .071), although further refinementsindicate positive and significant effects on wo-men’s productive assets (coef = 870; p = .014),including nonland farm assets (coef = 329;p = .100) and consumer durables (coef = 755;p = .026). F-tests shown in the last two columnsof Table 12 indicate that the impact on commit-tee members is significantly greater than that onrank-and-file on many of these wealth measures(household and women’s wealth, household andwomen’s nonland assets, household and wo-men’s productive assets, household and men’slivestock, and women’s consumer durables).

(ii) Savings, debt, and lendingThe impact of the village banks on household

savings is positive and significant for committeehouseholds (551; p = .073). This effect appearsto come not only from the impact on women’ssavings (217; p = .118), but also on men’s sav-ings (425; p = .161), although neither gender’sindividual coefficients are significantly differentfrom zero. F-tests show that the impact oncommittee household savings is significantlygreater than the impact on rank-and-file house-hold savings.

According to NGO staff, the primary goalof these village bank programs is to allowmembers to reduce their high-interest debt tomoneylenders. Hence, I estimated village bankimpact on ‘‘high-interest’’ debt, defined asdebt with an interest rate greater than 2% permonth, the rate charged on external accountloans. However, no coefficients were signifi-cantly different from zero, indicating thatvillage bank credit is not substituting forhigh-interest credit. 32 I also estimated the im-pact on ‘‘low-interest’’ debt (debt with an inter-est rate less than or equal to 2% per month) todetermine if village bank credit is merely substi-tuting for existing sources of low-interest credit,such as that from BAAC, or if it was allowing(or encouraging) households to mobilize addi-tional institutional credit. In fact, the resultsindicate that committee member householdsare also increasing their low-interest debt fromother sources available to men in the house-hold. In practice, this additional debt comesprimarily from BAAC. This is an interestingand useful result for the following reason. Thedifferential impacts measured between commit-tee members and rank-and-file members couldbe the result of two different factors: either itis the result of the differential access to fundsas discussed above, and this differential impactallows committee members to invest in differenttypes of projects, perhaps with different scaleeconomies; or it is the result of different unob-servable characteristics (e.g., entrepreneurship)that can be harnessed or realized only by accessto credit. The fact that access to village bankfunds appears to be allowing or encouragingadditional borrowing is one indication thatcommittee members are investing in differenttypes of projects, perhaps with larger fixed as-sets, that require larger capital. This is also con-sistent with the result discussed above thatwomen’s productive assets are positively af-fected by the village bank. Hence, differentialaccess to program loans does appear to matter.


In four of eight treatment villages and threeof six control villages, at least one (and oftentwo) committee member(s) engaged in money-lending to some degree, and some nonmem-bers and rank-and-file members complainedthat committee members borrowed from thevillage bank, then lent the money out at higherinterest rates. Therefore, I also estimated theimpact of village bank loans on moneylending.Indeed, there is evidence of positive impacton committee members’ moneylending. Thecoefficient on the impact of committee mem-bers on household moneylending is positiveand highly significant (2,653; p = .005),and this effect comes entirely through the wo-men in the household (2,093; p = .054). 33

The difference between coefficients on commit-tee and rank-and-file months for householdlending with interest is also significant (p =.035).

(iii) Production, sales, expenses, and labor timeAs Table 12 indicates, the village banks have

also had a positive and significant impacton women’s self-employment sales (2,245; p =.023) and expenses (1,520; p = .058) in commit-tee households. F-tests indicate that the impacton sales and expenses of women in committeehouseholds is significantly greater than the cor-responding measures for rank-and-file house-holds (p = .058 for sales, and p = .051 forexpenses). Refined impact estimates indicate apositive effect on household agricultural pro-duction, household livestock production, wo-men’s livestock sales, and women’s businesssales and expenses in committee member house-holds. All impact measures are significantlygreater for committee members than rank-and-file members.

The impact on total household self-employ-ment labor time in committee householdsis also positive and significant (coef = 26.9;p = .058), apparently impacting both women’slabor time (14.0; p = .112) and men’s time(12.9; p = .156), though the estimates by gen-der are not individually statistically significant.

(iv) Health care and educationMost estimates of village bank impact on

medical and school expenses (totals, by gender,and for children, by gender) 34 in both commit-tee households and rank-and-file householdsare insignificantly different from zero. The oneexception to this was educational expensesfor boys in committee member households(coef = 86.6; p = .035).

7. SUMMARY AND POLICYCONCLUSIONS

This paper has evaluated the targeting and im-pact of a women’s group-lending program inNortheast Thailand. To do so, it exploited aunique survey sample that included programparticipants from ‘‘treatment’’ villages that hadalready received program support, participantsfrom control villages that had not yet receivedprogram support, and nonparticipants fromboth types of villages. The results were presentedin terms of targeting (i.e., the processes of mem-ber selection and borrowing) to determine if theprogram has succeeded in reaching its targetgroup of the ‘‘poorest of the poor,’’ and in termsof impact on member households.

There is a strong evidence that, similar to theprevious efforts to deliver financial services tothe rural poor in developing countries, the pro-grams surveyed are not reaching the poor asmuch as the relatively wealthy. Weighted t-testsindicate that, prior to program intervention,participant households are significantly wealth-ier than those of nonparticipants, and thewealth difference is explained primarily by thevalue of female-owned land in the households.Moreover, the probability of the wealthiest vil-lagers selecting into the program is nearly twicethat of the poorer villagers. Weighted logit esti-mates confirm that the value of female-ownedland is a significant determinant of memberselection. The same estimates, however, indi-cate that ‘‘creditworthiness’’ of female house-hold members, as measured by a specialsurvey of village informants, is also a significantdeterminant of member selection. Hence, thereis evidence that both relatively public informa-tion on land holdings (the same informationcommonly used by commercial lending institu-tions in Thailand to select its rural customers)and local information on creditworthiness arebeing used to select village bank members.

