The socio-economic determinants of microfinance participation : an analysis of married women and unwed female household heads in Karnataka, India

(1)

THE SOCIO-ECONOMIC DETERMINANTS OF

MICROFINANCE PARTICIPATION:

AN ANALYSIS OF MARRIED WOMEN AND UNWED FEMALE

HOUSEHOLD HEADS IN KARNATAKA, INDIA

Nadja van ’t Hoff

June 26, 2018

Bachelor’s thesis

University of Amsterdam

Faculty of Economics and Business

Econometrics and Operational Research, 2017 - 2018

Student number: 11030720 Supervisor: S. Stephan

(2)

Statement of originality

This document is written by Student Nadja van ’t Hoff who declares to take full respon-sibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in

creating it.

The Faculty of Economics and Business is responsible solely for the supervision of

(3)

Abstract

The present study investigates the factors that influence participation of married women

and unwed female household heads in the Bharatha Swamukti Samsthe (BSS) microfi-nance programme in rural Karnataka, India. Women should be targeted more by

micro-finance institutions, because women who participate in micromicro-finance services are found to increase the family well-being, to yield a higher productivity in their business and the

participation is found to empower women in gender-biased cultures (Pitt & Khandker, 1998; Pitt, Khandker Cartwright, 2006; cited in Murshid, 2018, p. 1; Pitt, Khandker, &

Cartwright, 2003). Previous studies found different social and economic factors to in-fluence the microfinance participation decision. The results of a probit model reveal that

age and a quadratic term of age, the level of education, the caste and the household size are social factors that influence the microfinance participation probability. Economic

factors that influence this probability are savings and self-help groups. The employment status is not of significance. Effects of savings and participation in self-help groups

differ between married women and female household heads. Savings positively influ-ence married women, whereas this effect is negative for female heads. Participation in

self-help groups yields a greater positive effect on female heads compared to married women whose husband does not experience a demonstration effect. This indicates that

female spouses are influenced by their husbands. On the other hand, unwed female household heads seem to lack the support of a husband and appear to be constrained

by insecurity resulting from discrimination. The results further suggest that poor female heads use microfinance services as a last resort to escape from poverty. This implies

that unwed female household heads are a more vulnerable group that could especially benefit from microfinance services.

(4)

1 Introduction

Microfinance seems to be the costless solution to poverty. Microfinance institutions provide financial services to the poor who were rejected for financial services by banks.

Poverty is reduced and at the same time the institutions gain profit. To increase the success of this win-win situation, the United Nations announced 2005 to be the

“Inter-national Year of Microcredit”. However, only five years later, the BBC publishes an article entitled “India’s micro-finance suicide epidemic” where microfinance in India is

compared to the 2008 subprime mortgage crisis (Biswas, 2010). The title implies the possible consequences of offering the wrong and complicated financial products to poor

and mostly uneducated people.

In the past years, many researchers, amongst whom Eneji, Ifeoma, Umejiakwu and

Ushie (2013) as well as Rasheed, Xia, Ishaq, Mukhtar and Waseem (2016) supported neither of these two extreme opinions about microfinance. Instead, they find that

micro-finance has the potential for poverty reduction, but that changes in the offered services as well as the institutions are necessary to deal with the current challenges.

A long-standing challenge appears to be the take-up rate of microfinance products, which is below expectations (Eneji et al., 2013; Maazullah, 2016). Armend´ariz and

Mordurch (2005) referred to India as “possibly world’s largest untapped market for microfinance”. The demand for microfinance services is found to depend on

economic variables (Rasheed et al., 2016). Hence, exact information about the socio-economic characteristics of people who are currently enrolled in these projects is

impec-cable to increase the participation in microfinance projects. This information is crucial to efficiently address the target group and to adapt financial products and policies to the

participants.

The importance of knowledge of the characteristics of participants in microfinance

projects cannot be emphasized enough. Social and economic characteristics are found to influence the participation decision (Bhoj, Bardhan, & Kumar, 2013; Anjugam &

(6)

A characteristic that is very often found to be significant is the gender of participants (Olateju et al., 2018; Rasheed et al., 2016; Eneji et al., 2013). According to Bardasi,

Sabarwal and Terrell (2011), women participate less often than men in Asian countries. It is of importance to raise the microfinance take-up rate of women due to multiple

reasons. First of all, women account for a majority of the poor and though female house-hold heads are said to belong to the poorest of the poor, not much empirical research

has been done on this assumption (Chant, 2003). These unwed female household heads are particularly challenged, since they culturally and socially belong to a disadvantaged

group. Instead of being culturally or socially challenged, married women are found to be challenged on the household level due to their husband who might financially

con-strain them or influence their participation decision (Vonderlack-Navarro, 2010). Mi-crofinance is found to significantly empower these women (Pitt et al., 2003). On top of

that, women are found to be more trustworthy and productive lenders than men and they also invest in the education and nourishment of their families (Pitt & Khandker, 1998;

Pitt, Khandker & Cartwright, 2006; cited in Murshid, 2018, p. 1). Altogether, female heads and married women should be targeted more efficiently, because women are less

likely to participate as opposed to men, women represent a larger part of the poor, their participation increases the family well-being and they are empowered through

micro-finance. Hence, the factors that trigger the microfinance participation of female heads and married women need to be analyzed.

In this article, the socio-economic determinants that influence the participation in microfinance projects of unwed female household heads and married women in rural

southern India are investigated.

To investigate this issue, a dataset is used that was initially assembled to investigate

the diffusion of microfinance (Banerjee, Chandrasekhar, Duflo, & Jackson, 2013). It contains a variety of socio-economic variables on individuals as well as households from

rural Karnataka, India. For a thorough estimation of the relationship between the par-ticipation decision and the potential determinants, different hypothesis are established

(7)

and model that seem to capture the determinants of participation best is selected and the results are interpreted.

For this investigation, results from literature are compared and possible determi-nants of microfinance participation are identified and used to establish three hypotheses

in section 2. In section 3, the dataset is introduced and the methods and content of the research is amplified with a detailed comparison of the models encompassing the

hypotheses. The results of the preferred model are recorded in section 4 and analyzed in section 5. Finally, a conclusion about the determinants is drawn, the limitation of

the study is stated and a recommendation for changes and further research is given in section 6.

2 Literature Review

2.1 Demand side barriers

An insight into the characteristics of women who decide to engage in microfinance is

needed. This can allow for policies to adapt to the applicants and for the take-up rate to be increased, enabling the benefits of microfinance to spread. One of the main

ben-efits of microfinance is the empowerment of women (Armend´ariz & Mordurch, 2005). Through microfinance programmes, women can be empowered financially, culturally

and socially (Bhoj et al., 2013). Moreover, approving microloans to female applicants leads to a higher economic and social improvement in comparison with loan approval

to male applicants (Khandker, 2003; cited in Armend´ariz & Mordurch, 2005, p. 190). Hence, a lot of organizations and institutions specifically target women, like for example

the Al-Amal Islamic Bank where women account for 60 per cent of its services (Abhijit, Arun, Esther Mathew, 2012; cited in Eneji et al., 2013, p. 2).

