Evaluation of the impact of Village Savings and Loan Associations using a novel survey instrument

ABSTRACT

Formal microfinance institutions have been an important tool in the fight against poverty in developing countries, but their reach is necessarily limited. Village Savings and Loan Associations (VSLAs) are an alternative, informal mechanism for saving and borrowing that do not require external capital or ongoing financial or administrative support from a founding organisation. This article evaluates the impact of VSLAs on their members and finds that long-term members fare better along multiple economic, nutritional, and health dimensions compared with a control group of recent joiners.

KEYWORDS:

• Microfinance
• Village Savings and Loan Associations
• Africa
• Tanzania

1. Introduction

During the last decade, microfinance institutions have provided access to financial services for millions of people in developing countries. However, provision of financial services in rural areas remains a major challenge. With poor road quality and lower population density, it can be extremely costly for microfinance organisations to reach the rural poor and consequently the great majority of rural areas lack access to any formal financial services.

To address this difficulty, CARE International, a non-governmental humanitarian organisation, designed a unique savings-based programme called a Village Savings and Loan Association (VSLA). VSLAs allow the poor to become their own bankers. They are built entirely on member savings and interest from loans; they receive no direct capital investment from external organisations. Members, however, do receive a year of intensive training in group governance and money management, which allows them to become self-sufficient and even enables them to establish and train other groups.

As a self-sustainable and self-replicating approach, VSLAs have the potential to improve financial access in more remote areas, but the impact of these groups is not well understood. Relatively little research, and no randomised controlled trial, has been performed to evaluate the effectiveness of VSLAs. Previous research suggests that programme participants have higher household wealth, better food security and health, and higher education expenditures (Allen & Hobane, 2004; Anyango, 2005; Anyango et al., 2006), but these studies suffer from methodological limitations that preclude interpreting these associations as causal.

The purpose of this study is to expand and improve upon the existing research to better understand the impact of the VSLA programme. The study evaluates the impact of one of the oldest VSLA programmes in Zanzibar, Tanzania. The study is comprised of an individual questionnaire administered to 170 households, including those of current VSLA members, former members, and incipient members. The incipient members serve as a control group in order to isolate and assess the impact of the VSLAs on the longer-term members. The survey data are complemented by three focus group discussions as well as several interviews with key informants within CARE and its affiliated organisations. Finally, a thorough understanding of both the economic and social settings in which the programme operates, as well as of the institution itself, facilitates interpretation of the data from the survey and focus group discussions.

The results suggest that participation in the VSLA programme has an overall positive impact on various indicators of household and individual welfare, including asset expenditure levels, the development of income-generating activities, spending on education, access to health services, nutritional levels, and quality of housing. Overall, these results suggest that the VSLA model is both sustainable and successful at reaching those who do not benefit from traditional microfinance programmes.

The remainder of the article is organised as follows. Section 2 describes the functioning of VSLAs and surveys the literature on microfinance institutions generally with particular emphasis on VSLAs. Section 3 describes the survey instrument used to collect data and the empirical methods used to evaluate the impact of VSLAs. The results of the empirical analyses are presented in Section 4. Section 5 concludes and provides a discussion of the potential policy implications of the results.

2. Background

2.1. Village savings and loan associations

Proponents have hailed the provision of formal financial services to the poor as a promising tool for alleviating poverty and fostering development (Morduch, 1999). The underlying logic is that providing financial services to the poor enables them to manage their money differently; for example by investing, acquiring productive assets, acquiring new skills, or opening new businesses. Formal microfinance institutions, such as the well-known Grameen Bank, have arisen to funnel investor capital to individuals in developing countries in the form of small loans.

In addition to formal microfinance institutions, more informal mechanisms for saving and borrowing have been developed. One such approach is a rotating savings and credit association (ROSCA). In a ROSCA, a small group of people, generally between 15 and 30, form a group and contribute an agreed amount at regular meetings. The entire fund is then distributed to each member on a rotating basis, until everyone in the group has received a loan. Several studies have attempted to explain the motivation behind ROSCA participation. Anderson & Baland (2002) argue that ROSCA participation is an effective strategy used by women to protect their savings against claims from their husbands. Bauer et al. (2008), Gugerty (2007), and Dagnelie & LeMay-Boucher (2008) suggest that individuals use participation in a ROSCA as a device to commit themselves to save money and to deal with self-control problems. Besley et al. (1993) argue that individuals who have no access to credit may choose to join a ROSCA to finance the purchase of indivisible durable goods, taking advantage of the gains from intertemporal trade between individuals.