There is also a strong evidence that the rich-est village bank members become committeemembers (president, vice president, and trea-surer) and use their position to borrow signifi-cantly more from the village bank (both fromthe external account and the internal account)than rank-and-file members. One method com-monly employed by the committee members tocircumvent village bank external account bor-rowing limits is to use multiple village bankaccounts, each under a different name.

Hence, it is clear that, within the context ofNortheast Thailand, village banks’ small loan


size and frequent meetings, as well as the stigmaof belonging to a poverty lending program, donot discourage the relatively rich villagers fromparticipating in the village bank. Furthermore,although some of the poorer nonmembers re-ported choosing not to join the village bank,many others were excluded against their wishes.Some never knew of the village bank before thesurveys, while others felt intimidated to join be-cause they considered the village bank to be aprogram for the relatively wealthy in the vil-lage. Moreover, the process by which NGOsfirst contact the village political structure, rep-resented by the village chief, and then allowhim to organize the selection process likely con-tributes to the richest women in the villagebecoming committee members, who then bor-row the lion’s share of village bank loans.

Impact results indicate a positive impact of thevillage bank program on several measures ofhousehold welfare. Given the difference in accessto loans by committee members and rank-and-file members, however, it is found that theestimated impact on committee members is sig-nificantly larger than the impact on rank-and-filemembers. For example, positive and significantimpact is observed in committee member house-holds on many important measures of wealth,savings, income, productive expenses, and labortime. Impact is also positive on committee mem-bers’ moneylending activities, as they apparentlyborrowed from the program at lower interestrates and relent at higher rates. Impact on out-comes for rank-and-file members was largelyinsignificant, although men’s economic activityappears to have decreased as a result of the pro-gram, possibly the result of increased leisure con-sumption allowed by their perception of theirwives increased economic activity.

The differential impact measured betweencommittee members and rank-and-file memberscould be the result of the differential access toloans, with committee members’ increased ac-cess allowing them to invest in different typesof projects, perhaps with greater returns toscale; or it could be the result of different unob-servable characteristics (e.g., entrepreneurship)that can be harnessed or realized only by accessto credit. However, the fact that access to vil-lage bank loans appears to also encourageadditional borrowing from other institutionalsources is consistent with the first possibility,namely that committee members are investingin different types of projects, perhaps with lar-ger fixed assets, that require larger capital. Dif-ferential access to loans does matter.

There are several policy implications of theseresearch findings. First, these and other pro-grams whose intention is to reach the poorshould seriously consider imposing (and enforc-ing) membership eligibility criteria similar tothat used by the Grameen Bank (e.g., maximumallowable land holdings or other measuresof wealth) in order to more actively target thepoor. 35 Beyond this type of restriction, how-ever, villagers should still self-select for member-ship since evidence presented here indicates thatlocal information about each household’s cred-itworthiness is used to screen members. NGOsshould also enforce the rule that village banksare to select new committee members annuallyso that the committee members do not becomeentrenched in their positions, 36 which onlyencourages abuses—in the two villages wheresome committee turnover was observed, theuse of multiple names was greatly diminished,and borrowing was much more equitable be-tween committee and rank-and-file members.Clearer and more frequent public pronounce-ments of the village bank’s goals and targetgroup made by the NGO fieldworkers to the vil-lagers, as well as more strict enforcement bythem of existing rules against the use of multiplenames, would also assist in reaching the poor.One goal of village banks is to ‘‘empower’’ thepoor, especially poor women. But it is naıve tothink that the existing village power structureswill not pursue their own self-interest and usethe village bank to enhance their own power ifgiven the opportunity; and it is equally naıveto think that the relatively poor and powerlesscan be empowered without more active involve-ment of program administrators at least to en-sure that they are the beneficiaries of thesepoverty lending programs. Along these lines, vil-lage bank rules regarding eligibility, limits onborrowing, and election of committee memberscould be printed on passbooks that each mem-ber receives, or posted in the village, 37 thushelping to eliminate some of the informationasymmetries that currently exist within the vil-lage regarding what is and is not acceptablewithin the village bank. Informal interviewsindicated that the lack of such commonly recog-nized rules restricts efforts of the poor to gain ac-cess to the program and restricts efforts by therank-and-file to equalize access to loans by elim-inating the ‘‘legal’’ basis of their arguments thatthe wealthy committee members are abusingtheir positions.

Caution should be exercised before extrapo-lating these results to other contexts. Thailand


is a relatively wealthy developing country, andmany villagers already have access to low-inter-est credit from financial institutions such as theBAAC. The average wealth of survey house-holds was over 500,000 baht, and averagehousehold low-interest debt, excluding villagebank debt, was over 30,000 baht. In this con-text, loans of 1,500–7,500 baht may have a lim-ited impact. Many women surveyed stated thatthe size of village bank loans was too small, andsome women left the program for that reason.It may not be surprising, then, that the largestimpact was seen on committee members, whowere able to circumvent the loan ceilings. Infact, given the inappropriately small loan size,it is understandable that the influential villagers

would manipulate the system to obtain largerloans. It is arguable that this sort of manipula-tion would be reduced if program loan sizes in-creased.

Further research should be conducted onmicrofinance programs in other parts of theworld to determine if these results are uniqueto Northeast Thailand or are typical of micro-finance programs in general. The researchmethods and survey design used here could beeasily implemented elsewhere. Since mostmicrofinance programs regularly expand tonew villages, which could be used as con-trols, this type of survey could be widely under-taken.

NOTES

1. See, for example, The Economist, 1993; Malveaux,1997; New York Times, 1997; San Francisco Examiner,1999, 1990.

2. Recent studies of note on microfinance impact,targeting, or both include Alexander (2001), Coleman(1999), Hashemi (1997), Hulme and Mosley (1996),Khandker (2005), McKernan (2002), Montgomery(2005), Park and Ren (2001), Pitt and Khandker(1998), and Wydick (1999a, 1999b). For discussions ontheoretical aspects of group lending (see Armendariz deAghion, 1999; Armendariz de Aghion & Morduch, 2005;Besley & Coate, 1995; Coleman, 2005; Conning, 1999;Ghatak, 1999; Ghatak & Guinnane, 1999; Stiglitz, 1990;Van Tassel, 1999; Varian, 1990).