For the past decades, multiple actions to enhance the access to microfinance pro-grammes have been undertaken, but the attempt often failed because of deficient policies

(8)

an-ticipated level which can be due to barriers on the supply side as well as on the demand side (Eneji et al., 2013; Maazullah, 2016). Bardasi et al. (2011) recommend taking a

closer look at the participants on the demand side, since they equally do not encounter strong evidence for barriers on the supply side. In line with this finding and since the

BSS microfinance programme is assumed to be equally offered to all poor, there barely are constraints on the supply side and thus the supply is assumed to be elastic in this

study. Hence, the focus of this paper lies on the demand side.

On the demand side, poor women are excluded for different reasons. Agier and

Sza-farz (2013) find that even when the chances of obtaining microcredit are equal, a gap between genders exists and that this gap increases when the asymmetric information

declines. This indicates that women exclude themselves from microfinance projects be-cause of personal constraints like risk adversity and insecurity. Bardasi et al. (2011)

encounter that men participate in microfinance projects more often than women in East-ern Europe and Central Asia, because women are subject to environmental constraints

next to personal constraints. According to many researchers, social and economic con-straints influence the women’s microfinance decision (Bhoj et al., 2013; Anjugam &

Ramasamy, 2007; Kifle et al., 2013; Chant, 2003; Brett, 2006).

2.2 Social determinants

The human capital theory describes how the ability for economic productivity and thus successful entrepreneurship depends on factors like education, experience and social

aspects. A higher human capital enables entrepreneurs to recognize and consequently engage in business opportunities (Ucbasaran, Westhead, & Wright, 2008). This

illus-trates that factors concerning the individual characteristics influence the participation decision. Bardasi et al. (2011) mainly investigate the enterprise attributes and strongly

recommend taking a look at personal, social and non-economic characteristics.

A main factor affecting human capital is education. Many surveys find a positive

effect of education on the probability of participation in microfinance services (Rasheed et al., 2016; Eneji et al., 2013; Omonona, Lawal, & Oyinlana, 2010; Bhoj et al., 2013).

(9)

Bhoj et al. (2013) encounter that education has a significant positive effect on the parti-cipation probability in dairy microfinance self-help groups in Uttarakhand State of India.

A woman with one unit level of education more is 3 times more likely to join a dairy self-help group. Rasheed et al. (2016) find that the demand for microfinance programmes

of Pakistani farmers increases with 0.74 percentage points if education status increases with 1 per cent. Most studies are based on samples of microfinance entrepreneurs that

dispose over no or little education. On the other hand, Kangogo (2013) uses a sample in which 97.7 per cent of the participants in Kenya have had formal education. He finds

that taking part in microcredit groups negatively depends on education. For every addi-tional year of education, the probability of joining the groups declines with 8.7 per cent.

He assumes that well-educated people are more likely to have paid jobs and thus do not need financial aid. Olateju et al. (2018) come to the same conclusion with a highly

significant, negatively-signed coefficient for education. These findings suggest the exis-tence of a nonlinear relationship between education and the microfinance participation

probability.

Experience and thus human capital also can increase with age. However, Kangogo

(2013) finds that age has a negative effect on the participation probability in micro-credit groups in Kenya. Anjugam and Ramasamy (2007) examine the determinants of

women’s participation in microfinance self-help groups. They support that age has a significant negative effect on the participation decision. If age rises by one year, the

probability of participation will decline with 0.6 per cent. Older participants proba-bly dispose of less energy and motivation than younger ones (Shah, Bukhari, Hashmi,

Anwer, & Anwar, 2008). They also could be more likely to be discriminated against. However, Bhoj et al. (2013) observe the reverse effect. They find that age has a

signi-ficant positive effect and that a rise in age by one year makes participation 1.15 times more likely. In line with Bhoj et al. (2013), Maazullah (2016) finds a higher take-up rate

of the Akhuwat loan in Pakistan when age increases. He further states that application for the loan is a nonlinear function of age. This implies that the effect of age declines

(10)

programme in Nigeria into two subsamples containing the poor and the less poor. They find a positive effect of age on participation for the poor and a negative effect for the

less poor.

Next to factors like education and experience that are directly related to human

ca-pital, there are other social indicators that can influence entrepreneurship of an indi-vidual like the marital status, the caste and family size. Oletaju et al. (2018) find that

married people are more likely to participate than non-married people, but this finding is not significant for the subsample with the less poor.

Anjugam and Ramasay (2007) use the caste of an individual as an indication for the social backwardness. 45 per cent of the participants in their sample are scheduled caste

members and thus belong to the most disadvantaged caste in India. Surprisingly, they observe that the greater the social backwardness, the higher the probability of

partici-pation. They explain this by the low income and assets of the individuals in the lower caste which lead to a need for microfinance services. On the other hand, Imai, Arun and

Annim (2010) find that individuals that neither belong to the scheduled caste nor to the scheduled tribe are more likely to participate in microfinance projects. They suppose

that this finding can be based on supply side constraints.

Kifle et al. (2013) do not find an effect of age, education or religion on the

par-ticipation decision of women in Ethiopia. They do find that the household size has a significant negative effect. An additional family member decreases the probability by

1.76 percentage points. This can be due to the fact that larger households generate more household income, since it has more members, and thus an enhanced

involve-ment in formal bank services (Kifle et al., 2013). Oletaju et al. (2018) support that the household size is highly significant for the poor subsample and they observe that

the larger the household, the smaller is the probability of participation. Unlike Kifle et al. (2013), Kangogo (2013) finds a significant positive effect of the household size on

the participation in microfinance activities. He concludes that the probability of parti-cipation increases by 7.7 per cent when the household increases with one member. He

(11)

that comes with an additional family member. Rasheed et al. (2016) support the finding that a household containing more family members has a higher probability to engage in

microfinance programmes.

Altogether, an individual’s participation decision is often found to depend on social

variables like education, age, marital status, caste and household size. Though variables are observed to yield different effects in literature, they often are encountered to be

significant.