The VSLA model developed by CARE improves upon the ROSCA approach in several ways. First, borrowing from the group savings can be done at any time, making it easier for individuals to time their borrowing to better match their consumption-smoothing needs or investment opportunities. Second, members can contribute differing amounts, enabling those with greater means to save more and thus increasing the potential amount available to be lent. Finally, borrowers pay interest on loans to the group, which should encourage more savings from those with greater means while simultaneously discouraging borrowing for less productive purposes. The net effect should be that more capital is provided by better-off members and used by borrowers for more productive purposes.

This study examines the VSLA programme in Zanzibar, Tanzania. Tanzania has the highest rate of extreme poverty in the world, with 88.5% of the population subsisting on less than US$1.25 per day and 96.6% on less than US$2 per day (World Bank, 2009a). The Tanzanian population is also poorly educated – in 2007 only 69.4% were literate (World Bank, 2009b). Access to credit is severely limited. As of 2007, just 10% of the population had access to formal financial services, up from 6.4% in 2001 (World Bank, 2009a).

In the region of this study, a VSLA consists of 15 to 30 people who save a small amount every week. A share is usually TSh1000 (US$0.90)¹ with members contributing up to three shares per week, which corresponds to approximately 8% of the average weekly income. The value of each share remains low to allow the poorest members to participate. The group’s funds are kept in a cash box that is fitted with three padlocks, the keys of which are held by different officers in the group. This system improves transparency and makes it easier to refuse loans to non-members, such as one’s husband (Allen & Staehle, 2007). After several months, the savings accumulated by the group become large enough to launch the loan function. All members have the right to take out a loan regardless of the number of shares they have contributed, but can only take out a loan equal to at most three times the value of their shares. Most loans are short term, generally around one month, at an interest rate determined by the group, usually 5% per month. This is low compared with moneylenders who often charge up to 30% per month, but slightly higher than non-governmental organisation affiliated microfinance institutions, which generally charge less than 4% per month (Mutesasira, 1999). Each group is able to set their own repayment terms. However, a VSLA never fines borrowers for late loan repayment because this may aggravate any underlying crisis the household may be facing. It is assumed that the embarrassment of being late is sufficient penalty (Allen & Staehle, 2007).

On a date chosen by the members, usually after about a year, the savings and accrued interest are divided among the members in proportion to each individual’s savings. This event, known as an ‘auction audit’, is usually scheduled to occur when members are most likely to need money, such as at the start of the school year or before a major holiday, in order to encourage the use of savings to meet pressing needs and discourage their use for unnecessary expenditures. After the disbursement of funds, the groups normally re-form immediately and start a new cycle of savings and lending.

The VSLA model is lauded for its transparency and adaptability for illiterate members. All operations (deposits, withdrawals, loans, loan repayments) occur at weekly meetings with the entire group present so that all activities remain transparent. Record-keeping was also designed to be as simple and as transparent as possible. Each member has an individual passbook, which is stamped every week, with each stamp representing one share. Only loan disbursement is recorded in the group ledger (Allen & Staehle, 2007).

VSLAs are built entirely on member savings and interest from loans; they receive no direct capital investment from CARE or any other supporting organisation. CARE’s role is to supply extensive training on group dynamics, governance, and money management. Each VSLA elects a Community Contact Person (village trainer). After approximately a year of supervision, if the Community Contact Person passes a certification test, the field officer from CARE moves on to another group and starts the process again, leaving the Community Contact Person in charge. Meanwhile, a local umbrella organisation, called an Apex organisation, is created to support and monitor existing groups while fostering the growth of new groups. After several years, once the necessary systems have been put in place, the Apex organisation is left in charge of the continued promotion of the VSLA model and CARE is able to move on to new areas.

CARE first introduced the VSLA programme in Zanzibar in 2001, and created a local Apex organisation called the Jozani Credit Development Organization (JOCDO) to support the growth and development of VSLAs in the region around the Jozani Chwaka Bay National Park. After two years of successful implementation, CARE moved on to different regions in Zanzibar, leaving JOCDO to oversee the continuation of the model around the Park. Over the past seven years, under the guidance of JOCDO, the number of VSLAs in the area has grown to 233. CARE maintains an active presence on the island, expanding the VSLA programme to new regions and providing general oversight of the existing VSLAs. However, JOCDO is responsible for all day-to-day operations, including supporting existing groups, and fostering and training new groups, in the Jozani Chwaka region.

2.2. Previous literature

Recent evidence suggests that microfinance, and specifically microcredit, does not have a transformative impact on poverty (Karlan & Zinman, 2011; Angelucci et al., 2015; Attanasio et al., 2015; Banerjee et al., 2015; Crépon et al., 2015; Tarozzi et al., 2015). However, it can give low-income households more freedom in optimising how they make money, consume, and invest. Evaluations in India, Mexico, Mongolia, and Morocco indicated that expanded credit access allowed entrepreneurs to invest more in their business, but these investments rarely resulted in profit increases and never led to substantial increases in income (Angelucci et al., 2015; Attanasio et al., 2015; Banerjee et al., 2015; Crépon et al., 2015). There is also little evidence that microcredit access has a substantial effect on women’s empowerment or investment in children’s schooling, but it does not seem to have widespread harmful effects either (Karlan & Zinman, 2011; Banerjee et al., 2015; Crépon et al., 2015).