3. For a thorough econometric discussion of the issuesinvolved see Coleman (1999) and Pitt and Khandker(1998).

4. See also Grosh (1995) and Ravallion and Datt(1995) for methodological discussions on targeting.

5. See Foundation for International Community Assis-tance (1990) and Hatch (1989).

6. In 1995–96, US$1 = 25 baht.

7. In three of the older village banks surveyed, 4-monthloan cycles were used for the first year. Subsequently, theNGOs switched to 6-month loan cycles in response tomember demand.

8. Interestingly, final-period external account loanswere always paid back in full, contradicting the eco-

nomic theory that finite-period games would lead todefault. Clearly, the villagers either feel a moral obliga-tion to repay, are concerned with their long-termreputations with the NGOs, or have not been properlyeducated on the subject of finite-period games!

9. Originally, interest payments were made at weeklymeetings, but were eliminated due to member demand.Meeting frequency was also reduced from weekly tomonthly. However, with all payments due every sixmonths, there is often little business to conduct at themonthly meetings, which are therefore poorly attended.

10. There was occasionally some shifting within thecommittee—for example, the president and treasurerswapping positions.

11. This village bank opened in April 1995 and is,depending on the empirical test being conducted, some-times a treatment village and sometimes a controlvillage. For example, when estimating the impact ofvillage bank loans on various assets, as measured by thefirst survey, it is a control village since it had not yetbenefited from village bank loans; but when estimatingthe impact on income or expenditures over the 12-monthperiod of the surveys, it is a treatment village since itsfirst two 6-month loans would have impacted incomeduring this period.

12. Such households might drop out because they arenot benefiting from, or are being negatively affected by,the program, for example, if their investments fail. Buthouseholds might also drop out if they benefitedsufficiently that they either graduate to a more formallending institution or are able to use the profits from asuccessful investment to finance additional investments.


13. Questionnaires were checked for internal consis-tency in the villages immediately upon completion of aninterview. Approximately 20% were subsequentlychecked for content in follow-up interviews.

14. For an overview of the Thai rural credit system, seeSiamwalla et al., 1990.

15. For example, households owning assets that areworth more than the value of one acre of land aretheoretically ineligible to join the Grameen Bank. Twoother Bangladeshi group lending programs, the Bangla-desh Rural Advancement Committee (BRAC) and theBangladesh Rural Development Board (BRDB), deemineligible households that own more than 0.5 acres ofland. Morduch (1999), however, shows that these eligi-bility criteria are not perfectly enforced, and some 18–32% of participants do not meet the eligibility criteria.

16. Collecting data on land owned five years earlier isrelatively easy. Because land transactions tend to belarge and important for a household, yet relativelyinfrequent, households can easily recall land transac-tions made in the previous five years, and land owned(and its value) five years earlier can be deduced.

17. Land value five years before our surveys is also animportant regressor in estimating village bank impact onvarious outcomes as it controls for initial wealth in allhouseholds. (See Section 6.)

18. Observations from each of 28 strata (determined byvillage and village bank membership status) wereweighted by the inverse of their sampling probabilities.

19. Because of different survey methods used in thepresent study and those used in measuring poverty inThailand, no attempt is made in this study to assesswhether a household is above or below the officialpoverty line in Thailand. Instead, the study only makescomparisons among households surveyed. At the time ofthe survey, the official national poverty rate in Thailandwas 11.4%. The Northeast, where this survey wasconducted, had the highest poverty rate, at 19%. Hence,as a rough estimate, the poorest group in Table 3,representing the relatively poorest 17% of households inthe villages surveyed, could be considered below thepoverty line.

20. Whereas households sometimes had trouble desig-nating an individual owner of nonland assets, they rarelyhad trouble designating an individual owner of each plotof land. Moreover, since Thailand is traditionally amatrilineal society, with land being passed downthrough the woman’s side of the family, it should notbe surprising that women own more land than men.

21. Wenner (1995) also shows that groups thatscreened members and used local information had lowerdelinquency rates than those that did not.

22. Sampling weights (the inverse of the probability ofbeing sampled) were applied to the data for these logitregressions because stratification was done on thedependent variable (member status), whereas weightingis not required in the impact regressions (see below)because stratification was done on the exogenous vari-ables that appear there (Maddala, 1983, p. 170).

23. The estimates on the right-hand side of Table 7,however, are possibly biased because ‘‘creditworthiness’’is likely to be endogenous for the village bank membersin treatment villages: the village chief’s and village bankpresident’s assessment of their creditworthiness wouldbe influenced by their performance in the village bank.

24. This result is interesting, especially when comparedwith the methodology of Park and Ren (2001), whoevaluate targeting and impact in China using respon-dents’ own subjective determination of their ‘‘eligibility’’in programs that do not have explicit eligibility criteria.

25. Borrowing is actually measured in baht-months toaccount for the different lengths of time that money isborrowed. For example, 1,000 baht borrowed for 6months is 6,000 baht-months. This measure is necessarybecause early external account loan cycles lasted 4months but were subsequently adjusted to 6 months inresponse to client demand. However, the results arerobust to other measures of borrowing (e.g., averageloan size outstanding).

26. Members in this and other villages were primarilyangry over internal account borrowing, since the NGOstill provided the rank-and-file members with the exter-nal account loans they were entitled to. In principle,internal account lending policy is determined by thevillage bank membership, but in practice by thecommittee members. An examination of Table 9 showsthat in fact no household characteristics other than‘‘months as a rank-and-file member’’ and ‘‘months as acommittee member’’ have a significant impact on creditdemand from the internal account, demonstrating thedominant role of committee members in allocatinginternal account funds.