2.3 Constraints on a household level

Individual characteristics are often found to be of an influence on the participation de-cision. At the same time, microfinance programmes are regularly accused of targeting

women based on individual aspects instead of taking into account their household si-tuation and sisi-tuation in society, like the position of the man as primary wage earner

of the household in conventional societies (Vonderlack-Navarro, 2010). Chant (2003) states that women cannot be saved from poverty when no attention is paid to gender

relations. Though the husband of a married woman can support participation in microfi-nance, male partners can also be a constraining factor (Vonderlack-Navarro, 2010). This

can occur when they feel attacked in their position as the breadwinner (Armend´ariz & Mordurch, 2005; Vonderlack-Navarro, 2010). In a majority of the cases , the use of

ap-proved microloans is determined by the male members of a household (Pitt et al., 2003). Eneji et al. (2013) find that the sex of the household head is decisive for the

participa-tion decision, since most of the loans go to the household heads of which 85 per cent is male. Even when the female spouses determine the use and invest in a business, they are

likely to be unable to earn enough money from their business, and thus rely on financial support of their husbands (Brett, 2006). However, Vonderlack (2010) examines a

micro-finance institution that targets unmarried female household heads, but encounters that most participants are married. This indicates that female household heads might be even

more constrained as opposed to married women or they might be less willing to incur debt given their particular vulnerability. Eneji et al. (2013) affirms that being a female

(12)

household head negatively effects the appliance for a loan, especially in a single-parent household. 47 per cent of the households with a male leader participate in microfinance

against 35 per cent of female-headed households in the sample used by Bekele (2010). The percentage of female household heads in India increased from 9.2 per cent

in 1993 to 14.4 per cent in 2006 (The Worldbank, Demographic and Health Surveys, 2018). In 2001, 10 per cent of the households in India are headed by women, of which

6.6 per cent are widows (Census Commissioner of India, 2001; cited in Chudgar, 2011, p. 552). In India, early widowhood leads to poverty in most cases (Buvini´c & Gupta,

1997). Chudgar (2011) states that the widow-headed households belong to the poor-est of the poor and according to Chant (2003), the terms ”feminisation of poverty” and

”feminisation of household headship” are used interchangeably. This motivates a lot of institutions to target female-headed households, because these households are

con-sidered a proxy for poverty (Buvini´c & Gupta, 1997). Buvini`c and Gupta (1997) state that if female-headed households actually belong to the poorest, then they should be

targeted. However, they find that when the household size is controlled for, female household heads are not poorer than other households regarding the per capita income,

since these households on average contain less members. This indicates that female heads make at least as considerate financial decisions as male household heads.

Next to financial constraints, female heads are further constrained by time and flexi-bility, because they have to carry out domestic tasks next to generating income (Buvini´c

& Gupta, 1997). Female heads also can exclude themselves on purpose from micro-finance services instead of being constrained by environmental factors. Corsi and De

Angelis (2017) found that women who are married receive a higher loan disbursement as opposed to unwed women who are presumably discriminated against. On top of that,

De Mel, McKenzie and Woodruff (2008) state that women are less likely to participate in microfinance, because they believe to be rejected anyhow. This belief might be more

present amongst female household heads who are often discriminated against and thus more likely to be rejected by formal institutions than married women. This indicates

(13)

par-ticipation rate might be a question of a self-fulfilling prophecy in which female heads do not participate due to the fear of rejection.

All in all, married women seem to face different constraints than female household heads. While married women are mostly constrained by their husbands, female

house-hold heads especially have to deal with economic constraints and personal barriers.

2.4 Economic determinants

According to the theory of diminishing returns to capital, the same amount of money invested in capital yields higher returns for a poor entrepreneur than for a richer one.

This suggests that low income households have a high incentive to participate in micro-finance programmes, even in presence of the high interest rates. However, the poorest

households might need to secure basic demands like food and health before investing in a business (Eneji et al., 2013). Moreover, the poor probably do not want to invest due

to high risks that come with owning a business, because a lot of new businesses are un-successful (Hashemi & Rosenberg, 2006). Thus although the microfinance institutions

aim to reach the poorest, it seems to be mainly the less poor that participate in microfi-nance programmes, making previous income an important determinant of microfimicrofi-nance

participation (Hashemi & Rosenberg, 2006).

While Maazullah (2016) finds that the monthly household income and the monthly

business profits are insignificant, Sekyi, Nkegbe and Kuunibe (2014) find that small scale entrepreneurs in Ghana are more likely to apply for microcredit when their

in-come increases. In contrast to these findings, Olateju et al. (2018) find a highly signif-icant, negative coefficient for the income level. They state that an increase in income

lowers the probability of participation in the Cowries Microfinance Bank programme in Nigeria. This coincides with the findings of Rasheed et al. (2016) and Magboul and

Hassan (2016) who equally encounter a decline in microfinance demand when the in-come level increases. Nevertheless, a negative coefficient of inin-come can only hold up to

a certain limit of wealth, because wealthy households won’t be rejected by formal banks that charge lower interest rates than microfinance institutions. Thus the coefficient of

(14)

income may yield different results dependent on the level of poverty in the data sam-ple. Since female heads mostly belong to the poorest of the poor, they probably have a

lower demand for microfinance compared to married women. Due to their vulnerability, female heads also might be more risk-adverse and less likely to incur debt.

Rather than the amount of income, Anggraeni (2009) finds that a regular income source is significant and that it increases the possibility of participation. Farmers for

example have an irregular income due to crop seasons and cannot afford long peri-ods of high interest rates, whereas individuals in commercial activities probably can

generate lucrative opportunities (Eneji et al., 2013). This indicates that regular income-generating activities increase the microfinance demand. This can be due to the fact

that microfinance businesses are flexible and thus easily combined with other house-hold tasks (Poster and Salime, 2002; cited in Vonderlack-Navarro, 2010, p. 123). At

the same time, female heads are constrained by time more than married women, be-cause they are also involved in income-generating activities (Buvini´c & Gupta, 1997).

Thus working female heads might be less likely to apply for microfinance than married women who do not work due to a lack of time.

Brett (2006) encounters that a lot of participants engage in microfinance to save money and they intend to leave the programme if they have sufficient savings. On the

other hand, Kifle et al. (2013) state that the probability of a woman’s participation in-creases by 1.804 percentage points if she has monthly savings. Savings are supposed

to release women from dependency on family members and thus can lead to autonomy (Vonderlack & Schreiner, 2002). Nevertheless, according to Deere, Oduro,

Swami-nathan and Doss (2013), only 8 per cent of the women are allowed to decide over the household’s wealth in rural Karnataka. They further find that in India, men are in control

of the household possessions. This implies that married women are not used to mak-ing considerate financial decisions and thus are less likely to save as opposed to female

heads who are in charge of the household resources.

Next to income and savings, participation in small microfinance programmes like

(15)

demon-stration effect, an individual is more likely to participate in microfinance programmes if he becomes accustomed to using financial services (Anjugam & Ramasamy, 2007).