While the evidence on microcredit is somewhat uninspiring, savings products may offer more promise. Evaluations in Kenya, Malawi, Nepal, and the Philippines have found that improved access to and usage of savings accounts led to increased savings, income, entrepreneurial activity, and investment, and protected households against income shocks (Ashraf et al., 2006; Prina, 2013; Dupas & Robinson, 2013a, 2013b; Brune et al., 2015). Access to savings also led to increased investment in health and education in Kenya, Nepal, and Uganda, and to increased food security and better food consumption in Nepal (Prina, 2013; Dupas & Robinson, 2013a; Jamison et al., 2014). In the Philippines, access to savings led to increased female decision-making power (Ashraf et al., 2006).

Savings mechanisms may be especially beneficial for women (Kabeer, 2001; Dupas & Robinson, 2013a). Anderson & Baland (2002) argue that more formal savings mechanisms allow women to protect their savings against claims from their husbands. Dupas & Robinson (2013b) and Schaner (2012) suggest that women also face constant demands from relatives and neighbours and may find it difficult to refuse requests if the money is available in the house. Dupas & Robinson (2013a) find that four to six months after opening a savings account, women in Kenya had 45% higher daily investment in their business, 10 to 20% higher daily food expenditures, and were better able to afford medical expenses for serious illness.

Informal and semi-formal financial schemes are likely to have similar benefits to more formal savings mechanisms. Despite the prevalence of such schemes, the evidence on the impact of the mechanisms is very limited. In 2006, Decentralized Financial Services, a consulting group based in Kenya, carried out an impact study of a VSLA programme in Zanzibar to examine its long-term sustainability and its impact on its members (Anyango et al., 2006). The study finds that VSLAs in Zanzibar have performed well in terms of growth and sustainability, and suggests that participation in the programme led to improved living standards and housing, and increased income. Although these results are encouraging, the study does not have a baseline for members, does not use a control group, and no tests of statistical significance were performed.

Other impact evaluations, however, seem to find similar positive impacts to VSLA programme participation. Allen & Hobane (2004) and Anyango (2005) conclude that in Zimbabwe and Malawi, respectively, membership in a VSLA contributed to an increase in household productive and non-productive asset levels among the majority of participants, as well as some improvement in the quality of housing. The findings also suggest that programme participation led to an increase in the number of income-generating activities and to an increase in stability of such activities. In Tanzania, the Women’s Empowerment Strategic Impact Inquiry found that female VSLA participants had higher savings, more income-generating activities, greater food security and health, and increased education expenditures (CARE Tanzania, 2006).

3. Methods

3.1. Research design

This study uses data from an original survey of VSLA members in an attempt to elucidate the economic and social impacts of the programme with a particular emphasis on the impact on female participants. Economic impact is measured principally through expenditure levels, the accumulation of household assets, and the development of income-generating activities, such as fishing, tailoring, or carpentry. To estimate social impact, the study relies on a variety of indicators, including educational spending, access to health services, nutritional levels, and quality of housing.

It is likely that VSLA members systematically differ from the general population. The establishment of new VSLA groups involves a process of self-selection, in which the most energetic and highly-motivated men and women are more likely to become involved, while the marginalised or vulnerable may be overlooked. The poorest also may be excluded because of their inability to finance the purchase of shares. A comparison of VSLA members with the general population therefore may be biased towards finding beneficial effects of the programme. In order to address this problem and improve upon the methodology used in Anyango et al. (2006), we compare established VSLA members with a control group of new members who are still in the initial training phase. The use of new members as a control group offers two operational advantages. First, there is no need to identify and survey non-members in order to generate a control group – it can be particularly difficult to motivate such a group to take part in a time-consuming survey. Second, there is no need to follow clients over time, as in a longitudinal survey (Karlan, 2001).

Using new members as a control group requires three major identifying assumptions. The first of these is that no one drops out of the programme or dropping out occurs for reasons orthogonal to the variables of interest. Failure of this assumption could cause two problems: incomplete sample bias and attrition bias (Karlan, 2001). Incomplete sample bias refers to the fact that those who drop out may have been impacted differently than those who remained. Programme benefits would be underestimated if individuals who became more economically successful later dropped out; for example because they no longer needed access to the savings and loan functions. On the other hand, programme benefits could be overestimated if individuals whose economic situation did not improve dropped out, perhaps because they felt the programme had failed to help them. In order to address potential dropout bias, the study includes a group of dropouts in the treatment group, the size of which is based on the approximate attrition rate experienced in the programme. Attrition bias is addressed by controlling for client characteristics, such as age, educational attainment, and number of children at the time of joining the VSLA group.