27. Recent studies have applied this method, includingMontgomery (2005). Ashraf, Karlan, and Yin (2006) gofurther, by conducting a fully randomized experimentin the Philippines. Specifically, they worked with theGreen Bank to fully randomize new program placementand clients to whom a particular savings product wasoffered. Their method thus addresses self-selection and


endogenous placement bias more cleanly. Such fullrandomization is also advocated by Duflo and Kremer(2003), and to be sure if it is superior from a researchpoint of view to the method used by Coleman (1999) andin the current paper. However, in practice, such fullrandomization will be difficult to apply often. First, itrequires extraordinary cooperation from the programmanagers. Second, it can be in conflict with someprogram objectives. For example, as microfinance hasbecome more and more commercially oriented in recentyears, managers will care deeply where they locate newoperations, to ensure program sustainability, and maybe resistant to randomly choosing new locales. Hence,the method of Coleman (1999) will be more practical inmany situations.

28. As mentioned in Section 2, committee members areto be chosen every year, which in theory could bias theresults because in the control villages, we know thecommittee members only for the first year. In practice,however, the committee members rarely change. In six ofthe treatment villages, the committee has not changed atall; in the two other treatment villages, there have beenonly minor changes, with many of the core committeemembers only shifting committee positions (e.g., presi-dent to vice president).

29. Variables were entered in linear form. In the vastmajority of regressions, no additional explanatory valuewas gained by introducing nonlinearities. Note also thatthe dependent variables presented in Table 12 are notindependent of each other. Most aggregate variables(e.g., total household wealth) are broken down intomore refined measures (e.g., women’s wealth, totalhousehold productive assets, and men’s business assets)in order to identify exactly where the impacts occur.

30. Nonlinear specifications, however, did not add anyexplanatory value.

31. The advantage of having a saving facility in thevillage was also virtually eliminated in two village bankssurveyed, largely because of the policies set by thecommittee. In one village, members were not allowedonly to save—all members were required to borrow from

the NGO, even if they did not want to. The reason givenby the committee members was that it would be‘‘impolite’’ for members to refuse the assistance offered.Twenty of this village bank’s 30 members regularly hadto borrow from moneylenders to repay their village bankloans. In another village bank, the president decided topay members their end-of-year interest income, notbased on average yearly savings, but (ostensibly for easeof calculation) on savings on December 31. She thengreatly increased her own savings on December 28 so asto increase her share of interest income.

32. Estimation of impact on male high-interest debtin committee households was not possible becauseonly four such households had men with high-interestdebt.

33. Estimation of impact on men’s moneylending wasimpossible because of the small number of householdsreporting nonzero values.

34. ‘‘Female medical expenses’’ are expenses forwomen, not necessarily paid by women. No attemptwas made to learn which household members paid forthe medical care. The same is true for school expenses.

35. This is not to advocate a return to directed lending,and certainly not at subsidized interest rates. Sustain-ability of microfinance programs is of paramountimportance, and the poor are best served by sustainableinstitutions. Nor do I argue that all microfinanceprograms should target the ‘‘poorest of the poor.’’Rather, the argument here is that those programs thatdo wish to target the poor may have to be more active inensuring that the poor are indeed reached.

36. At least in Thailand, literacy is not an issue inchoosing committee members: approximately 80% ofwomen surveyed reported being able to read and write.

37. All surveyed villages had a village meeting placewhere other information on rice banks, buffalo banks, orgeneral village information was posted or could beposted.

REFERENCES

Adams, D., Graham, D., & von Pischke, J. D. (1984).Undermining rural development with cheap credit.Boulder, Colorado: Westview Press.

Adams, D., & Vogel, R. (1986). Rural financial marketsin low-income countries: Recent controversies andlessons. World Development, 14, 477–488.

Adams, D., & von Pischke, J. D. (1992). Microenterprisecredit programmes: deja vu. World Development, 20,1463–1470.

Alexander, G. (2001). An empirical analysis of micro-finance: Who are the clients? Paper presented at 2001Northeastern Universities Development ConsortiumConference.

Amin, S., Rai, A., & Topa, G. (2003). Does microcreditreach the poor and vulnerable? Evidence fromNorthern Bangladesh. Journal of Development Eco-nomics, 70(1), 59–82.


Armendariz de Aghion, B. (1999). On the design of acredit agreement with peer monitoring. Journal ofDevelopment Economics, 60, 79–104.

Armendariz de Aghion, B., & Morduch, J. (2005).The economics of microfinance. Cambridge: MIT Press.

Ashraf, N., Karlan, D., & Yin, W. (2006). Tyingodysseus to the mast: Evidence from a commitmentsavings product in the Philippines. Quarterly Journalof Economics, 121(2), 635–672.

Bank for Agriculture and Agricultural Cooperatives(1997). Annual Report. Bangkok.

Besley, T., & Coate, S. (1995). Group lending, repay-ment incentives and social collateral. Journal ofDevelopment Economics, 46, 1–18.

Chen, M. (1992). Impact of Grameen Bank’s creditoperations on its members: Past and future research.Mimeo, Cambridge: Harvard University.

Coleman, B. (1999). The impact of group lending inNortheast Thailand. Journal of Development Eco-nomics, 60, 105–142.

Coleman, B. (2005). Risk, mutual assistance, and mutualinsurance among village bank members. Mimeo,Asian Development Bank.

Conning, J. (1999). Outreach, sustainability and leveragein monitored and peer-monitored lending. Journal ofDevelopment Economics, 60, 51–78.

Duflo, E., & Kremer, M. (2003). Use of randomizationin the evaluation of development effectiveness.Mimeo prepared for the World Bank Operationsand Evaluation Department Conference on Evalua-tion and Development Effectiveness.

Foundation for International Community Assistance(FINCA) (1990). Promoting and supporting villagebanks: Outline for a procedures manual. Washing-ton, DC.

Ghatak, M. (1999). Group lending, local informationand peer selection. Journal of Development Econom-ics, 60, 27–50.

Ghatak, M., & Guinnane, T. (1999). The economics oflending with joint liability: Theory and practice.Journal of Development Economics, 60, 195–228.

Grosh, M. (1995). Toward quantifying the trade-off:Administrative costs and incidence in targeted pro-grams in Latin America. In D. Van de Walle, & K.Nead (Eds.), Public spending and the poor: Theoryand evidence (pp. 450–488). Baltimore: John HopkinsUniversity Press.