Anjugam and Ramasamy (2007) find that women who participate in self-help groups are more likely to join other microfinance programmes. This is in contrast with the

findings of Bhoj et al. (2013). They state that only a small proportion of the poor join microfinance self-help groups for this reason. Though the demonstration effect might

not affect married women who are under the control of their husband, participation in self-help groups can lead to an improvement on the household level. According to Pitt

et al. (2003), married women can increase their input in household decisions and their financial access by joining self-help groups. On the other hand, Kabeer (2005) only

finds a small effect of participation in self-help groups on decision-making processes. Altogether, the findings on economic determinants of microfinance participation

dif-fer in significance and sign. Furthermore, these determinants can yield difdif-ferent effects for married women as opposed to female heads.

2.5 Hypotheses

In this section, the findings from literature are summarized and three mutually exclusive

hypotheses are established based on these findings.

As stated before, knowledge of the demand side determinants that restrain women

to participate in microfinance programmes is needed to increase the current take-up rate. A variety of social and economic variables are found to influence the microfinance

decision, but very different results about the sign of their coefficients are observed at the same time. A higher human capital is supposed to increase the microfinance

de-mand and in theory, a poorer entrepreneur is more motivated to engage in microfinance projects. Yet both theories do not always seem to hold in practice. When estimating the

determinants of microfinance participation, it is of importance to include social factors like age, education, caste and household size.

Being a married woman or a female household head might influence the microfi-nance participation decision. Poor married individuals and male-headed households are

(16)

found to be more likely to participate, indicating that married women are more likely to apply for microfinance programmes. At the same time, the microfinance

participa-tion of female spouses might be constrained by their husbands. Nevertheless, female household heads might suffer from greater constraints as opposed to female spouses.

These constraints are mostly of financial nature and female heads are even used as a proxy for poverty by microfinance institutions. Furthermore, the less poor participate

more often than the poor, indicating that married women might be more likely to take up microfinance than female heads. Married women also might be more likely to

en-gage in microfinance activities because of the risk adversity and insecurity of female household heads who form a more vulnerable group. This leads to different financial

decision-making between the group of married women and the group of female heads. A distinction between these two groups is rarely made in literature and could form an

explanation for the differing results amongst papers.

While there is no inducement from the literature review that the effect of the

so-cial variables differ between married women and female heads, the economic variables might yield a different effect for these two groups. Female heads could have a higher

incentive to participate in microfinance programmes based on whether or not they main-tain savings, whereas married women are not used to making considerate financial

de-cisions and are thus less accustomed to saving activities. Moreover, the demonstration effect obtained through self-help groups probably has a stronger effect on female heads,

since married women are likely to need approval of their husbands who haven’t expe-rienced this effect. Nevertheless, self-help groups are expected to increase the demand

of married women as well, since these groups can increase their input on household decisions. Working female heads are suspected to have a lower positive demand than

married women, whereas non-working married women could have time as well as a regular income through their husband, indicating the possibility of a higher demand due

to less risk.

The aim of this study is to find the social and economic determinants of the

(17)

findings on the signs in combination with the different interpretations of the factors influencing microfinance participation increase the difficulty of estimating the

determi-nants and induce different possibilities. Based on the above observations, the following three mutually exclusive hypotheses can be formulated:

(1) The effects of social and economic determinants on the microfinance participation decision do not differ significantly for female heads and married women.

(2) Married women are on average more likely to take up microfinance as opposed to female household heads, when social and economic determinants are controlled

for.

(3) The economic variables comprising self-help groups, savings and employment

status have a different effect on the participation probability of female heads as opposed to married women, when social determinants are controlled for.

In the following sections, these three hypotheses will be further investigated.

3 Methodology

3.1 Data

To determine the factors that influence the microfinance participation decision, sec-ondary data is used. The data used in this study was originally used by Banerjee et

al. (2013) to examine the diffusion of microfinance. It was gathered through a survey that was held between 2006 and 2011 in multiple villages in rural Karnataka, India.

These villages only had little contact with microfinance programmes up to to that point (Banerjee et al., 2013). The microfinance institution Bharatha Swamukti Samsthe

en-tered the villages in 2006 and offered households with female members between the ages of 18 and 57 a group lending programme. The participation decision of 49 out

of the 75 initially aimed villages is recorded. The data further contains household and individual characteristics. The individual respondents were randomly selected based

(18)

on stratification of religion and location and the questionnaire was taken by eligible households (Banerjee et al., 2013).

For this research, the households that didn’t complete the questionnaire as well as the observations without reported participation decision are dropped from the data, leading

to 4714 remaining household observations. A respondent is defined as household head if this person is the only head in the household and a respondent is defined married if the

household contains exactly one household head and one spouse of the household head. Of every household, the observation of one individual is used who either is a married

woman or a female household head. Of the female household heads 97.28 per cent is unmarried. The eleven, seemingly erroneous, observations of married female household

heads are dropped. Finally, 91.94 per cent of the 4714 household observations is used for the investigation, consisting of 394 observations on single female household heads

and 3940 observations on female spouses.

3.2 Model specification

In this study, the determinants of the microfinance participation decision are estimated

based on three hypotheses. To test these hypotheses, three different models are esti-mated, each of which applies to one hypothesis. The first model does not account for

any difference between married women and female heads, but models the microfinance participation as a function of social and economic variables:

yi =β1+ β2x2i+ β3 x3i+ β4x4i+ β5 x5i+ β6x6i+ β7 x7i+ β8 x8i

+ β9x9i+ β10x10i+ β11x11i+ εi, (i = 1, . . . , n)

(1)

Where:

i = Household observation

yi = Participation in microfinance (one if household participates, zero otherwise)

x2i = Age of respondent (years)

x3i = Nonlinear term of age (age squared)

(19)

x5i = Unknown caste (one if caste is unknown or not recorded, zero otherwise)1

x6i = Scheduled caste (one if household belongs to the scheduled caste or tribe, zero

otherwise)1

x7i = OBC caste (one if household belongs to the OBC caste, zero otherwise)1

x8i = Household size (number of members)

x9i = Savings (one if household has savings, zero otherwise)

x10i = Self-help group (shg) participation (one if household participates, zero

other-wise)

x11i = Workflag (one if household’s respondent worked last week, zero otherwise)

The second model is based on hypothesis (2). To estimate the difference between married women and female heads, assuming that the effects of the variables in model

(1) are the same for both groups, we add a dummy to model (1):

+ β9x9i+ β10x10i+ β11x11i+ β12x12i+ εi, (i = 1, . . . , n)

(2)

Where:

x12i = Spouse (one if respondent is female spouse, zero if female household head)

The third hypothesis states that economic variables have a different effect on the participation decision of female heads compared to the decision of female spouses. By

including a dummy variable for married women and interaction terms of married women with the economic variables, a third model is established that accounts for differences

in intercept as well as slopes:

+ β9x9i+ β10x10i+ β11x11i+ β12x12i+ β13x12i× x9i

+ β14x12i× x10i+ β15x12i× x11i+ εi, (i = 1, . . . , n)