The second identifying assumption is that there is no change in how selection of VSLA members occurs over time. If the first to join the programme are wealthier, more entrepreneurial, or perhaps considered by their peers to be more reliable and trustworthy, programme impacts may be overestimated. The less well-situated community members who join later would not provide an accurate ‘baseline’ against which to measure the treatment group. However, the bias caused by changing selection effects over time may also run in the opposite direction – that is, programme impact may be underestimated if the poor are the first to join; if, for example, they are willing to take greater risks than their wealthier, more conservative neighbours. To control for changing selection effects, comparisons are made between time-invariant characteristics of the treatment and control groups.

The third assumption is that benefits to programme participation do not occur immediately upon sign-up, but rather accrue over time. If new members were able to change the outcome variables in anticipation of future resources, then comparing them with longer-term members may not find any differences. This seems unlikely here because of the liquidity-constrained nature of the study population, so this third identifying assumption appears quite reasonable.

In addition to the three major assumptions discussed, the use of new members as a control group potentially involves a problem of changing institutional dynamics, which would impact the composition of the new versus veteran participant pool. The credit or savings programme may change its strategy and/or client identification process. Programme placement also may change – for example, the programmes might prefer to start out cautiously and enter slightly more well-off communities, and then, only once they are successfully established, branch out into poorer neighbourhoods. Programme placement may also work in the other direction. Any of these changes might affect the relative make-up of the two different groups, thus biasing any comparisons. Karlan (2001) suggests that the best, and perhaps only, way to deal with these problems is through a solid understanding of the selection process involved and the institutional dynamics. From interviews with key informants, including employees of both CARE and JOCDO, it appears that the client identification process has not changed substantially within the past 10 years. The leadership of every village in an area is approached to explain the programme. The village leader is then responsible for informing his community of the opportunity. If there is a group of 15 to 30 people who are interested in becoming VSLA members, they are encouraged to contact JOCDO. No special effort is made to reach out to any particular subset of the community. Furthermore, because all villages in the area are informed of the programme, there is little reason to believe that the nature of the communities involved in the programme has changed over time. Although this evidence is unavoidably anecdotal, it suggests that changes in the selection process or institutional dynamics will not bias the results of this study.

3.2. Sampling

At the time of the survey there were 233 VSLA groups in Zanzibar (61 trained by CARE and 172 added since JOCDO took over the organisation and training of new groups). However, only groups that were included in the sample used by Anyango et al. (2006) were included in the final sample for this study. This includes the 73 groups that were formed before mid-June 2004. By relying on the sample used in the previous study, it is possible to ensure that only the most ‘mature’ groups are included in the study. This facilitates analysis of the long-term impacts of programme participation. The control group is made up of 50 individuals in five new VSLA groups that began training in early January 2010. The survey took place late in the same month; therefore, these five groups were still only in the very initial stages of the training process and had not begun saving in or borrowing from their new VSLAs.

From the sample of 73 groups, 25 groups spread across 13 different villages were chosen randomly. Four members (with two alternates) were then selected randomly from each of these groups to be interviewed. Although only groups that formed before mid-2004 were included in the sample, within each group the members were chosen randomly and therefore the average length of membership was only five years. In addition to the four current members from each group, 20 dropouts were selected randomly from the full set of 25 groups, based on JOCDO’s estimated attrition rate of 20%, to be interviewed in order to control for potential dropout biases. These dropouts were included with the current participants in all analyses. In total, 170 current, former, and incipient VSLA members were interviewed.

The questionnaire tool covered the basic socio-economic characteristics of the respondents and their households: participation in the VSLA programme, asset levels, housing characteristics, nutritional status, access to healthcare, and social impact. In order to facilitate comparisons, where possible, the questionnaire matched that used by Anyango et al. (2006).

Three focus group discussions, each with between 15 and 20 participants, were carried out to supplement the information gathered in the individual survey. The participants for the three groups were selected randomly from the original sample of 73 VSLAs, after excluding the 25 groups that were already included in the quantitative research so as not to recount the information gained through the individual survey. The focus group discussions covered issues such as group formation and membership; general group dynamics; challenges and limitations; behavioural changes; social and economic impact; benefits and/or negative consequences of participation; impact on the community; and the sustainability and effectiveness of the Apex organisation. In addition, each group was visited during its weekly VSLA meeting in order to observe the methodology and activities of the group as well as general group dynamics.