Hashemi, S. (1997). Those left behind: A note on targetingthe hardcore poor. In G. Wood, & I. Sharif (Eds.),Who needs credit? Poverty and finance in Bangladesh(pp. 249–257). Dhaka: University Press Limited.

Hatch, J. (1989). A manual of village banking forcommunity leaders and promoters. Washington, DC.

Hulme, D., & Mosley, P. (1996). Finance against poverty.London: Routledge.

Khandker, S. (2005). Microfinance and poverty: Evi-dence using panel data from Bangladesh. WorldBank Econonmic Review, 19, 263–286.

Maddala, G. S. (1983). Limited-dependent and qualitativevariables in econometrics. New York: CambridgeUniversity Press.

Malveaux, J. (1997). Another way to look at banking.San Francisco Chronicle.

McKernan, S. (2002). The impact of microcredit pro-grams on self-employment profits: Do noncreditprogram aspects matter? Review of Economics andStatistics, 84(1), 93–115.

Microcredit Summit Report (1997). Washington, DC,Results Education Fund.

MkNelly, B., & Watetip, C. (1993). Impact evaluation offreedom from hunger’s credit with education programin Thailand. Mimeo, California: Davis.

Montgomery, H. (2005). Meeting the double bottomline—the impact of Khushhali Bank’s microfinanceprogram in Pakistan. Asian Development BankInstitute.

Morduch, J. (1999). Does microfinance really help thepoor? New evidence from flagship programs in Ban-gladesh. Mimeo, Princeton: Princeton University.

New York Times (February 16, 1997). Micro-loans forthe very poor.

Park, A., & Ren, C. (2001). Microfinance with Chinesecharacteristics. World Development, 29, 39–62.

Perry, D. (1995). Constraints to women’s village-bankingactivities: An anthropological study of village banks inthe Entente des Groupements Associes de Nganda.Mimeo, New Haven: Yale University.

Pitt, M., & Khandker, S. (1998). The impact of group-based credit programs on poor households in Ban-gladesh: Does the gender of participants matter?Journal of Political Economy, 106, 958–996.

Ravallion, M., & Datt, G. (1995). Is targeting througha work requirement efficient? Some evidence forrural India. In D. Van de Walle, & K. Nead (Eds.),Public spending and the poor: Theory and evidence(pp. 413–444). Baltimore: John Hopkins UniversityPress.

San Francisco Examiner (December 2, 1990). Women’sWorld Banking makes dreams come true.

San Francisco Examiner (May 2 1999). A little credit, abig success.

Sebstad, J., & Chen, G. (1996). Overview of studies onthe impact of microenterprise credit. AIMS Project,Washington, DC.

Siamwalla, A., Pinthong, C., Poapongsakorn, N., Sat-sanguan, P., Nettayarak, P., Mingmaneenakin, W.,et al. (1990). The Thai rural credit system: Publicsubsidies, private information, and segmented mar-kets. The World Bank Economic Review, 4, 271–296.

Stiglitz, J. (1990). Peer monitoring and credit markets.The World Bank Economic Review, 50, 350–366.

The Economist (February 27, 1993). Banking on the poor.Van Tassel, E. (1999). Group lending under asymmetric

information. Journal of Development Economics, 60,3–26.

Varian, H. (1990). Monitoring agents with other agents.Journal of Institutional and Theoretical Economics,146, 153–174.

Wenner, M. (1995). Group credit: A means to improveinformation transfer and loan repayment perfor-mance. Journal of Development Studies, 32, 264–281.

World Development Report: Financial systems and devel-opment (1989). New York: Oxford University Press.

Wydick, B. (1999a). The effect of microenterpriselending on child schooling in Guatemala. EconomicDevelopment and Cultural Change, 47, 853–869.


Wydick, B. (1999b). Credit access, human capital, andclass structural mobility. Journal of DevelopmentStudies, 35, 131–152.

Table 6. List

Variable name

Independent variables

Months as village bank rank-and-file memberMonths as village bank committee memberValue male-owned land five years ago (baht)Value female-owned land five years ago (baht)Sex of household head (F = 1)Education of male (years)Education of female (years)Number generations family in villageNumber relatives in villageVillage chief or assistant? (0/1)Is female a civil servant? (0/1)Is male a civil servant? (0/1)Does household have village bank member? (0/1)Number of females age 22–39Number of females age 40–59Number of females age 60 and overNumber of males age 22–39Number of males age 40–59Number of males age 60 and over

Variable name (in bahtunless stated otherwise)

Weighted mean(n = 445)

Standarddeviation

Dependent variables

Household wealth 529,586 742,452 Ho

Female-owned wealth 267,272 654,629 F

Male-owned wealth 256,640 389,261

Household land value 390,330 686,081

Female-ownedland value

218,379 628,373

Male-ownedland value

171,951 338,945

Household nonlandassets

172,819 199,889

Female nonland assets 58,050 123,455Male nonland assets 107,713 159,159Household productive

assets43,052 56,071 H

Female productiveassets

10,459 27,582

Male productive assets 31,064 50,504

APPENDIX A

See Tables 6–12.

of variables


Standard deviation

35.9 21.24.18 12.6

169,763 340,890217,333 628,873

.212 .415.16 3.274.82 2.583.44 1.138.05 6.84.0323 .21.0207 .15.0383 .20.463 .47.604 .56.523 .53.264 .46.593 .57.438 .52.206 .38

Variable name(in baht unless

stated otherwise)


Standarddeviation

usehold self-employmentproduction

135,215 1,273,136

emale self-employmentsales

29,852 101,596

Male self-employmentsales

93,825 1,269,232

Household agriculturalproduction

24,254 21,974

Female agriculturalsales

6,160 13,145

Male agriculturalsales

8,163 14,513

Household livestockproduction

5,035 9,974

Female animal sales 2,839 7,332Male animal sales 2,195 6,602

ousehold business sales 104,791 1,271,791

Female business sales 20,853 100,446

Male business sales 83,466 1,267,907

Table 6—continued



Standarddeviation


stated otherwise)