(3)

(20)

The dependent variable in all three of the models is a binary variable for the participa-tion decision and thus a probability model is used for estimaparticipa-tion. The linear probability

model P [yi = 1] = x0iβ is not suited because of multiple disadvantages. The error

terms εi are not normally distributed and the estimated probabilities ˆP [yi = 1] = x0ib

can get values smaller than zero or larger than one. Instead, a nonlinear probability model of the form P [yi = 1] = F (x0iβ) is used, like the probit or the logit model. Most

researchers use the probit model (Eneji et al., 2013; Bekele, 2010; Maazullah, 2016; Magboul & Hassan, 2016; Anjugam & Ramasamy, 2007), though the logit model is

widely spread as well according to Maazullah (2016). These two models are estimated by maximizing the likelihood. The main difference between these two models is the

cu-mulative distribution function F (x0β), but they are often found to yield similar results. Even in case a probit model is used where a logit model would have been more

appro-priate, the estimates still are reliable to a moderate degree. When regressing the three defined models with probit and logit, the likelihood of the estimated probit models is

marginally higher (see Appendix I, Table 1). Since the likelihood is slightly higher, the probit model is preferred. Hence, the probit model is applied in this study and not the

logit model. The cumulative distribution function of the standard normal distribution is applied in the probit model:

P [yi = 1] = Φ(x0iβ) = 1 √ 2π x0β Z −∞ e−12s 2 ds

The sign of most coefficients except for β1 can directly be interpreted. To interpret

the magnitude, the marginal effects of the explanatory variables are estimated. Because

the models contain dummy variables, the mean marginal effects are estimated instead of the marginal effects at means:

1 n n X i=1 ∂P [yi = 1] ∂xji = βj 1 n n X i=1 f (x0_iβ), j = 2, . . . , k Where the coefficient βj is multiplied by the scaling factor _n1

Pn i=1f (x

0 iβ).

(21)

4 Results

4.1 Descriptive statistics

Table 1 depicts the descriptive statistics of the social and economic variables used in

the three defined models. The table shows the frequency and corresponding percentage of these variables for the subsample of married women and the subsample of female

household heads. Out of 4334 observations, 394 of the respondents are female heads and 3940 are married women. Of the female spouses, 18.15 per cent take up microfinance

against 16.75 per cent of the female heads. This seems to support the hypothesis that female spouses are more likely to participate.

Regarding the social factors, almost 50 per cent of the women are between 31 and 45 years old in both groups. Yet only 5.55 per cent of the female heads are younger

than 30 years, whereas this percentage equals 29.75 per cent for the female spouses. On average, female heads and spouses are respectively 46.96 years and 37.94 years old.

This supports the assumption that widows are over-represented in the subsample of fe-male heads. Furthermore, 73.86 per cent of the fefe-male heads has no education against

51.50 per cent of the female spouses. The lower educational level of female heads can be based on the exclusion from society of these women. This also is supported by the

descriptive statistics on the caste. 30.96 per cent of the female spouses are in the sched-uled caste or tribe against 39.09 per cent of the female heads, supporting the suspicion

that female heads belong more often to a group that is discriminated against. The house-hold size of female heads and female spouses respectively consists of 1.75 members and

2.57 members on average. This supports the finding of Buvini`c and Gupta (1997) that households headed by women are smaller. There are two apparent explanations for this

observation. Firstly, female heads are divorced or never married and thus do not have children or do not have them anymore. Second, female heads are older widows whose

children have left the house.

Regarding the economic variables, 47.46 per cent of the female-headed households

(22)

indi-Table 1: Descriptive statistics of social and economic characteristics

Variables Female spouses Female heads

Frequency Percentage Frequency Percentage

Participation Yes 715 18.15% 66 16.75% No 3225 81.85% 328 83.25% Age ≤ 30 1172 29.75% 21 5.33% 31 – 45 1969 49.97% 188 47.72% 46 – 60 723 18.35% 145 36.80% ≥ 61 76 1.93% 40 10.15% Education* No education 2029 51.50% 291 73.86% Lower education 1321 33.53% 81 20.56% Higher education 590 14.97% 22 5.58% Caste Unknown 15 0.38% 1 0.25% Scheduled 1220 30.96% 154 39.09% OBC 2217 56.27% 211 53.55% General 488 12.39% 28 7.11% Household size ≤ 2 2723 69.11% 312 79.19% 3 – 4 1034 26.25% 71 18.02% ≥ 5 183 4.64% 11 2.79% Savings Yes 1804 45.79% 187 47.46% No 2136 54.21% 207 52.54% Self-help group Yes 1650 41.88% 149 37.82% No 2290 58.12% 245 62.18% Workflag Yes 1887 47.89% 241 61.17% No 2053 52.11% 153 38.83%

* Lower education contains all levels below the Secondary School Leaving Certificate (S.S.L.C.). Higher education encompasses the respondents with the S.S.L.C. or higher.

(23)

cates that female heads make at least as considerate financial decisions as households with a married woman. Moreover, 61.17 per cent of the female heads worked during the

last week against 47.89 per cent of the female spouses. Using workflag as a proxy for the employment status, this implies that female heads work more often than female spouses

who probably focus on taking care of household tasks instead of income-generating activities.

All in all, female spouses appear to take up microfinance more often than female heads. Female heads are likely to be widows and belong to a group that is discriminated

against. Moreover, they seem to be more likely to work compared to spouses.

4.2 Model performance and selection

In this section, the three models are compared and the model that seems to define the relation between the participation probability and the social and economic variables best

for the female heads and married women is selected for further analysis.

The variable education2was suspected to have a nonlinear effect on the participation probability, but this variable is not found to be significant in either of the models and thus is removed (see Appendix II, Table 1). Considering model (1) in Table 2, the average

household with a female head or spouse with no education or savings and who belongs to the OBC caste, who does not participate in self-help groups and does not work has

a probability of 12.89 per cent to participate in the BSS microfinance programme (see Appendix II, Table 2).