3.3. Empirical strategy

Simple comparisons of the means across the treatment and the control groups allow for initial estimations of programme impact. Regression analysis is then used to further explore programme impact, while controlling for individual and household characteristics, which might also impact the outcome variables. The basic model used in the regression analysis is as follows:

where is an outcome of interest; is a binary variable equal to one for long-term VSLA members or dropouts and equal to zero for new members; is equal to one for female and zero for male; and represents a vector of control variables, including age, religious status, marital status, number of children, educational status, and prior savings or access to credit. Under the identifying assumptions discussed in Section 3.1, the coefficient reflects the impact on the outcome variable of VSLA membership on males, represents the difference in the outcome measure between non-member males and females, and represents the difference in the impact of VSLA membership on females compared with males.

3.4. Data

Initial comparisons of descriptive statistics, presented in Table 1, suggest that the treatment and control groups are similar along most dimensions. Existing VSLA members (the treatment group) are older than newly joining members (the control group) and also have more children, but these differences disappear if we compare age and the number of children at the time of joining a VSLA. The only other statistically significant differences are in educational attainment, and these favour the control group. This difference would be expected to bias the results towards finding the VSLA programme to be ineffective since most of the outcome measures are likely to be positively correlated with education.

Table 1. Demographic characteristics of treatment and control groups.

Control variable	Treatment group	Control group	Test statistic
N	120	50
Gender (% female)	0.68	0.72	0.58
Age	37.95	33.64	2.18**
Age at time of joining	33.19	33.64	0.23
Religion (%)
Muslim	0.97	1.00	1.31
Christian	0.03	0.00	1.13
Other	0.01	0.00	0.65
Marital status (%)
Married	0.75	0.70	0.67
Widowed	0.08	0.08	0.07
Divorced	0.05	0.08	0.76
Separated	0.02	0.00	0.92
Single	0.10	0.14	0.75
Educational attainment (%)
No education	0.13	0.20	1.10
Primary	0.48	0.20	3.34***
Ordinary level	0.16	0.32	2.38**
Advanced level	0.23	0.28	0.64
Number of children	3.73	2.62	2.72***
Number of children at time of joining	3.08	2.62	1.14
Savings prior to joining VSLA? (%)	0.48	0.36	1.38
Access to loans prior to joining? (%)	0.08	0.08	0.11

Notes: VSLA = Village Savings and Loan Association. Means for the treatment and control groups were compared using t tests for non-proportions and a proportions test to compare proportions. The reported test statistics are t statistics for non-proportions and z statistics for proportions. Statistically significant results are indicated as follows: ***p < 0.01, **p < 0.05.

To further address the concern that the first to join the VSLA programme might have been better off or more entrepreneurial than those who joined later, Table 2 breaks down the data by subdividing the treatment group by the median number of years of programme participation. Again, the basic characteristics of the treatment group do not appear to be statistically different from those of the new members in the control group. Furthermore, there does not appear to be any evidence that the older members of the treatment group – the ‘pioneers’ – are significantly different from more recent members, thus suggesting that the characteristics of VSLA programme participants have not changed over time. If anything, newer members appear to be of higher ‘quality’ than older members, in terms of both the education and savings. Therefore, any bias introduced by changes in the characteristics of VSLA participants over time should distort the results towards finding the programme to be less effective than it truly is.

Table 2. Comparison of control variables with treatment group divided by median years in VSLA.

Control variable	Mean/percentage			Test statistic
	Older	Recent	New	Recent–Older	Recent–New	Older–New
n	63	57	50
Gender (% female)	0.71	0.63	0.72	0.97	0.97	0.07
Age	41.03	34.54	33.64	3.20**	0.40	3.37***
Age at time of joining	34.37	31.90	33.64	1.24	0.77	0.33
Religion (%)
Muslim	0.97	0.96	1.00	0.10	1.34	1.27
Christian	0.03	0.02	0.00	0.50	0.94	1.27
Other	0.00	0.02	0.00	1.06	0.94	0.00
Marital status (%)
Married	0.75	0.75	0.70	0.11	0.63	0.54
Widowed	0.10	0.07	0.08	0.50	0.19	0.28
Divorced	0.06	0.04	0.08	0.71	1.01	0.34
Separated	0.00	0.04	0.00	1.50	1.34	0.00
Single	0.10	0.11	0.14	0.18	0.55	0.74
Educational attainment (%)
No education	0.17	0.09	0.20	1.40	1.67	0.34
Primary	0.40	0.56	0.20	1.80*	3.82***	2.25**
Ordinary level	0.17	0.14	0.32	0.51	2.22**	1.80*
Advanced level	0.25	0.21	0.28	0.56	0.84	0.31
Number of children	4.11	3.30	2.62	1.79*	1.49	3.36***
Number of children at time of joining	3.38	2.74	2.62	1.45	0.26	1.71*
Savings prior to joining VSLA? (%)	0.46	0.49	0.36	0.34	1.37	1.07
Access to loans prior to joining? (%)	0.03	0.12	0.08	1.90*	0.73	1.14

Notes: VSLA = Village Savings and Loan Association. The treatment group was divided into ‘Older’ members and ‘Recent’ members of a VSLA with the dividing line being the median number of years of membership. Means for these two groups and the control group of ‘New’ members were compared pairwise using t tests for non-proportions and a proportions test to compare proportions. The reported test statistics are t statistics for non-proportions and z statistics for proportions. Statistically significant results are indicated as follows: ***p < 0.01, **p < 0.05, *p < 0.1.