Standarddeviation

Household nonlandfarm assets

15,934 20,749 Household self-employmentexpenses

108,963 1,200,117

Female nonlandfarm assets

3,485 10,509 Female self-employmentexpenses

23,540 90,119

Male nonlandfarm assets

11,676 18,735 Male self-employmentexpenses

84,182 1,195,831

Household livestock 15,205 14,366 Household agriculturalexpenses

12,044 10,974

Female-ownedlivestock

4,391 10,147 Female agriculturalexpenses

4,634 7,026

Male-owned livestock 10,068 13,693 Male agriculturalexpenses

7,408 9,780

Household businessassets

11,913 45,793 Household animal-raisingexpenses

3,401 8,783

Female businessassets

2,583 20,890 Female animal-raisingexpenses

1,627 6,131

Male business assets 9,320 40,735 Male animal-raisingexpenses

1,653 5,531

Household consumerdurables

32,340 83,611 Household businessexpenses

92,715 1,196,996

Female-ownedconsumer durables

14,584 56,455 Female business expenses 17,279 88,911

Male-owned consumerdurables

15,585 60,394 Male business expenses 75,121 1,194,188

Value of house 82,551 109,238 Household self-employmentlabor hours

3,488 2,122

Household cashsavings

13,574 37,730 Female self-employmentlabor hours

1,695 1,221

Female cash savings 6,383 16,369 Male self-employmentlabor hours

1,792 1,352

Male cash savings 7,191 29,499 Household medicalexpenses

2,606 6,100

Household low-interestdebt (62%/mo.)

31,330 103,351 Medical expensesfor females

1,281 3,137

Female low-interestdebt (62%/mo.)

9,342 46,252 Medical expensesfor males

1,325 5,303

Male low-interestdebt (62%/mo.)

21,775 73,630 Medical expensesfor children

573 1,478

Household high-interestdebt (>2%/mo.)

7,386 22,842 Medical expensesfor girls

284 1,015

Female high-interestdebt (>2%/mo.)

3,928 14,810 Medical expensesfor boys

289 1,111

Male high-interestdebt (>2%/mo.)

3,458 16,577 School expensesfor children

2,430 3,918

Household loaningout at interest

3,823 27,027 School expensesfor girls

1,079 2,250

Female loaningout at interest

3,104 25,950 School expenses for boys 1,351 3,014

External accountborrowing,according to records(baht-months)

85,722 57,587 External accountborrowing, adjusted

(baht-months)

87,203 82,837

(continued next page)


Table 6—continued



Standarddeviation


stated otherwise)


Standarddeviation

Internal accountborrowing,according to records(baht-months)

88,250 102,275 Internal accountborrowing, adjusted

(baht-months)

101,910 157,371

Total VB borrowing(ext. + int. acct.),according to records(baht-months)

173,972 144,898 Total VB borrowing(ext. + int. acct.),

adjusted (baht-months)

189,112 226,822

Table 7. Logit estimates of the determinants of village bank member selection

Independent variable In control villages only In all villages

Coefficient p-Value Coefficient p-Value

Creditworthiness score .709 0.046 .384 0.097Value female-owned land 2.73 · 10�6 0.025 – –Value male-owned land 1.01 · 10�6 0.157 – –Value female-owned nonland assets �1.96 · 10�6 0.500 – –Value male-owned nonland assets �2.03 · 10�6 0.236 – –Value female-owned land five years ago – – 2.44 · 10�6 0.001Value male-owned land five years ago – – 2.40 · 10�6 0.509Number of males, age 22–39 �.0839 0.848 �.260 0.314Number of males, age 40–59 �.378 0.516 �.0803 0.823Number of males, age 60 and over �1.199 0.070 �.784 0.052Number of females, age 22–39 .320 0.519 .0445 0.874Number of females, age 40–59 .539 0.377 �.0106 0.976Number of females, age 60 and over .202 0.736 �.356 0.294Household head female? (0/1) �1.914 0.010 �.948 0.029Highest female education level .00204 0.986 .0744 0.249Highest male education level �.0172 0.814 �.0383 0.417Generations family in village .117 0.585 .0205 0.849Number blood relatives in village �.0347 0.385 .0191 0.312Household member chief or assistant? (0/1) .0100 0.993 1.905 0.071Female have civil servant job? (0/1) �1.008 0.609 �1.0912 0.380Male have civil servant job? (0/1) �.168 0.860 .893 0.244Constant �1.354 0.276 �1.191 0.086

Number of obs = 167 Number of obs = 444F(19, 148) = 1.16 F(17, 427) = 2.32

Prob > F = 0.2992 Prob > F = 0.0022

Table 8. Cumulative borrowing from village bank external account (in baht-months)

Independent variable Dependent variable:external account

borrowing according tothe records

Dependent variable:external account

borrowing, adjusted foruse of multiple names


Months as rank-and-file VB member 2,064 0.000 2,359 0.000Months as VB committee member 3,001 0.000 4,953 0.000Value female-owned land five years ago �.0014 0.743 �.00687 0.301


Table 8—continued

Independent variable Dependent variable:external account


Dependent variable:external account



Value male-owned land five years ago .0112 0.396 .0208 0.307Sex of household head (female = 1) 5,669 0.596 4,733 0.774Education of most highly educated woman 1,461 0.373 4,021 0.121Education of most highly educated man �759 0.533 �2,383 0.201Number of generations family in village �9,270 0.001 �6,434 0.149Number of blood relatives in village 724 0.082 1,512 0.018Is household member village chief or asst.? (0/1) 16,151 0.217 38,816 0.058Are any females civil servants? (0/1) �11,055 0.638 �58,571 0.149Are any males civil servants? (0/1) �1,509 0.926 15,146 0.549Number of females, age 22–39 31,970 0.000 27,311 0.031Number of females, age 40–59 13,855 0.091 21,354 0.092Number of females, age 60 and over �14,731 0.134 �18,903 0.212Number of males, age 22–39 �1,538 0.838 10,760 0.352Number of males, age 40–59 �943 0.914 �5671 0.674Number of males, age 60 and over 9,727 0.393 894 0.960Constant �7,972 0.581 �54,363 0.018