The output of the three models is given in Table 2. The LR test of joint significance of all explanatory variables indicates that the three models have explanatory power. The

coefficients of the social variables in model (1) and (2) are equal in sign and similar in magnitude. The same finding holds comparing the control variables of these two

models with those of model (3). However, the values of the economic variables as well as the sign of savings differ in model (3). For all three models it holds that all

variables except workflag, spouse and spouse×workflag are significant. At the same time, the McFadden’s R2of all three models is very low. The combination of significant

(24)

Table 2: Probit regression output when regressing on the participation decision

Model (1) Model (2) Model (3)

participation Coefficient S.E. Coefficient S.E. Coefficient S.E.

age – 0.039*** (0.013) – 0.038*** (0.013) – 0.038*** (0.013) age2 0.000** (0.015) 0.000** (0.015) 0.000** (0.015) education – 0.013* (0.007) – 0.013* (0.007) – 0.013* (0.007) unknown caste 0.661* (0.354) 0.661* (0.354) 0.652* (0.354) scheduled caste 0.587*** (0.091) 0.585*** (0.091) 0.587*** (0.091) OBC caste 0.330*** (0.086) 0.329*** (0.086) 0.329*** (0.086) household size 0.086*** (0.022) 0.090*** (0.023) 0.092*** (0.023) savings 0.191*** (0.065) 0.190*** (0.065) – 0.370* (0.223) shg 0.145** (0.065) 0.146** (0.065) 0.529** (0.225) workflag 0.069 (0.047) 0.068 (0.048) 0.155 (0.163) spouse – 0.038 (0.088) – 0.104 (0.159) spouse×savings 0.616*** (0.232) spouse×shg – 0.421* (0.234) spouse×workflag – 0.094 (0.169) constant – 0.721** (0.290) – 0.692** (0.298) – 0.635** (0.315) Observations 4334 4334 4334 Log likelihood – 1971.1616 – 1971.0681 – 1967.2995 LR chi2(df) 146.38 146.57 154.11 Prob > chi2 0.0000 0.0000 0.0000 McFadden’s R2 0.036 0.036 0.038 AIC 0.915 0.915 0.915 BIC – 32259.543 – 32251.356 – 32233.770

(25)

variables and low R2 is a common phenomenon when the model can estimate general relations in the data, but encounters problems estimating individual decisions.

For models (1) and (2), the Wald test cannot reject the hypothesis that workflag equals zero (see Appendix II, Table 4). It could be that the models perform better

without this variable. To compare models, information criteria can be used. In this paper, the Bayesian Information Criterion (BIC) is preferred to the Akaike Information

Criterion (AIC), because the BIC assumes that a correct model exists and punishes for extra regressors, while the AIC is in favor of correct prediction and thus punishes the

extension of the model less severe. Removing the variable workflag from the models decreases the BIC, but at the same time the maximized log likelihood declines (see

Appendix II, Table 3). Hence, this variable is not removed from the models.

Instead of using the Wald test, the LR test is preferred for testing the significance of

several variables simultaneously. The LR test is more exact when testing joint signifi-cance, because it refits a model while the Wald test uses a quadratic approximation to

the likelihood function. For model (3), the LR test for joint significance of spouse and all interaction terms marginally cannot reject the hypothesis that these coefficients are

zero at the 10 per cent level (see Appendix II, Table 6).

The hypothesis that spouse in model (2) equals zero can neither be rejected by the

Wald test nor by the LR test (see Appendix II, Table 4 and Table 5). The lack of sig-nificance could also be due to multicollinearity. Based on the descriptive statistics, the

variable spouse could correlate with the variables age, education, household size or workflag. However, based on the correlations and the variance inflation factors, there

is not much evidence of multicollinearity (see Appendix II, Table 8 and Table 9). This indicates that spouse is insignificant and thus model (1) and (3) are preferred. This

im-plies that there is not enough evidence to support hypothesis (2) of a differing constant for female heads and female spouses.

Model (1) is based on hypothesis (1) which states that the microfinance participation decision does not significantly differ between the two groups of women. To investigate

(26)

variables as well as economic variables is established and compared to model (1) (see Appendix II, Table 10 and Table 11). The interaction terms of spouse with the social

variables are not significant according to the LR test which indicates that the effects of social determinants on the microfinance participation decision do not significantly

differ for female heads and married woman (see Appendix II, Table 7). Model (1) and Model (3) are further compared to investigate the possibility of differing effects of the

economic variables.

Model (1) and model (3) have a very good hit rate regarding the expected hit rate

and they both correctly predict approximately 82 per cent of the values (see Appendix II, Table 12). Though model (1) is preferred when comparing the BICs, model (3) has

a higher likelihood. Furthermore, heteroskedasticity is tested for, since heteroskedastic error terms could lead to inconsistent marginal effects when estimation is done with

maximum likelihood. Both models have a low significance on the presence of he-teroskedasticity (see Appendix II, Table 13). Model (1) has homoskedastic error terms

at the 5 per cent significance level, whereas model (3) is homoskedastic at the 10 per cent significance level.

Concluding, model (3) is slightly preferred to model (1) because of the higher likeli-hood, the smaller evidence for the existence of heteroskedasticity and the similar

predic-tive performance. Though the variables spouse and the interaction terms are marginally not significant at the 10 per cent level, the interaction term of spouse with savings is

significant at the 1 per cent level and spouse with shg at the 10 per cent level. Hence, hypothesis (3) which states that economic variables differ between female heads and

(27)

5 Analysis

5.1 Social variables

This section presents the analysis of the social variables that determine the microfinance

participation decision of female heads and female spouses in the BSS programme in Karnataka. Model (3) accounts for different economic effects between female heads

and spouses and is preferred above model (1) and model (2). In Table 3, the average marginal effects of model (3) are given. The variables age and age2 _{are significant at}

respectively the 1 per cent and 5 per cent level. An increase in age of one year decreases the probability of participation by 0.97 per cent for the average household. This is in

Table 3: Average marginal effects of model (3)

participation dy/dx Std. Err. z

age – 0.0097*** 0.0033 – 2.96 age2 0.0001** 0.0039 2.16 education – 0.0033* 0.0017 – 1.96 unknown caste 0.1647* 0.0892 1.85 scheduled caste 0.1484*** 0.0229 6.47 OBC caste 0.0831*** 0.0218 3.82 household size 0.0232*** 0.0058 4.02 savings – 0.0936* 0.0562 – 1.66 shg 0.1336** 0.0567 2.36 workflag 0.0392 0.0411 0.95 spouse – 0.0264 0.0401 – 0.66 spouse×savings 0.1557*** 0.0587 2.65 spouse×shg – 0.1064* 0.0592 – 1.80 spouse×workflag – 0.0238 0.0428 – 0.56

(28)

line with the findings of Kangogo (2013) as well as Anjugam and Ramasay (2007) and can be explained by a lack of energy and motivation of older women (Shah et al., 2008).

Another reason could be that younger women live in more modern household structures where the husband does not form as much a constraint as in older households. Moreover,

younger women could be more likely to invest in their future and their family’s well-being, while older women are likely to be widows and thus less motivated to do such

investments. The significant quadratic term implies that the effect of age is nonlinear, meaning that the effect decreases with the age of women.

The social variable for the level of education is significant at the 10 per cent level. A one-unit increase in education decreases the probability of participation by 0.33 per

cent. This indicates that the higher the educational level, the lower is the probability that this person will take up microfinance. This supports the findings of Kangogo (2013)

and Oletaju et al. (2018), who assume that highly-educated people have well-paid jobs and are not in need of financial aid. Moreover, they are more likely not to be rejected

by formal banks and thus will find formal sources of finance with lower interest rates compared to microfinance programmes.