4. Results

We analyse the impact of VSLA participation on a variety of indicators of household and individual welfare, including the development of income-generating activities, asset expenditure levels, quality of housing, educational spending, nutritional status, and health expenditure levels. We first conduct simple mean comparisons of these various outcome measures across the treatment and control groups and then test our findings using regression analysis.

The results in Table 3 show several significant differences in outcomes between the treatment and control groups, all of which indicate a beneficial impact of VSLA membership. Compared with new VSLA members, current members engage in more income-generating activities, spend more on education and health, eat higher quality food (meat and fish) more often, and are more likely to own their own home and to have made improvements in their homes recently.

Table 3. Economic, health, and educational outcomes of treatment and control groups.

	Treatment group	Control group	Test statistic
N	120	50
Number of IGAs	1.91	1.39	4.58***
2009 asset expenditure (TSh)	138 078	31 289	4.73***
Housing tenure (%)
Owned by household	0.86	0.60	3.67***
Shared	0.08	0.34	4.22***
Rented	0.01	0.00	0.64
Other	0.05	0.06	0.04
Improvements in last 12 months? (%)	0.67	0.16	6.08***
2009 education expenditure (TSh)	105 580	33 809	1.88*
Average number of meals per day	2.54	2.46	0.93
Average number of days consumed meat in last week	0.50	0.16	2.64***
Average number of days consumed fish in last week	4.61	1.20	10.38***
2009 health expenditures (TSh)	69 521	36 948	2.17**

Notes: IGA = income-generating activity. Means for the treatment and control groups were compared using t tests for non-proportions and a proportions test to compare proportions. The reported test statistics are thus t statistics for non-proportions and z statistics for proportions. Statistically significant results are indicated as follows: ***p < 0.01, **p < 0.05, *p < 0.1.

Regression analysis allows us to investigate these correlations further while controlling for the various observable characteristics described in Table 1. Because there were no significant differences between the treatment and control groups in these observable characteristics, the basic results of the regression analysis are similar to those presented in Table 3. The regression analysis also provides a useful way to explore differential programme impact by gender, and allows standard errors to be corrected for possible heteroskedasticity and auto-correlation. The control variables included in each regression are derived from those shown in Table 1 and are related to demographics (age, religion, marital status, number of children), education (non-parametric indicators for level of educational attainment), and financial status (prior savings, prior access to loans).

The regression results for several economic outcomes are presented in Table 4. In this table and those that follow, the programme impact on men is given by the coefficient for the ‘Membership’ variable; the impact on women is given by the coefficient for the ‘Membership + (Membership*Gender)’ variable; and the difference in impact between men and women is given by the coefficient for ‘Membership*Gender’. The table confirms that there are statistically significant differences between the treatment and control groups for each of the variables presented, and suggests that these differences do not vary significantly between men and women.

Table 4. Economic outcomes.

	(1)	(2)	(3)	(4)
	Number of IGAs	2009 asset expenditure	Home ownership	Housing improvements
Membership	0.368* (0.205)	116 000*** (40 500)	1.000** (0.430)	1.582*** (0.509)
Marginal effect			0.301** (0.139)	0.552*** (0.130)
Gender	0.044 (0.086)	−21 200* (11 355)	0.389 (0.382)	−0.440 (0.421)
Marginal effect			0.107 (0.112)	−0.173 (0.161)
Membership*Gender	0.157 (0.237)	−29 100 (48 900)	0.008 (0.495)	0.096 (0.484)
Marginal effect			0.002 (0.128)	0.038 (0.193)
Membership + (Membership*Gender)	0.524*** (0.097)	87 200*** (26 700)	1.008*** (0.265)	1.678*** (0.301)
Marginal effect			0.304*** (0.082)	0.576*** (0.072)
Demographic controls	Yes	Yes	Yes	Yes
Education controls	Yes	Yes	Yes	Yes
Financial controls	Yes	Yes	Yes	Yes
Constant	0.913*** (0.304)	−400 (55 500)	0.693 (0.787)	0.056 (0.576)
Observations	162	133	170	170
R-squared/pseudo-R-squared	0.211	0.274	0.130	0.224

Notes: IGA = income-generating activity. The table reports the results of regressions. The column labels indicate the dependent variable while rows indicate explanatory variables. See text for a complete description of the control variables. Columns (1) and (2) were estimated using ordinary least squares regression, while columns (3) and (4) report the results of probit regressions. Standard errors are in parentheses and are heteroskedasticity robust and clustered by Village Savings and Loan Association group (n = 25). Statistically significant results are indicated as follows: ***p < 0.01, **p < 0.05, *p < 0.1.