Number of obs = 181 Number of obs = 181chi2(18) = 160.59 chi2(18) = 149.21

Prob > chi2 = 0.0000 Prob > chi2 = 0.0000Pseudo R2 = 0.0387 Pseudo R2 = 0.0359

Test rank-and-file = comm.:


F(1, 163) = 13.64 F(1, 163) = 44.63Prob > F = 0.0003 Prob > F = 0.0000

Table 9. Cumulative borrowing from village bank internal account (in baht-months)

Independent variable Dependent variable:internal account


Dependent variable:internal account



Months as rank-and-file VB member 3,397 0.000 4,236 0.000Months as VB committee member 5,294 0.000 8,704 0.000Value female-owned land five years ago �.000310 0.973 �.0129 0.389Value male-owned land five years ago .00732 0.802 �.0127 0.788Sex of household head (female = 1) �10,497 0.652 �9,450 0.802Education of most highly educated woman �4,230 0.238 �9,98 0.864Education of most highly educated man 2,355 0.370 2,427 0.566Number of generations family in village �820 0.896 �4,236 0.674Number of blood relatives in village 585 0.513 1,918 0.184Is household member village chief or asst.? (0/1) �46,852 0.103 �16,142 0.728Are any females civil servants? (0/1) 102,730 0.051 15,516 0.862Are any males civil servants? (0/1) �27,295 0.445 �19,705 0.732Number of females, age 22–39 13,811 0.434 18,333 0.518Number of females, age 40–59 9,892 0.576 21,287 0.454Number of females, age 60 and over �11,929 0.575 2,394 0.944



Table 9—continued

Independent variable Dependent variable:internal account borrowing

according to the records

Dependent variable:internal account borrowing,adjusted for use of multiple

names


Number of males, age 22–39 �17,856 0.276 909 0.972Number of males, age 40–59 �7,635 0.690 �6,403 0.836Number of males, age 60 and over �2,471 0.920 �12,858 0.747Constant �58,189 0.072 �128,615 0.014





F(1, 163) = 12.11 F(1, 163) = 25.77Prob > F = 0.0006 Prob > F = 0.0000

Table 10. Cumulative borrowing from village bank external and internal accounts combined (in baht-months)

Independent variable Dependent variable:external and internalaccount borrowing

according to therecords

Dependent variable:external and internalaccount borrowing,

adjusted for use of multiplenames


Months as rank-and-file VB member 5,063 0.000 5,960 0.000Months as VB committee member 7,814 0.000 12,982 0.000Value female-owned land five years ago �.000535 0.961 �.0198 0.298Value male-owned land five years ago .0281 0.405 .0192 0.744Sex of household head (female = 1) �4,847 0.859 �11,274 0.811Education of most highly educated woman �3,626 0.387 2,829 0.702Education of most highly educated man 1,538 0.621 �15.9 0.998Number of generations family in village �9,420 0.198 �9,209 0.468Number of blood relatives in village 1,196 0.259 3498 0.057Is household member village chief or asst.? (0/1) �26,790 0.421 18,052 0.759Are any females civil servants? (0/1) 105,358 0.080 �38,958 0.732Are any males civil servants? (0/1) �28,641 0.492 �6,312 0.931Number of females, age 22–39 38,282 0.067 44,018 0.221Number of females, age 40–59 18,433 0.376 40,069 0.266Number of females, age 60 and over �27,297 0.278 �18,338 0.673Number of males, age 22–39 �16,274 0.397 14,132 0.670Number of males, age 40–59 �10,425 0.641 �19,048 0.623Number of males, age 60 and over 4,423 0.879 �25,733 0.610Constant �34,634 0.344 �144,079 0.028





F(1, 163) = 18.03 F(1, 163) = 39.32Prob > F = 0.0000 Prob > F = 0.0000


Table 11. Sample regression: impact of village bank on household nonland asset value

Independent variable Coefficient p-Value

Months as rank-and-file member �367 0.554Months as committee member 3,187 0.018Does hh have a village bank member? (0/1) �5,885 0.760Does hh have a VB committee member? (0/1) 12,987 0.668Female-owned land value five years ago (baht) .0439 0.000Male-owned land value five years ago (baht) .0921 0.000Sex of household head (female = 1) 349 0.988Education of highest educated female (yrs) 3,437 0.311Education of highest educated male (yrs) 4,127 0.123Number generations family in village �7,678 0.247Number relatives in village 1,868 0.068Is household member village chief or asst? (0/1) 77,053 0.020Is female in hh a civil servant? (0/1) 471,308 0.000Is male in hh a civil servant? (0/1) 260,787 0.000Number of females, age 22–39 8,372 0.593Number of females, age 40–59 15,275 0.418Number of females, age 60 and over 17,968 0.360Number of males, age 22–39 �10,379 0.491Number of males, age 40–59 23,382 0.236Number of males, age 60 and over 19,040 0.415Village FA 1,896 0.955Village FB �26,836 0.475Village FC �20,317 0.557Village FD �53,542 0.143Village FE �20,083 0.577Village FF �9,731 0.783Village RA �12,073 0.778Village RB �30,246 0.439Village RC �87,836 0.031Village RD 17,076 0.652Village RF �49,209 0.190Village RG �25,185 0.480Village RH �11,448 0.755Constant 97,358 0.012

Number of obs = 444F(33, 410) = 14.25Prob > F = 0.0000R-squared = 0.5342

Adj R-squared = 0.4967

Table 12. Impact of village bank on household outcomes

Dependent variable(measured in bahtunless stated otherwise)


Coefficient onmonths as

rank-and-fileVB mbr (d1)

p-Value Coefficient onmonths ascommitteembr (d2)

p-Value F-testthat d1 = d2

p-Value

Physical assets

Household wealth �432 .528 3,010 .043 5.73 .017Women’s wealth �494 .265 1,705 .076 5.58 .019