The unknown caste variable is significant at the 10 per cent level, whereas the vari-ables for the scheduled and OBC caste are significant at the 1 per cent level. The lower

significance of unknown caste can be due to the low amount of observations on this variable. All three added dummy variables for the caste are positively-signed. This

indicates that a person belonging to the general caste is less likely to participate as op-posed to a person from any of the other castes. Women who declared not knowing their

caste or of whom no caste is recorded are most likely to participate in microfinance programmes, followed by members of the scheduled caste. Members of the scheduled

caste are 14.84 per cent more likely to participate than members of the general caste. It appears that the lower the caste, the higher is the probability of participation. This is

in line with the finding of Anjugam and Ramasay (2007) who state that the higher the social backwardness that is represented through the caste, the higher the microfinance

(29)

are discriminated against. Attaining microfinance could help them start a small busi-ness and earn money through their busibusi-ness, offering an opportunity for improvement

without the constraints of discrimination.

The variable household size is highly significant. An increase of the household size

by one member increases the participation probability by 2.32 per cent. This is in line with Rasheed et al. (2016) and also with Kangogo (2013) who finds that an additional

member increases the probability by 7.7 per cent. He assumes that the burden of an extra member is the reason for the higher participation probability. Another reason

could be that household spouses and heads do not apply for the loan for themselves, but instead they apply for other family members. In this case, a larger family increases the

probability that the women apply for a loan for another family member.

Altogether, the findings on age and education contradict the expected effects based

on the human capital theory which describes that well-educated people with more ex-perience are more likely to participate in entrepreneurial activities. A lower caste and a

larger household size increase the probability of participation.

5.2 Economic variables

The mean marginal effects of the economic variables are depicted in Table 3. Differ-ences between spouses and female heads are estimated through the variable spouse and

its interaction terms. The term spouse is negatively-signed, which indicates that mar-ried women are less likely to pick up microfinance. This observation is in contrast with

the expectation that female heads are more constrained as opposed to married women. Moreover, if female heads do belong to the poorest, this indicates that the poor rather

than the less poor participate in microfinance programmes. This finding supports the theory of diminishing returns to capital. However, the variable spouse is not significant.

Model (3) contains the economic variables savings, shg and workflag as well as the interaction terms with spouse. The variable savings is significant at the 10 per cent level

and the interaction term spouse×savings is significant at the 1 per cent level. The mi-crofinance participation probability for a female household head decreases by 9.36 per

(30)

cent if she engages in saving activities. In the case of married women, the probability increases by 6.21 per cent when she possesses savings as opposed to spouses who do

not have savings, which is in line with Kifle et al. (2013). This finding contradicts the suspicion that female spouses do not make as considerate financial decisions as female

heads do. One reason for this finding could be that female heads use microfinance as a last resort to escape from poverty. If they maintain savings, this implies that they

earn enough income to save and thus do not desperately need microfinance services. They might not be willing to take the risk of investing in a business and thus are less

likely to participate as opposed to female heads who do not have savings. On the other hand, married women might be willing to invest in a business to improve the household

conditions when the family has savings, because her husband can function as a finan-cial backup. Moreover, women in general are more risk-adverse than men (Hanley &

Schmidt, 2013). This indicates that a husband could positively influence the microfi-nance participation decision of his wife, because he is more likely to take the risk of

investing in a business. Hence, savings can positively influence spouses, whereas it negatively influences female heads.

The variable shg is significant at the 5 per cent level, whereas its interaction term is significant at the 10 per cent level. Participation in self-help groups has a positive

effect on the participation decision of both female heads and married women, which supports the finding of Anjugam and Ramasay (2007). However, the estimated effect of

participating in a self-help group is greater for female household heads than for spouses. Female heads who participate in self-help groups have an increased probability of 13.36

per cent to consecutively apply for microfinance as opposed to heads who do not par-ticipate, whereas this percentage is only 2.72 per cent for female spouses. This

sup-ports the expectation that the demonstration effect yields a smaller positive effect for female spouses than it does for female heads. This can be due to constraints imposed

by husbands, indicating that the participation of female spouses indeed depends on the approval of their husband. The positive effect supports the statement of Pitt et al. (2003)

(31)

within the household.

Female heads who worked during the last week have an increased probability of

3.92 per cent compared to female heads who did not work, while this percentage is 1.54 per cent for married women. Though a positive demand was expected for both groups,

the expectation that the demand of female heads is lower is not proven to be true. A possible explanation is that female heads have to fulfill more tasks in the household,

since they do not have the support of a husband, and thus the flexibility that comes with owning a business is especially beneficial to them. However, both the variable workflag

and its cross term spouse×workflag are not significant.

All in all, the findings of the negative influence of savings on the decision of female

heads and the smaller positive influence of workflag on married women as opposed to female heads are rather unexpected. The findings suggest that female heads use

microfinance as a last resort to escape from poverty and that female heads are more risk adverse and insecure as opposed to married women. Nevertheless, the expectation

on the demonstration effect is met, because female heads who participate in self-help groups are more likely to take up microfinance than married women who participate

in self-help groups. The findings on savings and shg support the statement that the participation decision of married women is influenced by their husbands.

6 Conclusion

The subject of this study was to determine the socio-economic factors that influence

women’s decision to participate in microfinance in rural southern India, while account-ing for possible differences between married women and unwed female household heads.

After comparing three probit models, the model that accounted for economic differences between the two groups of women was slightly preferred to the model that did not

con-tain differing effects for these two groups. This result affirms the third hypothesis of different economic effects. Social variables that are found to significantly influence

(32)

quadratic term of age, the educational level, the caste and the respondent’s household size. Economic variables are the savings and the participation in a self-help group as

well as the interaction of these two variables with the dummy variable for being a mar-ried woman. The workflag which was used as a proxy for a regular income was not

found to be significant.

From the average marginal effects of the social variables, it can be concluded that

human capital does not positively influence the participation decision. Factors like the caste and household size do have a positive influence. Furthermore, if female headship

can be used as a proxy for the poor, it is not the less poor that are more likely to partici-pate in the BSS programme which supports the theory of diminishing returns to capital.

Nevertheless, this finding is insignificant.

As expected, the demonstration effect seems to have a greater effect on female heads

as opposed to married women. This finding in combination with the finding that savings yield an opposite effect on married women against female heads leads to the suspicion

that the husbands of married women might influence the participation decision for the BSS programme. In a household with savings, husbands seem to have a positive

in-fluence, whereas in households where the woman participates in a self-help group, the husband appears to have a negative influence. Hence, attention should be paid to gender

relations when targeting households for microfinance as stated by Chant (2003). Fe-male heads seem to use microfinance as a last resort to escape from poverty and they

also appear to be more insecure and risk adverse than married women. This supports the idea that female heads belong to a more vulnerable group. Microfinance services

can be especially beneficial to these women, because it is found to empower women significantly and helps them improve their financial situation (Pitt et al., 2003).