For each of the variables in Table 4, the magnitude of the impact seems to have practical, as well as statistical, significance. New VSLA members participate in an average of 1.39 income-generating activities. Column (1) suggests that VSLA participation results in an increase of 0.37 income-generating activities for men and 0.52 for women, which is likely to be economically significant for the families involved. Column (2) indicates a significant programme impact among members of approximately TSh 100 000 (US$90). The average annual income in Tanzania is approximately TSh 1 367 300 (US$1243) (Human Development Report, 2009), so this increase represents a substantial 7.7% of annual household income. Finally, columns (3) and (4) suggest home ownership and improvement rates increase by 30 and 55 percentage points among VSLA members from a baseline among new members of 60 and 16% – clearly meaningful increases.

The regression results for the one education-related outcome variable, level of education expenditures, are presented in Table 5. The basic specification in column (1) shows no impact of the VSLA programme on educational expenditures for either men or women. While this differs from the simple mean comparison test shown in Table 3, it should be noted that even this result was only marginally significant and the regression has far fewer degrees of freedom. To explore the robustness of the results for education, a second specification with more degrees of freedom is presented in column (2). There are two differences here from the first column: the four non-Muslims, who had much higher than average educational expenditure, are omitted; and the educational controls were reformulated as a single semi-parametric variable. This specification eliminates some outliers in terms of educational spending and reduces the number of variables in the regression by four (two religion indicators omitted, three education indicators combined to a single variable). The results of this alternative specification are suggestive of a positive programme impact on educational spending, although this result is clearly not robust across specifications.

Table 5. Education outcome.

	(1)	(2)
	2009 education expenditures	2009 education expenditures
Membership	−19 400 (41 000)	29 900** (14 300)
Gender	−42 100 (33 000)	−23 900 (36 400)
Membership*Gender	46 300 (54 600)	28 000 (64 100)
Membership + (Membership*Gender)	26 900 (20 600)	39 000** (16 800)
Demographic controls	Yes	Yes
Education controls	Yes	Non-parametric
Financial controls	Yes	Yes
Constant	−130 000 (95 800)	−118 000 (90 600)
Observations	164	160
R-squared	0.294	0.128

Notes: The table reports the results of regressions. The column labels indicate the dependent variable while rows indicate explanatory variables. See the text for a complete description of the control variables. All columns were estimated using ordinary least squares regression. Standard errors are in parentheses and are heteroskedasticity robust and clustered by Village Savings and Loan Association group (n = 25). Statistically significant results are indicated as follows: **p < 0.05.

This weak result for education outcomes is somewhat surprising given the existing literature on the impact of savings on education (Prina, 2013; Dupas & Robinson, 2013a; Jamison et al., 2014). It is possible that educational expenses are not an appropriate proxy for the programme’s impact on education, especially considering that primary education is provided tuition-free by the Tanzanian government. The weakness of this result is also inconsistent with the results of the focus group discussions. The increased ability to finance the education of their children, including tuition fees, materials, testing fees, and so forth, was the most commonly cited benefit of programme participation by focus group participants. For example, one member took out a loan of TSh 100 000 (US$90) to send her two daughters to a secondary boarding school – an opportunity that would most likely have been closed to them without the programme.

The final table of results (Table 6) presents the regressions for health and nutrition outcomes. Column (1) suggests that membership in the VSLA programme increases the number of meals per day for male members’ households, but not for females’. However, the increase of 0.337 meals per day for male members’ households is almost exactly equal to the extra 0.348 meals per day that non-member females’ households consume. Thus, it appears that women might already be prioritising household meals so that there is little room for programme participation to have an impact.

Table 6. Health and nutrition outcomes.