Table 12—continued







p-Value

Men’s wealth 212 .684 1,382 .219 1.15 .284Household land value �96.1 .322 �20.8 .921 0.14 .712Women’s land value (T) 31.1 .725 263 .167 1.61 .205Men’s land value (T) �68.1 .532 �170 .453 0.21 .647Household nonland assets �367 .554 3,187 .018 7.44 .007Women’s nonland assets �535 .183 1,300 .135 4.74 .030Men’s nonland assets 362 .504 2,122 .071 2.40 .122Household productive assets �20.2 .935 981 .066 3.74 .054Women’s productive assets (T) 171 .326 870 .014 4.26 .040Men’s productive assets (T) �153 .578 351 .540 0.82 .367Household nonland farm assets 52.9 .543 307 .104 1.93 .165Women’s nonland farm

assets (T)88.9 .353 329 .100 1.58 .210

Men’s nonland farm assets (T) �7.22 .945 142 .512 0.50 .480Household livestock 56.4 .357 514 .000 12.7 .000Women’s livestock (T) 111 .151 187 .242 0.25 .621Men’s livestock (T) �5.31 .953 355 .061 3.88 .049Household business assets (T) �506 .319 872 .372 2.14 .144Women’s business assets (T) 50.0 .899 724 .231 1.46 .228Men’s business assets (T) �1,045 .245 1,044 .571 1.32 .251Household consumer durables �232 .341 437 .407 1.71 .191Women’s consumer

durables (T)�114 .477 755 .026 7.05 .008

Men’s consumer durables (T) �6.04 .985 �356 .599 0.28 .598House value 58.4 .882 1,220 .151 1.98 .160

Savings, debt, lending

Household savings(cash, bank deposits, etc.)

�172 .225 551 .073 5.91 .016

Women’s savings (T) �80.8 .215 217 .118 4.93 .027Men’s savings (T) �79.2 .590 425 .161 2.96 .086Household low-interest debt

(interest rate 6 2%/month) (T)244 .560 1,782 .036 3.54 .061

Women’s low-interest debt (T) �290 .445 256 .736 0.57 .451Men’s low-interest debt (T) 461 .299 2012 .022 3.34 .069Household high-interest debt

(interest rate > 2%/month) (T)218 .388 �254 .644 0.77 .380

Women’s high-interest debt (T) 373 .140 402 .445 0.00 .956Men’s high-interest debt (T) �91.3 .805 NA NA NA NAHousehold lending out at

positive interest (T)756 .172 2,653 .005 4.49 .035

Women’s lending out atpositive interest (T)

831 .198 2,093 .054 1.54 .216

Production, sales, expenses, labor

Household self-employmentproduction (sales and ownconsumption)

�6,319 .222 �3,495 .754 0.07 .795

Women’s self-employmentsales (T)

�380 .423 2,245 .023 7.51 .006









p-Value

Men’s self-employmentsales (T)

�11,023 .106 2,109 .886 0.83 .362

Household agriculturalproduction

24.2 .746 304 .061 3.15 .077

Women’s agricultural sales (T) 104 .279 206 .285 0.30 .587Men’s agricultural sales (T) �153 .145 200 .352 2.81 .094Household animal production

(sales and own consumption)�65.2 .106 156 .073 6.84 .009

Women’s animal sales (T) �58.8 .255 229 .031 7.97 .005Men’s animal sales (T) �17.2 .827 30.2 .879 0.06 .808Household business sales (T) �16,841 .061 4,101 .819 1.46 .228Women’s business sales (T) �640 .574 3,691 .079 4.73 .030Men’s business sales (T) �20,886 .121 2,635 .924 0.77 .382Household self-employment

expenses (purchase of inputs)�5,813 .233 �4,709 .655 0.01 .914

Women’s self-employmentexpenses (T)

�562 .139 1,520 .058 7.23 .007

Men’s self-employmentexpenses (T)

�9,986 .077 �9,368 .436 0.00 .958

Household farming expenses(purchase of inputs)

�41.3 .314 92.7 .295 2.42 .120

Women’s farming expenses (T) 9.86 .792 106 .166 1.70 .193Men’s farming expenses (T) �84.9 .086 5.78 .956 0.77 .381Household animal-raising

expenses (purchase of inputs)�26.1 .456 56.7 .453 1.27 .260

Women’s animal-raisingexpenses (T)

�12.5 .704 �48.4 .475 0.30 .583

Men’s animal-raisingexpenses (T)

�42.7 .309 106 .221 3.09 .080

Household business expenses(purchase of inputs) (T)

�15,030 .082 4,191 .808 1.34 .247

Women’s business expenses (T) �655 .515 3,209 .082 4.89 .028Men’s business expenses (T) �19,797 .134 3,631 .892 0.81 .369Household self-employment

labor hours�4.60 .481 26.9 .058 5.25 .022

Women’s self-employmentlabor hours

�1.94 .633 14.0 .112 3.47 .063

Men’s self-employmentlabor hours

�2.67 .526 12.9 .156 3.11 .079

Health care, education

Household medicalexpenses (T)

�43.7 .105 �47.6 .420 0.00 .947

Medical expenses madefor women (T)

�13.3 .436 �46.0 .229 0.77 .382

Medical expenses madefor men (T)

�23.5 .465 �66.9 .356 0.38 .541

Medical expenses madefor children (T)

6.09 .526 �10.7 .635 0.58 .445










p-Value

Medical expenses madefor girls (T)

�1.08 .918 8.38 .736 0.15 .694

Medical expenses madefor boys (T)

12.2 .267 �26.7 .349 1.88 .171

School expenses made forchildren in household

5.39 .771 86.6 .035 4.29 .039

School expenses madefor girls

�1.47 .929 2.79 .932 0.02 .889

School expenses madefor boys

6.80 .688 127 .002 10.14 .002

(T) indicates that tobitregression is used

NA = not applicable


Date post:	01-Dec-2021
Category:	Documents
Upload:	others
View:	4 times
Download:	0 times

Microfinance in Northeast Thailand

Documents