Altogether, when targeting women for participation in microfinance projects, so-cial determinants that influence their decision should be taken into account, like age,

education, caste and household size, as well as economic determinants like savings and participation in self-help groups. However, not only these constraints influence the

(33)

partic-ipation decision due to constraints on a broader level, namely the household level for married women and the cultural and personal level for unwed female household heads.

Microfinance institutions should be aware of these determinants, the husband as a pos-sible constraint and the vulnerability of female heads when deploying their policies and

targeting female applicants.

7 Limitations and recommendations for further research

The present study has a few limitations. First, the different effects of savings and

self-help groups on married women and female heads provide evidence for different barriers that restrain the women from participating. These barriers are very likely to be on a

household level for the married women with the husband as a main constraint, while

financial and personal insecurity as well as risk adversity might form main constraints to female heads. More research on these constraints is needed to measure their impact

on the participation probability. By conducting a survey with married women and fe-male heads that contains questions on insecurity and risk adversity, these constraints

can be measured. The survey for example can contain questions about how often the respondent got rejected by official banks and whether the respondent feels desperate or

not.

If the husband indeed influences or restrains their spouse, this can be done on

pur-pose or unconsciously. Knowing this difference is important for efficiently targeting married women for the microfinance programme, since the husband might need to be

involved in this process. The influence of the husband can be investigated through qual-itative research in which the husbands of the households are asked for their opinion on

the microfinance programme offered to their spouses. The positive or negative nature of their response can be an indication of whether the husband’s influence on the

parti-cipation probability of the household is positive or negative. To measure the nature of the response, a diverging scale can for example be used with options ranging between ”I

(34)

influencing the decision, the husband could also completely decide whether the house-hold participates or not. Hence, the husband and the spouse can also independently and

anonymously be asked whether it was the spouse who made the participation decision or not such that an insight can be obtained whether the husband makes the final decision

to participate in the programme.

In this research, the variable used for savings was a dummy variable based on the

question whether or not the respondent of the household has a savings account. This leads to two different issues that should be improved in further research. First, it is not

known whether a woman who stated to have savings does have control over the account and access to the savings. It could also be that she means the household’s savings

account that is in control of her husband or other family members of the household. If a woman has savings at her disposal, this can be an indicator for a liberal and modernized

household where the woman is less risk adverse and the husband is less of an influence. Hence, in further research, it should be investigated who possesses and controls savings

in the household. Secondly, the dummy variable does not reveal the amount of savings on the account. By gathering information on the amount of savings of the household

members, more information on the impact of this variable can be obtained. This can be done by asking for the amount of savings.

Another limitation of the study is that it is not certainly known whether it is the fe-male head or the spouse who applies for the loan, since the participation was recorded

at a household level. A daughter or other female family member in the household could also have applied for the loan. Furthermore, income is often found to significantly

in-fluence the participation decision in literature, but the data did not contain this variable. Collecting data on the monthly income of each household member is valuable for

fur-ther research and can give more insight into the reason of applying for microfinance services and if these are indeed used as a last resort to escape from poverty by female

heads. Another variable that can help to estimate the reason for applying for the mi-crofinance services is the loan usage. The loan can be used for a business as it mostly

(35)

Information on this variable can be gathered by asking households whether or not they will invest the loan in a business.

Another limitation of this paper is that the workflag was used as a proxy for regular working activities. This variable was insignificant which can be due to the fact that

it only captures whether a respondent worked during the last week or not. Hence, this variable can be highly unrepresentative for some households. The household’s members

should be asked whether they have a secure job to improve the quality of this dummy variable and to be able to investigate the impact of a steady job on the participation

decision.

The surveys held to originally gather the data used in this paper were held before

the BSS microfinance institution entered the villages and thus before the microfinance programme was introduced and offered to the households. In further research, it might

occur that only information on the final participants is available instead of information on all eligible households. In this case, the sample is non-random and there is a question

of sample selection bias. To still obtain consistent estimates the Heckman two-step method can be applied. The demanded loan size per household can be gathered and

used as the dependent variable. In the first step of this method, a probit model for the selection mechanism is estimated and the inverse Mills ratio is computed. In the second

step, the model for the selected sample is estimated, including the inverse Mills ratio which accounts for the selection bias. If the coefficient of this ratio is significant, then

there is indeed a question of sample selection bias, so the two-step method should be used.

Furthermore, the results of this research suggest that underlying factors such as risk adversity and insecurity due to for example discrimination influence the microfinance

participation probability of female heads. This indicates that female heads might not engage in microfinance projects on the basis of unobservable factors, leading to a

se-lection bias and endogeneity in the model. To account for these underlying effects, the model in the first step of the two-step method should include variables that capture the

(36)

measure this willingness without measuring the actual need for microfinance like for example variables that are based on personality traits or on the social network. The

per-sonality traits could account for risk adversity and insecurity while the social network can influence the willingness to participate in new activities as well as the respondent’s

feeling about the new activity. Personality traits can be measured by the URDU version of NEO-FFI test as done by Maazullah (2016) and network effects can be measured as

done by Banerjee et al. (2013). Furthermore, Maazullah (2016) spread a survey in his sample consisting of eligible women as well as men to gather information on their

opin-ion about microfinance institutopin-ions. If women do belief that microfinance institutopin-ions are not trustworthy, they are also less likely to apply for microfinance services regardless

of their need for these services. Moreover, as mentioned before, the number of rejec-tions by banks and whether the respondent feels desperate as well as the control over

the savings account in the household can indicate the risk adversity and insecurity and thus the willingness to participate in microfinance, without capturing the actual need for

microfinance.

Finally, in this research, the focus lied on the demand side. The supply side was

assumed to be constant, because the microfinance services are equally offered to all poor women. If this assumption could not be made, simultaneous equation bias would be the

consequence because of the simultaneity of demand and supply that leads to endogenous regressors. In further research where the assumption of a constant supply cannot be

made, using valid and relevant instrumental variables and applying estimation with two-stage least squares offers a solution. The instrumental variables for supply should be

chosen such that they do not affect demand, but that they do contain the constraints of the supply side. Discrimination often is a constraining variable on the supply side.

Hence a variable that captures this constraint could be used as an instrumental variable for the supply side. Dependent on the microfinance institution and area of supply, further

research should take into account the possibility of simultaneous equation bias and find instrumental variables that account for corresponding supply side constraints.

The socio-economic determinants of microfinance participation : an analysis of married women and unwed female household heads in Karnataka, India