	(1)	(2)	(3)	(4)
	Number of meals per day	Number of times had meat	Number of times had fish	2009 health expenditures
Membership	0.337** (0.151)	0.337 (0.311)	3.48*** (0.743)	28 000 (29 800)
Gender	0.348*** (0.111)	−0.209 (0.272)	−0.762 (0.765)	−23 500 (15 500)
Membership*Gender	−0.358** (0.149)	−0.0494 (0.355)	0.208 (0.814)	−7110 (33 700)
Membership + (Membership*Gender)	−0.020 (0.106)	0.287* (0.144)	3.69*** (0.341)	20 900* (11 900)
Demographic controls	Yes	Yes	Yes	Yes
Education controls	Yes	Yes	Yes	Yes
Financial controls	Yes	Yes	Yes	Yes
Constant	2.04*** (0.226)	0.580 (0.396)	0.876 (0.867)	1440 (29 400)
Observations	170	168	168	160
R-squared	0.133	0.137	0.491	0.145

Notes: The table reports the results of regressions. The column labels indicate the dependent variable while rows indicate explanatory variables. See the text for a complete description of the control variables. All columns were estimated using ordinary least squares regression. Standard errors are in parentheses and are heteroskedasticity robust and clustered by Village Savings and Loan Association group (n = 25). Statistically significant results are indicated as follows: ***p < 0.01, **p < 0.05, *p < 0.1.

In addition to meal quantity, VSLA participation seems to have a considerable impact on meal quality, evident in an increase in the quantity of meat and fish consumed in the past week. While VSLA programme participation seems to have no significant impact on meat consumption for the households of male members, column (2) suggests households of female VSLA members consume meat 0.287 more days per week than non-members. The coefficient on gender is negative, although insignificant, implying that control households of female respondents consume meat approximately 0.2 less times than those of male respondents. This runs contrary to previous research suggesting that women are more likely than men to invest in the household’s diet. It may be that female-headed households are more resource constrained than male-headed households. This conjecture is supported by the fact that women in control households spend less on household assets, as shown in column (2) of Table 4.

Alternatively, given the relatively high price of meat in Zanzibar, women may spend a greater proportion of their resources on more cost-effective food items such as grains or fish. They may be more concerned with meal quantity than quality – a hypothesis supported by the significant and positive coefficient on gender in column (1) in Table 6. Correspondingly, female VSLA members experience a significant programme impact on household meat consumption, because they begin to spend more on relatively expensive meat only when participation in a VSLA increases the quantity of available resources.

The regression results presented in column (3) are also consistent with this theory and indicate that participation in the VSLA programme has a substantial impact on fish consumption. Programme participation increases weekly servings of fish by approximately 3.5 for both men and women. This is a very significant increase of a healthy protein source compared with the 1.2 servings per week consumed by the control group.

The final column of Table 6 suggests that the VSLA programme had a moderately significant impact on the health expenditures for the households of female members only. This is consistent with the literature that suggests women are more likely to spend additional resources on the health and welfare of their families than men.

These results for health and nutrition are broadly consistent with the results of the focus group discussions. Many of the focus group participants named nutrition as one of the primary uses of both savings and loans. Several participants also listed improved access to health care as a one of the major benefits of programme membership.

5. Discussion

The provision of financial services has expanded rapidly in developing countries over the past decade, but millions of people remain without access. Poor road quality and lower population density makes the provision of formal financial services in rural areas prohibitively expensive. The VSLA model offers a promising way of increasing financial access in such remote areas. In particular, VSLAs are entirely self-sufficient, requiring no external contributions to the loan portfolio and only limited outside support beyond the initial years.

The results of this study suggest that the VSLA programme is beneficial to participants. Compared with new members, those with several years of VSLA membership participated in more income-generating activities and had higher asset expenditure. They were also more likely to own homes and to have made improvements to those homes in the past year. Membership was also associated with eating more and higher quality meals. Finally, there is evidence that members spent more money educating their children. Discussions with VSLA members were also consistent with membership being causally related to these economic improvements.

This work serves as the first controlled evaluation of VSLAs of which the authors are aware and suggests two main directions for future research. First, larger studies should be performed to further elucidate programme impacts and to allow higher powered subgroup analysis; for example by gender. The current study did not find significant differences by gender, but with a relatively small sample consisting of 70% women it had insufficient power to detect economically meaningful differences. Second, analyses using alternative control strategies, ideally with randomised assignment, would be helpful to confirm that the beneficial impacts found here are indeed causally related to programme participation.

The VSLA programme may not have as substantial an impact on its members as many of the larger microfinance programmes, such as the renowned Grameen Bank in Bangladesh or BancoSol in Bolivia. These organisations have substantial donor resources at their disposal and, therefore, are able to provide much larger loans at slightly lower interest rates, which may facilitate greater impacts. But they are also constrained by the need for more formal infrastructure and are unable to reach more rural areas. Inasmuch as the VSLA approach does not rely on outside donor funding and does not require continued support of the founding organisation, it may prove to be more cost-effective, sustainable, and easily replicated than alternative approaches. Overall, the VSLA model appears to be both successful and sustainable – it is a promising means of improving access for those not otherwise reached by traditional financial services.

Notes

¹The exchange rate at the time of data collection (2010) was approximately 1110 Tanzanian shillings (TSh) per US dollar (US$).